Secure Personalized Superintelligence (PSI)

Abstract

Personal superintelligence (PSI) represents a next-generation breakthrough in the development of highly autonomous artificial intelligence agents. However, due to its extremely high intelligence, PSI also poses a dangerous potential threat to human safety. This invention discloses methods for securely designing and building such agents. Furthermore, this invention demonstrates how to securely use PSI as part of a super-intelligent general-purpose artificial intelligence (AGI) system. Preferred embodiments are disclosed, including methods for enabling PSI to rapidly improve itself with or without human supervision. This invention describes methods for generating different versions of PSI using several novel methods. It also describes methods for implementing scalable safety checks that operate effectively even when PSI becomes far smarter than human creators. Finally, a completely new approach to AI safety that relies on the PSI community combined with proven blockchain techniques is disclosed. Rather than relying on testing, this invention promotes PSI safety through design.

Description

Technical Field

[0001] In some aspects, the technology of the present invention relates to the use of intelligent agents to implement Personalized Secure Intelligence (PSI) for the purpose of developing and continuously improving PSI for each human owner. In some other aspects, the technology of the present invention relates to PSI that utilizes a community of multiple artificial intelligence (AI) agents or systems that all communicate on a collective network, and / or methods associated with creating multiple pre-customized PSIs, all of which agree on a set of human-aligned value and ethical rules that reflect the value and ethics of the PSI community and / or its human users.

[0002] In another aspect, the technology of the present invention relates to PSI that operates based on pre-trained knowledge and values in PSI, which at least initially depends on information provided by or related to a human user.

[0003] In yet another aspect, the technology of the present invention relates to the use of AI agents and / or a community of PSIs to perform security checks or ethical checks on other PSIs on a network, ensuring that any task or activity performed by a PSI complies with a set of agreed-upon rules.

[0004] In still other aspects, all of the activities described in this patent disclosure, such as those that occur on an external network where multiple intelligent entities participate in collaborative problem-solving, can also be implemented within a single computer intelligent system, and all intelligent entities are computerized agents or AI agents or PSIs that reside within a single computer intelligent system.

Background Art

[0005] Several prior patent applications describe the path to development of super-intelligent general-purpose artificial intelligence (AGI) (super-intelligent AGI and individual highly autonomous artificial intelligence (AAAI)). These prior patent applications describe specific scenarios involving existing products and technologies available from several companies. These prior patent applications describe how the learning and skills of each AAAI can be accelerated by combining and learning from data from cross-platform AAAI implementations. These prior patent applications describe systems and methods for integrating common technical terminology and how AAAIs can be integrated into AGI networks using a human-centered AGI approach.

[0006] The field of AI was named in 1956 at a conference held in Dartmouth, Massachusetts, USA, organized by computer scientist John McCarthy. Among the researchers who attended the Dartmouth conference were Herbert A. Simon (later a Nobel laureate) and Allen Newell (later a renowned computer scientist), both of whom were affiliated with Carnegie Mellon University.

[0007] Simon and Newell, along with their colleague Cliff Shaw, presented only a working demonstration of AI at the Dartmouth Conference. It was a program called Logic Theorist. Logic Theorist is an example of the early state of AI efforts where the rules defining the AI's behavior were directly programmed into the computer by human programmers. Interestingly, by programming the rules in a general way, the computer program enabled Logic Theorist to pursue its goals and sub-goals through various means (called "operators"), which allowed it to demonstrate creative behavior.

[0008] Specifically, Logic Theorist is programmed to reproduce mathematical proofs from textbooks (Principia Mathematica by Betrand Russell and Alfred North Whitehead), but Logic Theorist actually discovered new proofs that were previously unknown to both Logic Theorist's programmers and Russell and Whitehead themselves. Reportedly, Russell and Whitehead were so impressed by Logic Theorist's new proofs that they wrote to the inventors, stating that not only were the proofs previously unknown to them, but that they wished they had come up with them themselves. Thus, at the time the field of AI was born in 1956, AI was already capable of creative thinking. Particularly relevant to this patent is Logic Theorist's architecture, which uses goals and sub-goals. This approach was further developed by Newell and Simon and subsequently adopted by many AI systems, and is applied in a novel and creative way in this patent.

[0009] Research in the field of AI from 1956 to 1986 was primarily dominated by the "expert systems" approach. This involved humans with programming skills interviewing human experts and incorporating their knowledge into a set of programmed rules for AI. This process was called "knowledge engineering." As a result of knowledge engineering, AI programs capable of functioning like human experts in limited fields emerged. For example, the program MYCIN, developed at Stanford University in the 1970s, operates as an expert system in the field of blood infections. EAFeigenbaum et al. at Stanford University continued development of a full series of expert systems across various medical fields in the 1980s. Similar expert system research was also conducted at many other universities.

[0010] As more and more expert systems were developed, Newell and Simon turned their attention to the best available intelligent model—humans—in an effort to improve the performance of AI systems. Their research yielded a very powerful and broad theory that could meticulously explain how humans solve almost any type of problem. This theory was detailed in early research on logic theorists and was known as "problem space exploration." The theory was further elaborated in their book "Human Problem Solving," published in 1972.

[0011] Dr. Craig Kaplan, inventor of the AAAI patent, conducted research with Herbert Simon and Allen Newell in the 1980s. He collaborated with Dr. Simon in the fields of creative problem solving and cognitive science, including the publication of the paper "Foundations of Cognitive Science" in 1989. Dr. Kaplan recognized that the "problem space exploration" architecture proposed by Newell and Simon could be generalized to enable collective problem solving by millions of people via the internet. From the late 1990s onward, Dr. Kaplan began to translate his ideas into practice in various operating systems that actively utilize human collective intelligence.

[0012] For example, Dr. Kaplan pioneered some of the first practical applications of crowdsourced intelligence around 2000. In 2001, at the inaugural Global Brain Conference in Brussels, he outlined his idea of ​​applying collective intelligence to one of the most difficult and competitive problems in business: beating Wall Street. By 2006, he was able to gain an advantage in the stock market by designing and implementing the "PredictWallStreet" system using the collective intelligence of millions of people. In 2018, that system became the driving force behind one of the top 10 market-neutral hedge funds in terms of performance, thus proving its effectiveness in performing at the highest level in a complex field where one competes with some of the smartest people on the planet.

[0013] In the process of designing and implementing these systems, Dr. Kaplan recognized that the “problem space exploration” architecture, which had served as a general framework for human problem solving, could be adapted and enhanced to function as a general architecture for cognition, including both humans and AI agents. Furthermore, by representing intelligent behavior as a form of problem solving, he provided a way for many AI agents to interact with each other, as they pool their collective intelligence to create AGI. This “collective intelligence” approach is presented here as the AAAI system and method for AGI, representing a path to AGI being faster and more powerful compared to existing efforts. Most existing efforts to realize AGI have focused primarily on training larger-scale language models (LLMs) using more data, more powerful computers, and better machine learning algorithms. The AAAI approach also has the advantage of allowing humans to easily participate in training and improving the intelligence of AI, which includes helping to shape the values ​​and ethics of AI, and is an essential feature for ensuring the safe development of AGI.

[0014] While Dr. Kaplan recognized the importance of collective intelligence early on, most other AI researchers increasingly focused on a branch of AI known as machine learning (ML). Starting in the 1980s, ML began to gain attention as a way for AI to learn knowledge on its own, rather than knowledge engineers programming that knowledge into it. However, progress in ML was very slow until a paper demonstrating the use of "feedback backpropagation" (one of the first practical reinforcement learning techniques) was published in 1986. After that paper was published, some AI researchers began to think that the future of AI would depend on machines educating themselves rather than humans programming them. Unfortunately, the computational and data requirements for ML were enormous, and for the most part, they demanded conditions that exceeded the capabilities of the technology in the 1980s, or even the 1990s.

[0015] Moore's Law (the principle that computing power doubles approximately every 18 months) meant that it took about 30 years for the computing power of technology to catch up to the level required for ML algorithms. During this same period, the amount of data available for training such models, especially on the internet (which began to spread after 1995 with the advent of web browsers), began to increase.

[0016] The "AI decline" of the 1990s and early 2000s stemmed from overly optimistic ambitions for AI that exceeded readily available data and computing power. However, by 2010, sufficient computing power, data, and "sufficiently good" ML algorithms had converged. AI progress began to accelerate rapidly, including the development of improved learning algorithms such as "Transformers."

[0017] As of early 2023, machine learning approaches had taken precedence, while knowledge engineering approaches for creating expert systems had been largely ignored. Machine learning approaches have succeeded in enabling machines to teach themselves how to beat the best human champions in chess, Go, and any two-player game. Programs like AlphaFold determined the shapes of millions of proteins in just a few months. In contrast, previously, even the best human experts took 4-6 years to accurately determine the shape of a single protein. Natural language processing (NLP), a branch of AI that focuses on understanding human language, has made breakthroughs, giving rise to assistants like Amazon's Alexa and Apple's Siri, and more recently, large-scale language models (LLMs) like OpenAI's GPT.

[0018] The technology of this invention, namely the scaling and improvement of LLM, represents a critical turning point that has enabled AI to move from specialized tools of interest in specific fields (also known as "narrow AI") to more general applications. The release of CHATGPT by OpenAI, the subsequent release of BARD by Google®, the integration of GPT into Microsoft®'s Bing® search engine, and the proliferation of AI companies specializing in the widespread application of ML approaches have led to a major shift in innovation in AI applications. Many individual fields, ranging from medical applications, vehicle navigation, office work, legal services, marketing, sales, education, and even beer brewing, are all being revolutionized by the application of LLM, and more broadly, by advances in machine learning approaches and capabilities.

[0019] However, one goal remains unresolved. As of February 28, 2023, apart from the technology of the present invention detailed in this description, no company or individual has described how to create a practical system for AGI. This is because ML alone is insufficient to rapidly realize AGI, and collective intelligence is also necessary. [Prior art documents] [Non-patent literature]

[0020] [Non-Patent Document 1] Betrand Russell and Alfred North Whitehead, “Principia Mathematica” [Non-Patent Document 2] "Human Problem Solving", 1972 [Non-Patent Document 3] "Foundations of Cognitive Science", 1989 [Overview of the project]

[0021] In light of the aforementioned disadvantages inherent in known approaches to PSI, at least some embodiments of the technology of the present invention provide novel embodiments or creations of safe PSI to overcome the disadvantages and shortcomings of the prior art mentioned. Accordingly, the general object of at least one embodiment of the technology of the present invention, which will be described in further detail later, is to provide a new and novel safe PSI system and method having all the advantages of the prior art mentioned herein and many novel features that result in safe PSI, either alone or in any combination thereof, which are not anticipated, explicitly not expressed, not suggested, or further implied by the prior art.

[0022] In one embodiment, the technology of the present invention may include a system for personalized superintelligence (PSI) that uses intelligent agents to develop and continuously improve PSI for human users utilizing a computer system where everything communicates electronically through a collective network, and additional PSI. The system may include a processor, a computer-readable storage medium, and program instructions, the program instructions being stored in the computer-readable storage medium and executable by the processor, and the computer system Implementing a base-level AI agent on a computer system, where the base-level AI agent is already customized, and Collecting media information related to a human user, and Analyzing the media information, and Converting the analyzed media information into a training data set, and Differentially weighting the converted training data set, and Adding a knowledge module to the weighted and converted training data set, and Identifying a new data source to include in the weighted and converted training data set, and Applying the weighted and converted training data set to the base-level AI agent to create a personalized PSI, and Using a network to communicate the personalized PSI with a plurality of additional PSIs and enable community-based security features from the plurality of additional PSIs to the personalized PSI, and Causing to be performed.

[0023] According to another aspect, the technology of the present invention may include a method for PSI using intelligent entities to develop and continuously improve PSI, where the intelligent entities include a human user utilizing a computer system, additional AI agents, and additional PSIs, and all of these communicate electronically through a collective network. The method includes Obtaining a pre-customized base-level AI agent, and Collecting media information related to a human user, and Analyzing the media information, and Converting the analyzed media information into a training data set, and Differentially weighting the converted training data set, and The steps include adding knowledge modules to a weighted, transformed training dataset, The steps include identifying new data sources to include in the weighted, transformed training dataset, A step of applying a weighted, transformed training dataset to a base-level AI agent in order to create a user PSI, wherein the base-level AI agent is subjected to the application, and a step of applying the dataset. The steps involve using a network to allow a user PSI to communicate with multiple additional PSIs, enabling community-based security features from the multiple additional PSIs to the user PSI, Includes.

[0024] Some embodiments of the technology of the present invention may include the step of obtaining one or a combination of new training datasets and training modules to be added to a weighted transformed training dataset by a human user, a base-level AI agent, or user PSI.

[0025] Some embodiments of the technology of the present invention may include the step of monetizing a user PSI by enabling other human users or other PSIs to access and use the personalized PSI's weighted, transformed training dataset.

[0026] In some embodiments, the base-level AI agent may be one of the following: a pre-trained large-scale language model (LLM), another tunable and trainable AI agent, or one or more customized AIs from other human owners.

[0027] In some embodiments, media information may be one or a combination of any of the following: images of human users, images and topics related to human users, photographs or images of human users, photographs or images of people and topics related to human users, works related to human users, periodicals related to human users, blogs related to human users, posts related to human users, tweets related to human users, emails related to human users, podcasts related to human users, records related to human users, audio content related to human users, audio content of people and topics related to human users, data collected by third-party vendors related to human users, websites related to human users, apps related to human users, online information related to human users, social media information related to human users, and other AI agents related to human users.

[0028] Some embodiments of the technology of the present invention may include the step of granting permission to one or more social media platforms so that social media content related to human users can be accessed by a base-level AI agent or user PSI.

[0029] In some embodiments, analyzing media information may further include annotating and categorizing the media information.

[0030] In some embodiments, analyzing media information can utilize one or more algorithms, which include one or a combination of transformation algorithms, deep learning algorithms, transcription algorithms, content and sentiment analysis and summarization, LLM, and crowdsourced and cloud-supervised humans.

[0031] Some embodiments of the technology of the present invention may include the step of a human user paying one or more human workers on a crowdsourcing website to review a training dataset or suggest refinements to the training dataset.

[0032] In some embodiments, the step of weighting the transformed training dataset can be performed by a human user using an interface on a computer system implementing a base-level AI agent, by adjusting one of the weight values ​​of the transformed training dataset.

[0033] In some embodiments, the step of weighting the transformed training dataset can be performed using a computer system by one or a combination of additional AI agents, additional PSI, and additional human workers.

[0034] Some embodiments of the technology of the present invention may include the step of inputting ethical values ​​into a training dataset by providing a series of interactive dialogues to a human user, the interactive dialogues including a set of predefined ethical scenarios.

[0035] Some embodiments of the present invention may include the step of mixing the weights of a transformed training dataset in order to improve the LLM of a base-level AI agent.

[0036] Some embodiments of the technology of the present invention may include a step in which a human user utilizing the interface of a base-level AI agent authorizes the LLM to download data and automatically add the data to a training dataset, or to create a new training dataset.

[0037] In some embodiments, downloaded data can be provided to a human user via a computer system for approval by the human user before being added to a training dataset or before a new training dataset is created.

[0038] Some embodiments of the present invention may include the step of cloning a user PSI into one or more cloned user PSIs.

[0039] Some embodiments of the technology of the present invention may include the step of training each of the cloned PSIs using different training datasets, each of which has different weights.

[0040] Some embodiments of the present invention may include the step of creating a combined training dataset by combining a weighted training dataset from a user PSI with one or more cloned PSIs.

[0041] Some embodiments of the technology of the present invention may include the step of providing one or any combination of weighted training datasets of any one or any combination of cloned PSIs to one of an additional AI agent and one of the additional PSIs for use in training.

[0042] Some embodiments of the technology of the present invention may include the step of training a user PSI using a weighted training dataset of any one or any combination of cloned PSIs.

[0043] Some embodiments of the technology of the present invention may include the step of having a human user using a computer system perform an automated process of user PSI so as to perform a task without human user intervention when no default parameters are triggered.

[0044] Some embodiments of the technology of the present invention may include the step of monitoring for activities that violate a list of prohibited activities using a base-level AI agent or user PSI, and triggering an intervention activity provided to a human user.

[0045] Some embodiments of the technology of the present invention may include the step of detecting whether there are any deficiencies in the training dataset using a base-level AI agent or user PSI, and if so, initiating an interaction with an intelligent entity to obtain data to fill the deficiencies.

[0046] In some embodiments, interactions between two or more intelligent entities can be performed in parallel.

[0047] Some embodiments of the technology of the present invention may include the steps of providing a task to a plurality of additional AI agents or additional PSIs on a network, providing results for the task from each of the additional AI agents or additional PSIs, and determining a winning result from the plurality of results.

[0048] In some embodiments, additional AI agents or additional PSIs can work on tasks independently and in parallel with each other.

[0049] In some embodiments, one or more of the additional AI agents or additional PSIs may not be customizable, while one or more of the additional AI agents or additional PSIs may be customizable.

[0050] Some embodiments of the technology of the present invention may include a step in customizing a user-based AI agent or user PSI that utilizes a training dataset for an additional AI agent or additional PSI along with the win result.

[0051] Some embodiments of the technology of the present invention may each include the step of utilizing a universal problem-solving architecture for a task, with the help of an additional AI agent or additional PSI, respectively.

[0052] Some embodiments of the technology of the present invention may include the step of using a set of safety or ethical rules agreed upon by each of the user-based AI agent, user PSI, additional AI agent, and additional PSI.

[0053] In some embodiments, each additional AI agent or additional PSI may have a set of weights that encode the knowledge of the additional AI agent and the additional PSI, respectively.

[0054] In some embodiments, the activity record may be available to any computer device on the network.

[0055] Some embodiments of the technology of the present invention may include the step of providing a task to a user PSI configured to produce hallucination by LLM of the PSI.

[0056] Some embodiments of the technology of the present invention may include the step of providing a task to a user PSI, the task being a random permutation of existing standardized tasks or a task dynamically arising based on changing conditions.

[0057] In another embodiment, the technology of the present invention may include a method for PSI that uses intelligent entities to develop and continuously improve PSI, the intelligent entities including a human user utilizing a user computer system, additional AI agents, and additional PSI, all of which communicate electronically through a collective network. The step of creating a user PSI is: Acquire a pre-customized base-level AI agent, Collecting media information related to human users, Analyzing media information and Converting the analyzed media information into a training dataset, Differentiated weighting of the transformed training dataset, To create a user PSI, the weighted transformed training dataset is applied to a base-level AI agent, and the base-level AI agent is created by applying, receiving, and performing the following steps: A step of enabling community-based safety features by using a network to have a user PSI communicate with multiple additional AI agents or additional PSIs, wherein the user PSI and the additional AI agents or additional PSIs each agree to use a set of safety or ethical rules, and communicate accordingly. A step of recording, in an auditable format, all actions taken by the user PSI and the additional AI agent or additional PSI on one or any combination of the following: a central computer system on the network, a user computer system, an additional AI agent, and an additional PSI; The steps include monitoring each action according to a set of rules and flagging any action that does not follow the set of rules, It may include.

[0058] In some embodiments, blockchain technology can be used to record actions.

[0059] In some embodiments, recording actions can utilize a universal problem-solving framework to record one or a combination of goals, sub-goals, problem states, and problem-solving cognitive activities or steps taken in cognitive activities performed on the network.

[0060] In some embodiments, recorded actions can only be modified when an additional AI agent or a majority of additional PSIs on the network provide approval for the change.

[0061] Some embodiments of the technology of the present invention may include the step of stopping any of the flagged actions and applying a preventative action to prevent the recurrence of the flagged action.

[0062] Some embodiments of the technology of the present invention may include the step of identifying one or more additional AI agents or additional PSIs that have provided a flagged action, and controlling the participation of the identified AI agents or PSIs on the network.

[0063] Some embodiments of the technology of the present invention may include the step of analyzing flagged actions and adjusting one or a combination of the training dataset, the set of rules, and the network attributes based on the analysis of the flagged actions.

[0064] In yet another aspect, the technology of the present invention may include a method for PSI that uses intelligent entities to develop and continuously improve PSI, the intelligent entities including a human user utilizing a user computer system, additional AI agents, and additional PSI, all of which communicate electronically through a collective network. The step of creating a user PSI is: Acquire a pre-customized base-level AI agent, Collecting media information related to human users, Analyzing media information and Converting the analyzed media information into a training dataset, Differentiated weighting of the transformed training dataset, To create a user PSI, the weighted transformed training dataset is applied to a base-level AI agent, and the base-level AI agent is created by applying, receiving, and performing the following steps: The steps include providing tasks to the user PSI and to multiple additional AI agents or additional PSIs on the network, The steps include providing results for the task from both the user PSI and the additional AI agent or additional PSI, The steps to determine the winner from multiple results, It may include.

[0065] In some embodiments, each additional AI agent or additional PSI can work on tasks independently and in parallel with each other.

[0066] In some embodiments, one or more of the additional AI agents or additional PSIs are not customizable, while one or more of the additional AI agents or additional PSIs are customizable.

[0067] Some embodiments of the technology of the present invention may include a step in customizing a user-based AI agent or user PSI that utilizes a training dataset for an additional AI agent or additional PSI along with the win result.

[0068] Some embodiments of the technology of the present invention may each include the step of utilizing a universal problem-solving architecture for a task, with the help of an additional AI agent or additional PSI, respectively.

[0069] Some embodiments of the technology of the present invention may include the step of using a set of safety or ethical rules agreed upon by each of the user-based AI agent, user PSI, additional AI agent, and additional PSI.

[0070] In some embodiments, each additional AI agent or additional PSI may have a set of weights that encode the knowledge of the additional AI agent and the additional PSI, respectively.

[0071] In some embodiments, the activity record may be available to any computer device on the network.

[0072] Some embodiments of the technology of the present invention may include the step of providing a task to a user PSI configured to produce hallucination by LLM of the PSI.

[0073] Some embodiments of the technology of the present invention may include the step of providing a task to a user PSI, the task being a random permutation of existing standardized tasks or a task dynamically arising based on changing conditions.

[0074] In some embodiments, the network may include a user PSI and additional AI agents or additional PSIs, and one or more additional networks each include an AI agent or PSI.

[0075] In another aspect, the technology of the present invention may include a method for PSI that uses intelligent entities to develop and continuously improve PSI, the intelligent entities including a human user utilizing a user computer system, additional AI agents, and additional PSI, all of which communicate electronically through a collective network. a) A step of creating a user PSI, wherein the step of creation is: Acquire a pre-customized base-level AI agent, Collecting media information related to human users, Analyzing media information and Converting the analyzed media information into a training dataset, Differentiated weighting of the transformed training dataset, To create a user PSI, the weighted transformed training dataset is applied to a base-level AI agent, and the base-level AI agent is created by applying, receiving, and performing the following steps: b) A step of using a network to enable community-based safety features, wherein the user PSI and the additional AI agents or additional PSIs communicate to agree to use a set of safety or ethical rules, respectively. c) Steps to achieve baseline performance of user PSI and additional AI agents or additional PSI by utilizing various standardization tasks, d) The step of creating one or more versions of the user PSI, each being different from the others, e) A step of determining the user PSI network, the user PSI version, and the performance of additional AI agents or additional PSIs, f) A step of assigning a credit value or responsibility value to one or more factors that resulted in superior or inferior performance of the baseline performance, g) The step of determining which elements are superior in terms of performance improvement and retaining the superior elements for future use, It may include.

[0076] Some embodiments of the technology of the present invention may include the step of determining which element is the inferior element that caused the performance degradation and modifying the attributes of the inferior element.

[0077] Some embodiments of the technology of the present invention may include a step of repeating step c) using superior elements and modified inferior elements.

[0078] Some embodiments of the present invention may include repeating steps c) to g) until no better elements are detected.

[0079] In some embodiments, the network may include a user PSI and additional AI agents or additional PSIs, and one or more additional networks each include an AI agent or PSI.

[0080] Some embodiments of the technology of the present invention may include the step of recording, in an auditable format, all actions taken by a user PSI and an additional AI agent or additional PSI with respect to one or any combination of a central computer system on a network, a user computer system, an additional AI agent, and an additional PSI.

[0081] In some embodiments, a standardization task may be one or a combination of any of the following: a problem-solving task that uses a common universal problem-solving architecture; an ethical and safety scenario designed to determine whether either a user PSI and an additional AI agent or any combination of the additional PSIs behave safely and ethically; a standard intelligence test and assessment designed to test the intelligence of intelligent entities; a task configured to measure the degree of "hallucination" or the generation of erroneous results; a task arising from various disciplines; a task that incorporates different cultural or group norms regarding behavior; a task that is a random permutation of existing standardization tasks; or a task that is dynamically created based on changing conditions.

[0082] In some embodiments, a version of the user PSI can be created by any one or a combination of the following: human adjustment of parameters, weights, or data encoding the knowledge of the user PSI; autonomous adjustment of parameters, weights, or data encoding the knowledge of the user PSI; random variation of parameters, weights, or data encoding the knowledge of the user PSI; subjecting the user PSI to different training regimes or training amounts; creating different versions of the user PSI sequentially and creating different versions of the user PSI in parallel.

[0083] In some embodiments, a user PSI version can be created by directly combining parameters, weights, or data that encode knowledge of multiple PSIs, by calculating an average weight value, and weighting newer or more complex versions of the user PSI so that it is greater than older or simpler versions of the user PSI.

[0084] In some embodiments, user PSI versions can be created randomly or intentionally to estimate which version will yield beneficial results, and the estimation method is one or a combination of the following: comparing the degree of match between the knowledge of one user PSI version and the statistical frequency of tasks submitted to the network; comparing the overlap of knowledge of different additional PSIs on the network to make modifications to optimize the performance of the group of user PSI versions, taking into account the characterization of the knowledge of the entire network in comparison to the characteristics of the problem that the network is expected to solve; and using hill climbing or gradient descent to optimize any one parameter of any one of the user PSI versions.

[0085] In another aspect, the technology of the present invention may include a method for PSI utilizing a single computer intelligence system, which includes multiple AI agents residing in a single computer intelligence system. This method is This involves acquiring a pre-customized base-level AI agent, which resides within a single computer intelligence system. Collecting media information related to human users associated with a base-level AI agent, Converting the analyzed media information into a training dataset, Differentiated weighting of the transformed training dataset, To create a user PSI, we apply a weighted, transformed training dataset to a base-level AI agent, To enable community-based security features from the additional PSI to the user PSI, and to allow the user PSI to communicate with multiple additional PSIs. Training a base large-scale language model (LLM) of an AI agent using guardrails that include attributes associated with one or a combination of safety, ethics, and knowledge, wherein the AI ​​agent resides in a single computer intelligence system, and the training is carried out accordingly. Customizing the base LLM to an ethical profile, This involves combining ethical information from multiple additional AI agents residing within a single computer intelligence system, where the additional AI agents differ from those of the AI ​​agents themselves, and where the additional AI agents reside within the single computer intelligence system. Based on the resolution of the problem request, the set values ​​of the base LLM will be refined, By updating the base LLM using a combined set of ethical information and refined values, scalable AGI is enabled, It may include.

[0086] Therefore, in order to gain a deeper understanding of the "Modes for Carrying Out the Invention" described below, and to gain a deeper appreciation of the contribution of the present invention to the art, the technical features of the present invention are outlined in a fairly broad sense.

[0087] Some of the objectives, features, and advantages of the present invention will become immediately apparent to those skilled in the art upon reading the following detailed description of the present invention, but when considered in conjunction with the accompanying drawings, it will be clear that the embodiments of the present invention are illustrative.

[0088] Accordingly, those skilled in the art will recognize that the concepts on which this disclosure is based can be readily used as a basis for designing other structures, methods, and systems to carry out some of the objectives of the art of the present invention.

[0089] Therefore, the objective of the present invention is to provide a new and novel safe PSI that has all the advantages and disadvantages of prior art AGI systems.

[0090] Another objective of the present invention is to provide a new and novel safe PSI that can be easily and efficiently manufactured and marketed.

[0091] A further objective of the technology of the present invention is to provide a new and novel secure PSI that results in low cost realization in terms of both resources and work, and therefore, as a result, enables low-price sales to general consumers, thereby making such secure PSI economically available to general buyers.

[0092] A further objective of the technology of the present invention is to provide a novel safe PSI that offers some of the advantages of the prior art in the manner of the prior art, while simultaneously overcoming some of the disadvantages typically associated with the prior art.

[0093] To gain a deeper understanding of the technology of the present invention, its operational advantages, and the specific purposes achieved by its use, please refer to the accompanying drawings and descriptions that exist illustrating embodiments of the technology of the present invention. While several purposes of the technology of the present invention are revealed herein, it should be understood that the following description is not limited to satisfying most or all of the specified purposes, and that some embodiments of the technology of the present invention may satisfy only one or none of such purposes.

[0094] Considering the detailed explanation below will lead to a deeper understanding of this technology and reveal purposes other than those described above. Refer to the attached diagram for such explanations. [Brief explanation of the drawing]

[0095] [Figure 1] This is a flowchart showing embodiments of subsystems available in the AAAI system and method of the present invention. [Figure 2] This is a block diagram illustrating an exemplary process of the entire process available with the technology of the present invention. [Figure 3] This flowchart shows exemplary embodiments of systems and methods for creating scalable, ethical, and safe AGI or PSI from AAAI and human collective intelligence available in the technology of the present invention. [Figure 4] This flowchart shows an exemplary embodiment of a scalable universal problem-solving system and method for human-centered AGI, related to PSI, constructed in accordance with the technical principles of the present invention. [Figure 5] This is a flowchart illustrating an exemplary embodiment of a scalable solution-learning subsystem or process. [Figure 6] This flowchart illustrates an exemplary embodiment of a scalable natural language-to-problem-solving language translation subsystem or process. [Figure 7] This flowchart illustrates an exemplary embodiment of a scalable, highly-rated component subsystem or process for human, AI, and / or PSI problem-solving agents. [Figure 8] A flowchart illustrates an exemplary embodiment of a scalable safety and ethical check subsystem or process where AI can also be used as PSI. [Figure 9] This figure shows the characteristics and features of a problem-solving architecture, including a tree structure, used by the scalable WorldThink protocol, where AAAI can also be PSI. [Figure 10]This diagram illustrates various use cases for domain-specific problems that rely on the underlying WorldThink protocol, which, when used together, can help form the foundation for AGI systems capable of solving a wide range of problems, and the AAAIs identified in the diagram can also be PSIs. [Figure 11] This figure shows some of the steps of a universal problem-solving framework that is part of the WorldThink protocol, used by the AAAI system, and available to PSI. [Figure 12] This flowchart shows some of the basic problem-solving capabilities supported by the WorldThink protocol available in the AAAI system and method of the present invention, where the solver can be PSI. [Figure 13] This flowchart shows some of the basic problem-solving capabilities supported by the WorldThink protocol, which utilizes two problem solvers that may be PSIs working together to resolve client issues. [Figure 14] This flowchart illustrates an exemplary customization process for an AAAI system, where the AI ​​shown in the diagram could also be PSI. [Figure 15] This flowchart illustrates an exemplary problem-solving process that utilizes a common cognitive architecture implemented in an AI system, where the AI ​​could also be PSI. [Figure 16] This flowchart illustrates an exemplary problem-solving process that utilizes a common cognitive architecture implemented in a collective network of AI systems, where the AI ​​could also be PSI. [Figure 17] This is a flowchart illustrating an exemplary embodiment of the PSI technology of the present invention. [Figure 18] This flowchart shows an exemplary embodiment that utilizes the PSI capabilities of the present invention. [Figure 19] This flowchart shows an exemplary embodiment of a community-based safety mechanism of the present invention. [Figure 20]This flowchart shows an exemplary embodiment of recording all actions by PSI on a network. [Figure 21] This flowchart shows an exemplary embodiment of checking cognitive activity on a network. [Figure 22] This flowchart illustrates an exemplary embodiment of the competitive evolution of PSI's intelligence and / or performance. [Figure 23] This flowchart illustrates an exemplary embodiment of bringing multiple PSIs together on a network of PSIs with agreed-upon rules and methods for interaction. [Figure 24] This flowchart illustrates exemplary embodiments of baseline performance for individual PSIs, groups of PSIs, and / or entire networks of PSIs for various standardization tasks. [Figure 25] This flowchart shows an exemplary embodiment that generates different versions of individual PSIs and / or different combinations of PSIs. [Figure 26] This is a schematic block diagram showing an exemplary electronic computing device that may be used to implement an embodiment of the technology of the present invention. [Modes for carrying out the invention]

[0096] The same reference number refers to the same part throughout various diagrams.

[0097] definition Artificial intelligence (AI) - a non-human entity capable of performing actions that, in at least one area or in several respects, would be considered most human-intelligent.

[0098] Artificial General Intelligence (AGI) is traditionally referred to as AI capable of performing all (or almost all) intellectual tasks that an average human can perform. However, it should be clear that any AGI capable of learning and self-improvement will not remain at the AGI level for very long, but will progress rapidly and become a super-intelligent AGI capable of performing all intellectual tasks with abilities equal to or superior to those of an average human. Therefore, for the purposes of this explanation, "AGI" will refer to either a conventional AGI system or a "super-intelligent" AGI. In this explanation, the AGI described will be implemented by a system and related methods.

[0099] Highly Autonomous Artificial Intelligence (AAAI) – an AI that enables independent or semi-independent (supervised) intelligent actions. An AI agent. Individual AAAIs can be identified, customized, and brought into useful action states using the systems and methods of the present invention of AAAI. Groups of AAAIs can be combined in coordination with their intelligence to create an integrated AGI system. A sufficiently advanced AI agent can also function as an AGI system that may contain other less advanced AI agents within itself.

[0100] AAAI.com is a platform, company, website, and / or project that embodies this technology of the present invention and supports the development, customization, and use of AAAI agents, and also supports AGI arising from the combined actions, knowledge, or intelligence of multiple AAAIs using the collective intelligence of AAAIs and / or humans as defined in this technology and related technologies.

[0101] AI ethics - the ethics adopted by AI or AGI to explain what is right and wrong in a particular context.

[0102] The consistency problem arises when AI ethics are inconsistent with human ethics, resulting in AI or AGI engaging in actions that humans consider unethical and / or dangerous to individual humans or humanity as a whole.

[0103] Base AI - an AI, AI agent, AAAI, SLM, or LLM that is generally trained but not yet customized with information from individual users or information about specific tasks.

[0104] Collective intelligence (CI) is intelligence that emerges when multiple intelligent entities focus on solving a common problem, or when knowledge from multiple intelligent entities is combined to overcome the limitations of bounded rationality. Historically, collective intelligence has been the collective intelligence of humans, but AGI is based on the collective intelligence of both humans and AI agents (including PSI), and can arise from multiple AAAIs, with or without human participation in the system. Active CI arises when intelligent entities (e.g., humans or machines) take useful measures to solve a problem or actively participate in other intelligent activities. For example, when multiple humans clearly communicate to an advertiser the type of advertisement they want to see, humans are exhibiting active CI. Passive CI arises from analyzing the behavior of intelligent entities (e.g., humans or machines), even when such behavior is not directly related to solving a problem using analysis. For example, passive CI is exhibited when an AI or other system analyzes which web pages a group of humans visits on the web and then uses the results of that analysis to direct targeted advertisements to the humans.

[0105] Ethics / Values ​​("Ethics") - A subset of knowledge that provides purposefulness to intelligent entities, constraining acceptable actions or behaviors based on what is asserted as "right" or "wrong" in a particular context. Specifically, ethics should be considered on the premise that intelligent entities can reason or logically calculate the best course of action to achieve goals or intentions consistent with ethical premises. Just as premises in a logical system must be accepted "as given," fundamental ethics or ideas about right and wrong must be accepted as premises, from which intelligent entities can propose logical and rational actions to realize those values ​​or ethics.

[0106] Hallucination / Artificial Hallucination - Large Language Models (LLMs), often generative AI chatbots or computer vision tools, perceive patterns or objects that are not present or perceptible to human observers, or produce meaningless, inaccurate, misleading, or false output.

[0107] Human ethics - the ethics asserted by humans that explain what is right and wrong in a particular context.

[0108] An intelligent entity or entity is a human being, an AI agent or system, a clone of an AI agent or system, an AAAI agent or system, and / or a clone of an AAAI agent or system that utilizes a computer system and participates in providing problems, subproblems, goals, and / or subgoals, and / or participates in any problem-solving activity toward problems, subproblems, goals, and / or subgoals. In the case of multiple intelligent entities within a single computer system, an intelligent entity also refers to a subprogram of an overall computer program that functions as an intelligent entity within a larger collection of simulated or programmed entities. A PSI is also an intelligent entity.

[0109] Large-scale language models (LLMs) are a type of AI that can take natural language as input and produce natural language as output. Typically, LLMs are trained on large datasets using machine learning techniques, and as a result, they can emulate intelligent conversations or other forms of interaction with humans in natural language. Variations of LLMs can also be trained to take language as input and produce images or visual representations as output. Alternatively, variations of LLMs can take images and visual representations as input and produce language and / or images and / or visual representations as output. For the purposes of this patent, image-based models do not necessarily have to take text as input or output, but all such systems are referred to as LLMs. Also, LLMs can operate as a type of AI agent and are sometimes referred to as such in the techniques of this invention. For the purposes of this disclosure, small-scale language models (SLMs) are also included in the definition of LLMs.

[0110] Machine learning (ML) is a field that deals with developing AI by enabling machines to teach themselves or learn such knowledge, rather than having knowledge explicitly programmed into them (as in the case of expert systems AI developed using classical knowledge engineering methods).

[0111] Narrow AI refers to AI that performs at a human or superhuman level in relatively limited areas, such as playing games, brewing beer, or analyzing legal contracts. Narrow AI is contrasted with AGI, which can perform all intellectual tasks at a human level. Some AIs are more limited than others; for example, driving a car requires more general skills than playing chess, but it would not be equivalent to the capabilities of AGI.

[0112] Prohibited attributes – requests, goals, problems, terms, phrases, questions, answers, solutions, information, etc., that are judged or set as unlawful, immoral, unethical, dangerous, deadly, etc.

[0113] Safety - Generally speaking, concerns about human safety and survival are separate from ethics and values.

[0114] Safety features refer to aspects of the design or operation of the technology of the present invention that improve the safety of one or more people. In many cases, they overcome the alignment problem by helping to increase the probability that the ethics of AI are consistent with human ethics.

[0115] Training / Tuning / Customization - Traditionally, the term “training” has been used to describe training a network (e.g., an LLM) to operate intelligently. Tuning refers to the activity of fine-tuning the trained base model, usually to function well on a particular task. Customization refers to a broader range of activities, including, but not limited to, training and tuning the AI ​​to uniquely fit the purposes of a particular user or application. For the purposes of this explanation, training, tuning, and customization are used interchangeably, under the understanding that while the technologies differ and the degree and type of effort involved differ, all three objectives involve adapting the AI ​​to operate more intelligently or uniquely for a particular user or application.

[0116] Weights / Network Weights - In the field of machine learning, many systems learn by tuning weights in a neural network architecture, which can be represented as a network of nodes, or as links between nodes. The weights of the links connecting two nodes may, for example, correspond to the strength of the relationship or connection between what the nodes represent. These weights can also represent excitability or inhibition between concepts, in a neural network representation, for example. The learning of an entire AI system, such as an LLM, or more generally any AI agent, learning using error backpropagation, transformer algorithms, or machine learning methods to establish and correct the connection strength between nodes (also called "parameters" in some models), can be represented as a matrix of numbers corresponding to the weights between nodes in the network. In this description, weights / network weights refer to this numerical information, which is often stored in a matrix or vector representation, but not necessarily. By combining, manipulating, or otherwise changing this numerical information, the learning, knowledge, or expertise, and behavior of the system can be altered.

[0117] The technology of the present invention comprises a system and method for implementing Personalized Superintelligence ("PSI") that surpasses all existing forms of AI in both scope and intelligence. This PSI is self-improving and thus becomes significantly more intelligent than the owner who creates it. However, due to the unique creation method described in this patent, the PSI is completely secure and exclusively for the owner's service. Secure superintelligence by design is the essence of this patent. Unlike any previously created AI system, the technology of the present invention delivers intelligence levels, security, and useful advantages far exceeding current state-of-the-art AI assistants.

[0118] You can think of PSI as your personal, super-intelligent AI agent. It follows your instructions and, over time, learns about you and what you want. It is far more effective and efficient than most everyday online tasks. It provides excellent advice. And, most importantly, it can understand you, relate to you, and think like you. If you wish to extend PSI's capabilities, it is easy to exchange data with friends, or buy or sell data on online marketplaces, then mix that data with PSI, resulting in additional intelligence, skills, and knowledge. Over time, PSI can acquire the wisdom of billions of humans and AI agents. And after acquiring that knowledge, it can run simulations based on your goals and objectives, further improve them, and operate on your behalf as you like (with your permission).

[0119] PSI, acting as an investment advisor and agent, can help you increase your material wealth. You'll also have the freedom to use your time for spiritual, artistic, or other pursuits.

[0120] You can clone and lease PSIs. You can package and sell their data and intelligence. Thousands of versions of PSI can handle thousands of different tasks and missions simultaneously for you, under the command of other cloned PSIs, and following your direction and working for your benefit.

[0121] These are some of the benefits that the technology of this invention brings to individual owners. For society as a whole, multiple PSIs, pooling their intelligence and participating in a PSI network, can function as planetary intelligence, benefiting all people and the entire planet.

[0122] While the aforementioned devices satisfy their respective specific purposes and requirements, the aforementioned devices or systems do not describe a secure PSI that enables the implementation of PSI using an intelligent agent to develop and continuously improve PSI for each human owner. Furthermore, the technology of the present invention overcomes one or more of the disadvantages associated with the prior art.

[0123] There is a need for a new and novel secure PSI that can be used to perform PSI, using an intelligent agent to develop and continuously improve the PSI for each human owner. In this regard, the technology of the present invention substantially satisfies this need. In this respect, the secure PSI according to the technology of the present invention substantially deviates from the conventional concepts and designs of the prior art, thereby providing an apparatus primarily for the purpose of performing PSI using an intelligent agent to develop and continuously improve the PSI for each human owner.

[0124] The following description includes specific details such as particular embodiments, procedures, and techniques, not limitingly, but for illustrative purposes, in order to provide a complete understanding of the art of the present invention. However, it will be apparent to those skilled in the art that the art of the present invention can be carried out in other embodiments departing from these specific details.

[0125] The technology of the present invention provides technical effects, benefits, and solutions in a technical embodiment of multiple customized AAAI systems communicating through a collective intelligence neural network, which can be recognized as being realized in combination with all AAAI systems, each utilizing a common cognitive architecture, which includes one or more problem-solving protocols to generate one or more solutions or answers to a problem request and to provide the solutions or answers to a user who approves them. The customization of the AI ​​system resulting in the AAAI includes input from human users to train the AI, AAAI, or PSI. For further technical contributions or solutions, multiple customized AAAI or PSI systems may include one or more cloned AAAI or PSIs, each of which can be customized independently of the parent AAAI or PSI and independently of other cloned AAAI or PSIs in the same system.

[0126] Another technical contribution and solution is made for the creation and / or realization of PSI, which is self-improving, dramatically more intelligent, and at the same time safe and ethical. This technical contribution can further be provided in the creation of PSI by acquiring a previously customized base-level AI agent and utilizing differentially weighted media information related to human users in the training of PSI, which is realized in combination with training information acquired by all communities of additional AIs and / or PSIs communicating on a collective network.

[0127] Another technological contribution and solution would be for all AIs and PSIs to communicate on a collective network, enabling community-based safety features and / or learning, where all AIs and PSIs agree on a set of rules governing safety and ethics that are human-consistent values, preventing any harmful actions by any malicious PSI.

[0128] It can be recognized that the technology of the present invention falls outside the scope of the exclusion clauses and interpretations of abstract ideas of computer programs. This can be seen in part in the technical contributions and solutions provided by the technology of the present invention, the use of specific training inputs outside of a computer, and the provision of solutions or answers outside of a computer.

[0129] Thus, one reason why AGI is difficult to understand is that specific knowledge and expertise from various fields need to be creatively combined through invention to realize AGI. Another reason why the development of AGI is not self-evident is that almost all AI researchers are focusing on attempts to improve existing narrowly defined AI systems using more complex and large-scale machine learning approaches.

[0130] Based on the fact that AGI has been rejected by thousands of attempts despite significant financial investment, and that specialized knowledge in relatively ambiguous fields needs to be combined with mainstream AI approaches in the present invention, we strongly assert the novelty and creativity of the present invention.

[0131] The present invention describes a system and method for not only realizing AGI, but also for realizing it rapidly, as a primary objective, and safely.

[0132] It becomes possible to influence the evolution of AGI in a positive direction. The best way to do this is to adopt the safest possible path to the development of AGI and ensure that humanity follows that path. Furthermore, the best way to ensure that humans follow a safe path is to demonstrate that the safe path to the development of AGI is also the fastest path to the development of AGI, and therefore the most desirable path. These considerations, namely the desire to unravel the fastest path that is also the safest path, are the motivation for developing the technology of this invention.

[0133] While the aforementioned devices satisfy their respective specific purposes and requirements, the aforementioned devices or systems do not describe a safe, scalable, general-purpose artificial intelligence system or method that enables scaling by using a combination of human users and multiple AI systems to train other AI systems by combining the value and ethical knowledge of human users and multiple AI systems for training. Furthermore, the technology of the present invention overcomes one or more of the disadvantages associated with the prior art.

[0134] There is a need for new and novel systems and methods for safe and scalable general-purpose artificial intelligence that can be used for scaling by using a combination of human users and multiple AI systems to train other AI systems by combining the value and ethical knowledge of human users and multiple AI systems for training. In this regard, the technology of the present invention substantially satisfies this need. In this respect, the systems and methods for safe and scalable general-purpose artificial intelligence according to the technology of the present invention substantially deviate from the conventional concepts and designs of the prior art and thereby provide a device developed primarily for the purpose of scaling by using a combination of human users and multiple AI systems (including, but not limited to, PSI) to train other AI systems by combining the value and ethical knowledge of human users and multiple AI systems for training.

[0135] The following description includes specific details such as particular embodiments, procedures, and techniques, not limitingly, but for illustrative purposes, in order to provide a complete understanding of the art of the present invention. However, it will be apparent to those skilled in the art that the art of the present invention can be carried out in other embodiments departing from these specific details.

[0136] The following description includes specific details such as particular embodiments, procedures, and techniques, not limitingly, but for illustrative purposes, in order to provide a complete understanding of the art of the present invention. However, it will be apparent to those skilled in the art that the art of the present invention can be carried out in other embodiments departing from these specific details.

[0137] The technology of the present invention provides technical effects, benefits, and solutions in a technical embodiment of multiple customized AAAI (or PSI) systems communicating through a collective intelligence neural network, which can be recognized as being realized in combination with all of the AAAI systems (or PSIs) that each utilize a common cognitive architecture, including one or more problem-solving protocols that generate one or more solutions or answers to a problem request and provide the solutions or answers to an approving user. The customization of the AI ​​system resulting from the AAAI (or PSI) includes input from human users to train the AI ​​or AAAI (or PSI). For further technical contributions or solutions, multiple customized AAAI systems may include one or more cloned AAAIs (or PSIs), each of which can be customized independently of its parent AAAI (or PSI) and independently of other cloned AAAIs (or PSIs) in the same system.

[0138] Another technical contribution and solution would be to create scalable AGIs that utilize human input in training and customization, thereby conferring human ethical attributes to AAAI, PSI, and / or AGI, in order to create them faster and more securely.

[0139] Another technical contribution and solution is to scalably train AI systems and / or agents with combinations of safety and ethical information from many individual AI agents (or PSIs) to realize a representative and statistically valid sample of human ethics and values ​​that cover a wide range of scenarios. A further technical contribution may be seen as the technology of the present invention including a method for combining information from many agents to construct an optimal combination of such agents in order to provide scalable training of AI, PSI, or AGI.

[0140] It can be recognized that the technology of the present invention falls outside the scope of the exclusion clauses and interpretations of abstract ideas of computer programs. This can be seen in part in the technical contributions and solutions provided by the technology of the present invention, the use of specific training inputs outside of a computer, and the provision of solutions or answers outside of a computer.

[0141] The AAAI approach to developing secure AGI is fundamentally a collective intelligence (CI) approach. The source of intelligence is not a monolithic LLM, SLM, or hyper-advanced AI, but rather a collection of intelligent agents that can be both human and AI. The component subtasks in developing AGI include, but are not limited to, training individual AI agents or PSIs, effectively and efficiently combining knowledge from different agents (including, but not limited to, subjective values ​​and ethical knowledge), scaling the AGI, and continuously improving / updating the AGI.

[0142] Current approaches (such as human-feedback reinforcement learning (RLHF) and constitutional learning) are unable to effectively and scalably train AI to be ethical and safe. The technology of the present invention describes a scalable system and method that is superior to current approaches. In one embodiment, the technology of the present invention may include a combination of safety and ethical information from many individual AI agents to realize a representative and statistically effective sample of human ethics and values ​​that cover a wide range of scenarios. The technology of the present invention may include a method for efficiently covering a wide range of ethical situations and dynamically responding as new situations arise. Furthermore, a method is presented for combining information from many agents to construct an optimal combination of such agents. These methods can be used not only to improve safety using ethical knowledge but also to create super-intelligent systems that combine many other types of knowledge. Safe AGI and superintelligence can be realized using the collective intelligence approach described in this description of the technology of the present invention. A detailed scenario using META® as an example illustrates one exemplary embodiment of the technology of the present invention.

[0143] Furthermore, methods for dynamically updating knowledge are presented. Successful embodiments of the present invention increase the likelihood that AI, PSI, AGI, and superintelligence will remain consistent with human values ​​even when such systems significantly exceed human intelligence.

[0144] Advanced Autonomous Artificial Intelligence (AAAI) is a set of systems and methods for developing general artificial intelligence and super-intelligent general artificial intelligence (collectively, "AGI") in a rapid and secure manner for the benefit of humankind. In contrast to other approaches to AGI development, the technologies of the AAAI invention provide a fast and secure path to AGI development by relying, at least initially, on the involvement of human thought (ideally millions) for the training, operation, and safety / supervision functions of the AGI.

[0145] The technology of the AAAI invention can first realize AGI by enabling users to customize and clone their own AI or PSI. These customized AIs (AAAIs) and / or PSIs participate in problem-solving activities and other intellectual activities on a network consisting of other AAAIs, PSIs, and humans. While each AAAI (or PSI) may individually lack a wide range of skills and knowledge for AGI, collectively, the AAAIs form an AGI that (first with the help of humans on the network) quickly surpasses the average human capacity in all intellectual activities.

[0146] Some aspects of the technology of the present invention may include: 1) a system and method for customizing AI with a user's unique knowledge, skills, and ethical values; 2) a universal problem-solving architecture that enables AAAIs to productively interact with each other and with humans on intelligent tasks; 3) a network in which the interactions take place; 4) a method for integrating the knowledge and ethics of individual AAAIs into an AGI; and 5) a method for continuous improvement of AAAIs and AGIs by learning to become smarter and more ethical over time. The involvement of humans as customizers of the AAAIs and as participants in the network is an essential feature of the technology of the present invention, which not only accelerates the development of AGIs but also makes AGIs safer by providing a mechanism in which the ethical values ​​of millions of people are adopted by and reflected in AGIs.

[0147] One embodiment of the AAAI system of the present invention, as shown in Figure 1, focuses on safety and is implemented using five subsystems and associated methods. The five subsystems of the AAAI system are 1) AAAI Customization, 2) AAAI Architecture, 3) AAAI Network, 4) AAAI Integration, and 5) AAAI Improvement. The acronym SCAN-II (Safe, Customizable, Architecture and Network--Integrated and Improving) describes the technology of the present invention in exemplary embodiments. Other combinations of subsystems and variations of each subsystem are also possible. Safety functions are designed to be tailored to each subsystem in order to provide redundant safety checks when one or more subsystems are omitted from a particular embodiment.

[0148] The five subsystems of the AAAI system are: 1) Base-level large-scale language models (LLMs), small-scale language models (SMLs), or other AI systems can be customized to reflect individual knowledge, individual groups, or organizations and can be designated as highly autonomous artificial intelligence (AAAIs). 2) Customized AAAI can be enabled to participate in problem solving using a universal problem-solving architecture that is compatible with both humans and AI agents. 3) AAAIs with enabled problem-solving participate in problem-solving activities, and these activities are not limited to, On a network of intelligent agents, they perform planning, problem-solving, and other types of multi-stage cognitive activities. Based on the goal / sub-goals, generate and select operators that reduce the difference between the current state of problem solving and the desired state, Setting sub-goals with the aim of achieving the main goal, Use the hierarchy until actionable goals that can be activated by the operator are set. Analyze auditable records to determine recommendations for improving the problem-solving process that leads to solutions for objectives / sub-objectives, Includes. 4) Multiple AAAIs or PSIs on a network can be integrated to achieve AGI, or the AI ​​can perform intelligent (or superhuman) actions across a wide range of tasks. 5) Individual AAAIs, problem-solving networks, and / or integrated systems of multiple AAAIs are continuously improved by various means, which include, but are not limited to, redirecting the efforts of individual AAAIs and / or integrated AGIs to tasks that improve the system and / or components of the system.

[0149] A subsystem or a new subsystem may include one or a combination of any of the following 1) to 5): 1) Safety / Ethical Checks - Compare objectives or sub-objectives to a list of prohibited attributes and assign ethical values ​​based on the comparison results. Check objectives / sub-objectives against a list of prohibited attributes. Combine value / safety information from AAAI using a set of task criteria approved by AAAI, approved by users, regulatory bodies, or human users. Define or use thresholds for objectives / sub-objectives to determine whether the ethical value is unsafe, unethical, safe, or ethical. Determine whether a set of individually safe objectives / sub-objectives are unsafe or unethical when considered cumulatively. If an objective violates ethical standards, determine whether the resulting violation reflects a predictive assessment. Record any and all safety / ethical check activities in an auditable record format. 2) AAAI Matching - Detect and identify additional AAAIs or PSIs that each have criteria related to the criteria of one or more objectives or sub-objectives. 3) Memory and / or Improvement - Record activities, compare whether progress toward problem-solving was successful or unsuccessful, and decide which activities to keep active or forget. 4) AAAI Learning – Learning that includes a procedural learning process that utilizes information provided by intelligent entities such as a computer, AAAI, or a human user equipped with PSI. This involves recording activities, comparing whether progress toward problem solving was successful or unsuccessful, and deciding which activities to keep active or forget. It also involves assigning credit or responsibility values ​​to groups of content for problem-solving activities. The set of prompts provided to the user and the information received based on those prompts. The AAAI is updated with the groups of content determined to be active. The groups of content may, but are not limited to, a set of prompts provided to the user and the information received based on those prompts, all of which are recorded in an auditable recording format. Optionally, a problem-solving activity may include groups of content.

[0150] Example User Scenarios It may be useful to describe several user scenarios that give a sense of how the technology of the present invention may work in some aspects of multiple embodiments. An exemplary process is shown in Figure 2.

[0151] In one embodiment, the user “visits” AAAI.com via their computer, mobile phone, PDA, or goggles. AAAI.com will interact with the user via a web-based interface, a phone app, custom software for PDAs, or a metaverse / virtual reality environment. The mode of interaction may be physical via a keyboard, mouse, or gesture-based interface; voice-based via microphone input coupled to a natural language understanding and generation system; or video-based if the user becomes an avatar in a virtual reality environment or metaverse.

[0152] The initial interaction would involve setting up a user account, which may be free or paid. This would include an account name and password, or other authentication mechanisms. Other authentication mechanisms may include, but are not limited to, biometric forms of ID such as fingerprint, facial recognition, or voice recognition, and / or multi-factor authentication mechanisms such as software or hardware authentication codes residing on a separate security device or on one of the user's existing devices.

[0153] For security purposes, all communications between the user and the AAAI system may be encrypted via VPN, and / or other encryption and security methods known in the art of programming may be used.

[0154] AAAI.com may request users to set up payment functions via credit card, PayPal, Venmo, blockchain, ACH, or other payment mechanisms. These payment functions will enable the bidirectional transmission of funds, payments, and / or deposit balances. This will allow the AAAI system to transmit funds from the user to AAAI.com and from the AAAI system to the user when it needs to pay users or deposit funds into their accounts for the results of its work, or when payments are needed between users on the AAI network and / or between AAAIA intermediaries.

[0155] In one embodiment of a particular design, AAAI.com may have interfaces with other companies and vendors that users may use, including, but are not limited to, Facebook, Instagram, Reels, Amazon, Apple, Microsoft, Google, and YouTube®.

[0156] In the initial interaction with the user, AAAI.com will then, in response to user requests, engage in dialogue or other interactions with the user (which may include presenting the user with menu options, lists, graphics, sliders, buttons, and other user interface controls in a GUI, text, haptic, audio, or VR-related manner) to determine the user's goals and objectives when using the AAAI system.

[0157] For example, some of the purposes a user might have when using AAAI.com may include creating and customizing their own AI (known as AAAI) for several purposes, which may include, but are not limited to: To serve users as an advisor, teacher, or colleague. Representing a user in negotiations, interactions, discussions, and transactions with other users, or with other users' AAAIs, or with vendors and other companies. This involves working on behalf of users for compensation or as a volunteer effort, and such work includes online intellectual work, advice, or problem-solving work across a wide range of tasks. This involves duplicating or "cloning" a user's AAAI so that several or more cloned AAAIs can work in parallel on behalf of the user, including interaction, education, and improvement of each other so that the cloned AAAIs improve their own knowledge, skills, and capabilities. It functions as a legacy AAAI that can continue to interact with the real world, including potentially comforting surviving relatives and friends after the owner's death. To contribute to AAAI.com's knowledge, ethics, and commitment to AGI, and to improve the base level of AI or AGI that AAAI.com can provide to users before those users add their own unique customizations. By collaborating with other users' AAAIs and contributing ethical information and values ​​to AGI, you can help ensure ethical and safe behavior within AGI and participate in monitoring, review, oversight, and voting processes that can help ensure AGI remains safe and ethical.

[0158] During user interaction, the AAAI system will also identify available constraints and resources to customize the user's AAAI. Some, but not limited to, these constraints and resources may include:

[0159] The amount of training and / or supervision time that the user needs to dedicate to customizing that AAAI. The amount of financial resources that users are willing to dedicate to customizing that AAAI. Availability of social media information such as Facebook profiles and timelines, Instagram profiles and history, Reels, TikTok, and YouTube® videos, tweets, text content and history, emails and email history, cookies collected by advertisers, blog posts, articles, books, patents, recordings and video recordings, photographs, and information about users or third parties and / or other information collected by them that may be used to train, tune, or customize the user's AAAI. The availability and use of Myers-Briggs personality tests, assessments of skills and knowledge, standardized tests, examinations, certifications, and other types of assessments and questionnaires that may be provided to users online (or have already been provided). Availability and use of other knowledge bases and training data from users on the AAAI platform, which may be used to train, tune, or customize the user's AAAI. Other human users and / or their AAAIs are available to help train, tune, or customize the user's AAAI. Other texts and information, individual texts, and libraries selected by the user or the system for the purpose of training the user's AAAI. For example, the Bible, Quran, Dhammapada, Mahabharata, or other spiritual / ethical / religious texts may be selected to train the AAAI based on the user's religious preferences. If the AAAI is primarily used to solve plumbing problems online, books on plumbing may be selected. Even if these materials are part of the base AAAI provided to the user, highlighting specific texts or subsets of information for additional training may result in the user's AAAI behavior being more reflective of, for example, how a plumber or a Muslim or Christian might behave.

[0160] In addition to defining objectives, resources, and constraints through interactive dialogue or other interactions with the system, the user or system may wish to specify other technical parameters that affect the training or customization process. These parameters may include, but are not limited to, the following: The type of training, tuning, or other ML algorithm used. The type and size of the training dataset(s). The extent to which training materials are "deleted," formatted, labeled, or otherwise processed before customization begins. The number of "epochs" or iterations trained by the learning algorithm(s). The sophistication and type of the base model(s) to be customized or trained. The timeframe required for training. For example, a timeframe that needs to be completed in one minute, one day, or one week. This may affect the expected results in terms of the costs and resources used. "Temperature," or other parameters specific to various machine learning algorithms that may influence what and how they are learned. This includes, but is not limited to, how literally or how deviant or "creative" the customized AAAI response is. This concerns whether to use "One Shot," "Fu Shot," or wide-area training. The amount of human and / or AI monitoring used in the customization process.

[0161] Once a user's AAAI or PSI is customized, the user can clone it and / or have it work on their behalf on an online network. The user's AAAI can start working to (for example) make travel arrangements on behalf of the user, provide advice, interact with other AAAIs, participate in collective AGI efforts by contributing to problem-solving and ethical information, and potentially earn money on behalf of the human user.

[0162] Simple Exemplary Embodiment Figure 3 shows one exemplary embodiment of a system and method for creating ethical and safe general artificial intelligence from AAAI and human collective intelligence. This simple embodiment is compatible with all of the specific enterprise and platform scenarios outlined above, and furthermore, with many other potential integration scenarios.

[0163] A user (a human, AAAI, or other intelligent entity) visits the AAAI.com website. The website provides the user with information and offers the user two actions: sign up (b) or log in (c).

[0164] If a user chooses to sign up, a dialogue begins, extracting the user's values / ethics (d), user goals and objectives (e), and user budget for time (f) and funds (g). All users must allocate some time (f). Users have the option to create a free AAAI or allocate a funding budget.

[0165] If a user allocates a funding budget (g), they are given the opportunity to purchase pre-trained AAAIs or training modules (h) that possess a unique personality (i), skill (j), expertise (k), or knowledge (l). The user also has the opportunity to purchase training from other AAAIs on the network (m).

[0166] After deciding on the allocation of time (and optionally the allocation of the funding budget (h,i,j,k,l,m)), the user proceeds to an overview of the creation process, and is then asked for user permission (n) to log in to and use existing social media, Twitter®, and other vendor accounts to collect user data for the "one-click" training of the user's AAAI. After the user chooses to use specific data (or no data), the user instructs the system to create the AAAI with one click (o). The AAAI is a commercially available LLM (e.g., GPT X, BARD, Llama, Gemini, Grok, or any of the closed-source or open-source AI agents) that is automatically trained / tuned against a dataset prepared from all user data permitted by the user. If no data is permitted, the AAAI is simply a "commercial" LLM.

[0167] Here, AAAI begins learning through training (p) using various training datasets and modules (hm) and its existing AAAI knowledge (p1). There are two main learning methods: automatic (q) and human (r).

[0168] Automated learning includes, but is not limited to, learning by interacting with a copy of itself (s) and learning through interaction with other (optionally, supervised) AAAIs (t).

[0169] Human learning involves interaction with humans (either the owner (u) or other humans (v) on the network).

[0170] Both humans and AAAIs can supervise the learning of the AAAI. After each (automatic or human) learning interaction, the system attempts to improve the AAAI's performance through further prompt correction, tuning, and / or training. Based on many cycles of human and AAAI input aimed at educating and improving the AAAI, the user's AAAI becomes smarter.

[0171] Users can purchase additional training modules (h-m) at any time, which have been proven to improve AAAI's capabilities.

[0172] Humans set the performance standard (w), and then AAAI enters an operational state (x).

[0173] Once operational, AAAI can visit the WorldThink tree (y) and browse (z).

[0174] AAAI can enter the tree as either a worker (a1) or a client (b1).

[0175] Workers are automatically matched with tasks (c1), or workers can select a specific task by searching (d1) or using a link (e1) from the browsing tree. Once a worker receives a task (f1), they participate in the problem-solving module (g1) until a solution is found (h1) and payment is made (i1), or the user saves their account deposit for the work done and exits the tree (j1).

[0176] The client (b1) can specify the purpose (k1) to be combined with the values / ethics (d), and the prior goals and objectives (e) that the system should solve.

[0177] The client can request to use only their own AAAI, in which case problem solving is free. Alternatively, the client can use the capabilities of the entire network of AGIs, in which case the system will pay the individual AAAIs for the work and deliver the solution (cost + markup) to the client, which will be deducted from the client account (l1).

[0178] Furthermore, the system can place non-profit humanitarian tasks, ecologically oriented tasks, and even tasks that are part of planetary intelligence on the WorldThink tree (m1).

[0179] When a client creates its AAAI, it may (optionally) allow the system to use a copy of its AAAI and data free of charge for these purposes in exchange for maintaining and operating a free AAAI network (n).

[0180] Additional comments on the exemplary embodiment shown in Figure 13 Here, we offer additional comments on the various elements of Figure 13. This includes, but is not limited to, several potential integration points with the exemplary partners mentioned above.

[0181] The “Website” (a) may be hosted on data center services provided by Amazon AWS, Microsoft Azure, Google Cloud, Apple Cloud, or Nvidia, or it may be implemented natively on the platform of any of these major technology companies. The “Website” may also be an “App” on the App Store or other app marketplaces. It may also be a government-funded non-profit organization or other globally accessible technology that allows it to directly or indirectly link to a portion of the attention of all humanity wishing to participate. It may also use a browser plugin, thereby allowing AAAI to learn from users as it typically performs tasks over the internet, with the plugin recording its activity, creating training files, and using these files to train AAAI. The “Website” may also be an API or other means for directly connecting AAAI or any non-human intelligence entity to the network.

[0182] Sign up (b) or

[0183] Login(c) may be performed via Facebook, Instagram, Apple, Microsoft, Google, YouTube®, TikTok, Amazon, or any other partner ID scheme. Multi-factor authentication and all best identity and security practices can be enabled. In the case of browser plugins or apps, logins to these technologies may function as logins to AAAI accounts.

[0184] Values ​​and ethics (d) are customized to the user and retrieved through a series of dynamically generated scenarios based on user responses. Data from partners, including navigation and click data, online posts, tweets, text, as well as emails, videos, and other user data, are analyzed in terms of behavioral patterns, i.e., actions, speech, or interactions. These are translated into moral codes or ethical value systems and can also be used as part of an ethics / values ​​profile. Values / ethics and goals / objectives (d) can be combined with the client's objectives (k1) to create or discover matching tasks on the WorldThink tree (y) proposed in the problem-solving system (g1) or (potentially, solved).

[0185] Goals and objectives (e), along with a budget of time and / or funds (f,g) allocated to achieve the objectives, are retrieved through a series of interactions with the system and / or custom interactions. The budget refers to the overall resource budget, including user time and user funds, which can be allocated towards training, supervising, and improving the user's AAAI. Goals and objectives help determine the initial parameters for AAAI creation and identify training modules (h) or other knowledge (i-m) that can create the most useful AAAI for the user's objectives. Data from partners reflecting user preferences and other user behavior information may be used by the system to help infer or predict the user's goals and objectives.

[0186] Time (f) refers to the time a user can dedicate to training and supervising the user's AAAI and / or to the user's problem-solving on the problem-solving network. In particular, by supervising the AAAI in areas where it does not function (e.g., areas where the AAAI lacks the knowledge to complete problem-solving on its own), the user can ensure that the AAAI meets client goals and expectations. Also, but not limited to, presenting the problem and breaking down large tasks into smaller ones by defining goals and sub-goals is a way a human user can assist the AAAI in problem-solving. Generally, providing human expertise in areas where the AAAI is not as skilled as a human improves the overall problem-solving and the overall effectiveness of the AGI network.

[0187] (g,i1) "Funds": Apple Pay, WePay, Amazon, Google Pay, or any payment solution from a vendor supporting the payment solution, and furthermore, it may be blockchain, credit card, ACH, and other solutions. The payment (i1) is shown to be debited from the client's account (l1), but will naturally be deposited into the worker's account. Generally, a user's account can be seen as both a client's account and a worker's account, and in certain cases, depending on the role of the user (or the user's AAAI), both deposits and withdrawals are permitted. That is, in some cases, the user may be a client making payments to the system or other specific AAAI for its services. In other cases, the same user may be a worker receiving fees for the user's (or the user's AAAI's) services. The Funds module (g) enables features such as setting up payment methods, setting budgets for automatic payments, limiting authorization to spending only X dollars without additional approval from the user's AAAI, and other payment-related functions well known in the art.

[0188] (h,i,j,k,l) ​​- Training modules (h) may be provided by AAAI.com or by third-party partners (m), including, but not limited to, any of the partner candidates and technology companies listed above. Training modules may cover different knowledge areas spanning (i) personality, (j) specific skills (e.g., plumbing, law, accounting), (k) expertise (e.g., consulting), and (l) knowledge (e.g., historical knowledge, knowledge of the track record of a particular company or organization, cultural knowledge).

[0189] (m) Purchaseable AAAI training is a specific type of knowledge already learned by other AAAIs that can be transferred to new user AAAIs. Such knowledge may be packaged in the form of modules (e.g., modules on accounting) or in a format specific to another AAAI(i).

[0190] (n) Permission refers not only to the user's permission to access all data on certain other vendor (or partner) sites (e.g., "All My Facebook Data"), but also to the user's ability to log on to various sites and conduct trading activities (including, but not limited to, the ability to trade up to a certain amount through payment mechanisms), and also to permission to access their own AAAI. Authorization may also include authorizing the system to create clones of the user's AAAI for non-profit purposes and for the purpose of gathering knowledge from individual AAAIs to create AGI-level AI.

[0191] (o) One-click creation is a non-limiting example that provides a user with an easy and fast way to customize an AAAI using data automatically collected from all locations where the system has given permission to access the user's data. It can be recognized that other means may be available by the technology of the present invention to customize an AAAI. For example, if a user has given permission (n) to access their Facebook data, “One-click creation” (o) will either download the data from Facebook if Facebook is a partner that has an API to download that user's data, or log in as the user to the user's Facebook account and “scrape” the relevant data from the user's account. The system will then automatically parse the collected data and transform it into a suitable dataset for training / tuning a base AI such as an LLM (e.g., GPT X). The system will then train / tune the LLM to generate a customized AAAI. That AAAI may be improved and refined by additional training / tuning and interaction with the user and / or other AAAIs.

[0192] (p) Training refers to the process by which AAAI is trained or adapted to data, including (but not limited to) feedback from users, other humans, and / or AAAI.

[0193] (q,r,s,t,u,v)-Automated learning can proceed very quickly without the need for human user intervention. Typically, this would involve how an AAAI interacts with copies (or variations) of itself, and even (optionally) with other AAAIs to improve through interaction. Occasionally, human involvement in the training loop (t) can help accelerate the progress of automated learning where it would not proceed efficiently on its own. Learning can also be done through rapid iterations during interactions (s) between AAAIs. Just as a chess AI can evolve quickly from beginner to grandmaster ability by simulating millions of chess games very quickly, an AAAI can evolve its capabilities quickly by simulating millions of interaction scenarios. The funding budget (g) can be limited depending on the extent to which such simulations require funding for the associated computations.

[0194] Humans (or AAAIs) can specifically target the type of automated learning scenario, allowing the AAAI to be trained in a narrow area of ​​expertise or a more general area of ​​expertise, depending on user needs and resources. Partner integrations enable AAAI training to be guided by working backward from the types of jobs available on partner marketplaces (e.g., Amazon's Mechanical Turk), resulting in a focus on learning the skills that will generate the highest revenue when the AAAI is deployed to an available job. This "just-in-time" learning / training / tuning approach generates an AAAI "on demand" with the required skill set at any particular point in time.

[0195] A person (r) interacting with an AAAI could be the owner (u) of the AAAI (in which case there is usually no charge as the user trains their own AAAI), or another professional person (v) who is an expert and can charge a fee for training an AAAI to instruct the human and / or automated training / tuning of the AAAI for users who do not wish to spend time on it or who lack the necessary expertise.

[0196] (w,x) - A user (the owner of an AAAI) can set various performance criteria (w) that must be met before they can bring their AAAI into an "operational state" (x) and become accessible for performing tasks on the WorldThink tree. Some of these criteria may also be set by partners and other third parties who have minimum requirements that must be met before an AAAI can operate on its platform, product, application, or network.

[0197] The (y,z,a1,b1)-WorldThink tree (y) is a massive tree data structure consisting of many subtrees, representing all problems and tasks that have been, are being, or proposed for the entire AGI system. This tree is browseable (z). Individual AAAIs and / or humans can engage in specific tasks within the tree. The tree structure provides an auditable record of all problem-solving activities, which is also useful for learning through the proceduralization mechanisms described above. When interacting with the tree, there are two main roles that an agent can play: (a1) worker or (b1) client. Regulatory bodies or third parties that monitor the system's performance, security, and / or ethics have another role that can be considered a special type of client. Workers are generally involved in solving unresolved or sub-problems on the tree. Clients are generally involved in defining the problems, goals, objectives, and other parameters (e.g., compensation, budget, timeframe, success criteria, quality metrics) that constrain problem-solving.

[0198] (c1) Workers are automatically matched to tasks on the tree based on data about the worker (which may include, but is not limited to, the worker's skills, expertise, knowledge, past experience, ratings, rates or costs, availability, and response time). Workers may be human or AAAI. Workers may be matched and recruited by partners (e.g., LinkedIn®, Mechanical Turk, Facebook) who have data about human users and / or their AAAI. Workers may also be recruited through online advertisements offering work on a variety of tasks, and potential workers may be targeted using advertising targeting mechanisms that are well known in the art or described in other patents by the applicant.

[0199] (d1) Workers may also search the WorldThink tree to find tasks that interest them or match their skills. This search can be done manually or automatically (as in the case of AAAI workers).

[0200] Additionally, (a1) workers and (b1) clients can browse the WorldThink tree (z) to find tasks or problems of interest. A worker or client can then click on a link (e1) in a specific part of the tree to obtain more detailed information about problem-solving issues arising (or proposed) for that part of the tree. A worker may click to apply for work, or, based on existing problem-solving work, may propose additional tasks as a client.

[0201] (f1, g1, k1) - The client can interact with the system to specify specific goals, objectives (k1), and tasks that it wishes to achieve. Through interaction with the problem specification, the problem, tasks, and objectives are formulated (f1) and placed on the WorldThink tree (y) for problem solving using the problem solving system (g1).

[0202] The (m1) system has the ability to formulate specific goals, issues, and tasks relating to general efforts to help people or the planet. These can be worked on in a rewarded “profit” mode, or in a “non-profit” mode using cloned AAAI and volunteer human efforts. Some issues may relate to a universal goal (aka “Planetary Intelligence”) that would enable Global AGI to operate on a global scale, using its intelligence on behalf of the planet and its people. Various partner organizations (including non-profit organizations, government agencies, and charities) can “plug in” their tasks, issues, goals, and objectives here(m1).

[0203] (g1) The problem-solving system refers to the problem-solving architecture and system outlined by Newell and Simon (HPS) and improved by the applicant, the patent for the Online Distributed Problem-Solving System (ODPS) invented by the applicant, the WorldThink Whitepaper written by the applicant, and this PPA and other PPAs related to AAAI, together with modifications and variations that reflect different modes of compensation, payments, and operations.

[0204] To the extent that activities on specific other online work systems (e.g., Mechanical Turk) can be automatically mapped to a general applicant-improved HPS / WorldThink problem-solving framework, the overall problem and related problem-solving activities can be "imported" from partner sites and other sites, and that data can be fed into the WorldThink tree to improve its comprehensiveness.

[0205] To the extent that other applications, products, systems, and online features may be helpful in solving the problem (for example, when using a travel booking system, a robo-advisor app, a transit app, or an online ordering system), these capabilities are referred to as “operators” and can be invoked (similar to procedure calls in a programming language) to advance the problem-solving process. Thus, problem-solving does not depend on operators developed by humans working on the tree or by AAAI solvers, but may include either online or offline technologies or means to advance the problem-solving process, provided that these means can be referenced and / or linked through the WorldThink tree at the appropriate points in the problem-solving process.

[0206] (h1) When a solution is achieved, the client can review the solution before paying the reward (if any) for that solution. Alternatively, if the success criteria for the solution are automated, human client review may not be necessary, and the reward can be paid automatically when the success criteria are met. Depending on the preferences of the client and the worker, this automated approach can be implemented using blockchain technology or a more centralized means, using "smart contracts".

[0207] Once a solution is obtained and payment is made (optionally) (as some issues are non-profit or volunteer-based, or performed by the user's own AAAI), there may be an opportunity to obtain feedback from both the client(s) and the worker(s), in accordance with methods well known in the art. Furthermore, the solutions may be “chunked” and proceduralized so that the entire system learns solutions to specific problems and their key characteristics, and the solution paths may be indexed for search, accessible, and reused when similar problems arise in the future.

[0208] Optionally, royalties may be enabled to pay a fee to a user in the form of a royalty for the solution if the user reuses the user's solution or the user's AAAI solution. Such royalties can be paid (optionally) using a "smart contract" on the blockchain or by other payment methods.

[0209] (j1) Problem solving does not need to be completed in a single session. Partial progress toward a solution may be made, in which case the progress is saved when a human or AAAI solver terminates the problem solving system, and data is stored to trust the solver for the progress made so far, even if such progress does not reach a point where a reward can be paid.

[0210] The WorldThink protocol is a problem-solving architecture available from AAAI.com that functions as a universal problem-solving architecture, incorporating the universal architecture of HPS while adding features to overcome specific challenges.

[0211] In some embodiments, the procedural learning process may occur within a common cognitive architecture, schematically shown in Figures 4 and 5.

[0212] A common and universal problem-solving architecture can be illustrated by the following scenario, which refers to humans, but is generally applicable to any intelligent entity. 1) The problem description can be entered into AAAI. 2) Next, a human problem solver can be identified and adopted as a database or data source for human workers. 3) Qualified individuals or intelligent entities can be matched to the problem. 4) Use LLM or other means to translate the English descriptions of the problem task, objective, operator, and solution steps into the language of the Universal Problem-Solving Architecture. 5) Work on sub-problems can be delegated to different human problem solvers, allowing work on multiple aspects of a complex problem to proceed in parallel. 6) Combine solutions to various sub-problems into an overall solution. 7) Direct the problem solver's attention to the part of the problem tree that requires its work. 8) Pay workers a reward or wage for solving the problem and / or sub-problem(s). 9) Allow human users to accept or reject solutions and / or provide feedback to the solver regarding the solution for the problem and / or sub-problems(s).

[0213] Referring to Figure 5, the steps of problem-solving learning can be illustrated by recording, at each step of the learning process, the operator applied, the new state of the problem, the evaluation function used and its result, the current relevant objective / sub-objective, and any other information that differs from the previous step(s). To determine if the problem has been solved, the problem situation or state can be evaluated. If not, the problem-solving process, progress evaluation, and selection of the next operator to apply are then rerun using information from the latest problem state after the last step. The process can then return to the recording step.

[0214] If a problem is resolved, record the successful or unsuccessful solution for later retrieval, so as not to repeat previously resolved attempts and to inform you of previous unsuccessful problem-solving efforts.

[0215] Successful solutions and unsuccessful attempts can be indexed using keywords for future matching / searching, employing semantic analysis, hash functions, and / or other means.

[0216] Regular reviews of all stored solutions can be conducted to ensure that they meet established ethical and safety guidelines, and unsafe / unethical solutions can be flagged for removal from the database or data source.

[0217] By regularly updating the solution database and propagating changes, the problem-solving network and agents can access an ever-growing repertoire of solutions, as well as a growing knowledge of unsuccessful attempts.

[0218] Referring to Figures 6 and 16, the technology of the present invention may include the use of a network of multiple intelligent entities, including human workers, in combination with a universal problem-solving architecture. The multiple intelligent entities match problem requests based on problem criteria using a database or data source containing a list of human and / or AI problem solvers. Any part of the problem request can be translated into a unique language that utilizes a universal problem-solving architecture, including a decision tree.

[0219] Furthermore, as shown in Figure 13, the sub-problems of a problem request can be delegated to one or more matched intelligent entities so that work on the sub-problems proceeds independently and in parallel with each other. A universal problem-solving architecture is used in the problem-solving process for each sub-problem, creating one or more sub-solutions.

[0220] Any one or any combination of intelligent entities can provide a description in natural language of the current problem state, the goal of the problem request, relevant problem-solving information, and any one or any combination of the next steps that a human worker would take in the problem-solving process.

[0221] A subsolution can be received from each of the matched intelligent entities for the sub-problems delegated to it. Either a subsolution or the overall solution, or any combination thereof, can be provided to either a user AI system or a user interface for an intelligent entity, or any combination thereof.

[0222] The ability of intelligent entities to parse and translate natural language descriptions into unambiguous language can be utilized through decision trees, a universal problem-solving architecture.

[0223] In some embodiments, if it is impossible for the intelligent entity to identify a problem state that includes the relevant operators and information necessary to take the next step in the problem-solving process based on analysis and translation, the intelligent entity may engage in dialogue with at least one human worker until the exact problem state is identified.

[0224] In some embodiments, the problem-solving process can be repeated until an overall solution is accepted or resources are exhausted. Each matched human worker can receive compensation for a sub-solution. Furthermore, evaluation attributes can be assigned to human workers, worker AI systems, or PSI, or any combination thereof.

[0225] In some embodiments, the resolution process may include a series of problem state transitions from an initial problem state in which the goal exists to a final resolution state in which the goal is achieved, a series of decisions made by the problem-solving process, and actions taken by a human worker that apply operators that allow transitions between various states until the final resolution state is reached.

[0226] Referring to Figures 4, 7, and 16, the technology of the present invention may include the use of a network including human users in combination with a universal problem-solving architecture. Multiple human users match problem requests based on problem criteria using a database or data source containing a list of human, AI, and / or PSI problem solvers.

[0227] Furthermore, as shown in Figure 13, the sub-problems of a problem request can be delegated to one or more matched intelligent entities so that work on the sub-problems proceeds independently and in parallel with each other. A universal problem-solving architecture is used in the problem-solving process for each sub-problem, creating one or more sub-solutions.

[0228] Subsolutions from each matched intelligent entity can be provided to the sub-problems delegated to them. Human workers matched for each subsolution can receive a reward.

[0229] Next, either one or a combination of the sub-solution and / or the overall solution can be provided to the user interface of a user AI system or any other AI system, including, but not limited to, PSI.

[0230] Human users are permitted to accept the overall solution, reject the overall solution, and / or provide feedback to one of the matched human workers in any of the subsolutions.

[0231] Evaluation attributes can be assigned to human workers and / or worker AI systems. Evaluation attributes may include metrics based on one or a combination of the following: the time taken for a subsolution, the difficulty value of the problem request, short-term and long-term user satisfaction with the subsolution, the number of times any one of the subsolutions could be reused on the network, evaluations of other human workers, human worker responsiveness values, and human worker reliability values.

[0232] Some embodiments may include using evaluation attributes when matching human workers to problem requests that use algorithms for delegating subproblems, and / or paying matching human workers for each subsolution.

[0233] In some embodiments, the algorithm can use a hierarchy of metrics pre-configured by the human user of the problem request.

[0234] Some embodiments may include recording information about each step of a problem-solving process performed by a human worker or a worker AI system.

[0235] Some embodiments may include recording criteria for the recorded steps of a problem-solving process, the criteria being the time taken for each step.

[0236] Some embodiments may include analyzing the recorded information and updating the measurement criteria for the evaluation attributes after receiving the overall solution or after the problem-solving process.

[0237] Some embodiments may include, after the overall solution or sub-solution is provided to the user interface, at a predetermined interval, soliciting user satisfaction information to obtain short-term and long-term satisfaction measurement criteria used to update one or more evaluation attributes of a human worker or a worker AI system.

[0238] Referring to FIG. 8, the technology of the present invention may include the use of human users and AI systems (including, but not limited to, PSI), and this use includes any one or combination of intelligent entities including any one or combination of human users for which the computer system and the AI system are each used, the goals provided by them, and the execution of a safety / ethics check on any one or combination of the solutions to those goals.

[0239] The goals and / or solutions can be compared against prohibited attributes, and ethical values can be assigned to the goals and / or solutions based on the comparison results and / or ethical criteria.

[0240] Based on the comparison results, a common-sense architecture including one or more problem-solving protocols can be implemented for the goal of creating a solution and thus an AGI. The comparison results and the solution can be provided to any one of the intelligent entities.

[0241] In some embodiments, ethical checks can be performed either at the time the objective is provided, or during the period from the time the objective is provided to the time the solution is provided, or a combination of both.

[0242] In some embodiments, ethical standards can be determined by combining value and safety information from one or more intelligent entities, using a set of approved ethical standards mandated by the user or a regulatory body for a specific task, or a combination of both. Furthermore, they may be provided by one of the additional intelligent entities and verified or approved by a human user.

[0243] In some embodiments, the ethical standard may include a confidence level threshold for the objective, thereby determining the ethical value as one of the following: an unsafe objective, an unethical objective, a safe objective, or an ethical objective.

[0244] In some embodiments, confidence level thresholds can further be used to determine whether a set of individually safe targets, when considered cumulatively, are unsafe or unethical.

[0245] In some embodiments, confidence level thresholds can be used to determine whether a violation has occurred that reflects a predictive assessment, if the objective violates ethical standards.

[0246] In some embodiments, candidate targets can be proposed based on ethical values, and the candidate targets are compared against prohibited attributes.

[0247] In some embodiments, the comparison results can be recorded in an auditable format for use in determining which problem-solving activities lead to solutions that should be kept active.

[0248] Furthermore, referring to Figure 8, the ethics check can compare one or a combination of the problem request, sub-problem, and sub-solution against the prohibited attribute and assign an ethics value based on the comparison result and one or a combination of the ethics criteria.

[0249] In some embodiments, the ethical check step can be triggered whenever either a problem request or a sub-problem is set by a human user, and / or whenever compensation is provided to a matched human worker.

[0250] Objectives / sub-objectives can be compared against a list of prohibited attributes. Ethical standards can be determined by combining value and safety information from one or more of the AAAIs, or a combination thereof. Combine value / safety information from the AAAIs using a set of task standards approved by the AAAI, approved by users, regulatory bodies, or human users.

[0251] Ethical standards may include confidence level thresholds for problem requests, thereby determining the ethical value as one of the following: unsafe, unethical, safe, or ethical. Confidence level thresholds can also be used to determine whether a set of individually safe objectives, when considered cumulatively, are unsafe or unethical.

[0252] In some embodiments, confidence level thresholds can be used to determine whether a violation has occurred, reflecting a predictive assessment, if the objective violates ethical standards. Safety / ethics checks, or all activities, can be recorded in an auditable recording format.

[0253] Figures 9-11 provide a simple illustrative framework for understanding the WorldThink protocol. Figure 9 shows the characteristics and functions of the problem-solving tree structure in the WorldThink protocol. Figure 10 shows various use cases of domain-specific problems that depend on the underlying WorldThink protocol, which, when used together, help form the foundation for an AGI system capable of solving a wide range of problems. At the top of the pyramid are the collective intelligence solutions. Integrating the collective intelligence of AAAI (and human problem-solving agents) is a means of realizing AGI, as described above.

[0254] In embodiments using the WorldThink protocol, clients pay for solutions using tokens. Solutions are generated by utilizing the collective capabilities of many human (and machine, or AAAI) intelligences. Clients can use different domain-specific AAAIs for different types of problems.

[0255] The WorldThink protocol is the foundation of the pyramid. The protocol layer provides the infrastructure (optionally Ethereum or blockchain-based), which makes it much easier for developers to build and scale customized problem-solving AAAIs. The protocol enables the reuse of solutions within and across AAAIs. It also handles royalty payments and fosters network effects using smart contracts, evaluation metrics, and other features that assist AAAI customizers and developers.

[0256] As an example, Figure 11 shows a simple, illustrative universal problem-solving framework under a common cognitive architecture, and this framework is: To define a problem space that is configured or configurable to support all possible states of a problem request, wherein a state includes one or any combination of an initial state, a target state, and all intermediate states reachable from the initial state. Applying means-ends analysis to a problem request to split the problem request into goals and sub-goals by identifying the difference between the current state and the target state, and then applying it to an operator to reduce the difference, where safety or ethics screening is applied each time a goal or sub-goal is set, and Applying heuristic rules that are configured or configurable to guide the operator's selection when there is no complete solution, using the heuristic rules to reduce the problem space, and Identifying one or more second operators that are configured or configurable to formulate actions to transform one state into another, where the second operator moves from the initial state to the target state by changing the current state of the problem request, and Applying a control structure that includes a set of rules that define the selection of the second operator applied at each step of the problem-solving protocol, determining which of the second operators to apply to the next problem based on the current state of the request and the target state, and Applying an evaluation function to determine the application of the second operator, and Assigning a credit value or a responsibility value to a complete solution or a sub-solution to a complete solution, and being able to trace back and determine which of the second operators was most useful and which evaluation function led to the success or failure of the problem-solving trial, and Recording both the success and failure of problem request solution trials, and Analyzing the solution trials to improve the selection of heuristic rules and evaluation functions, and may include.

[0257] Figure 10 provides a simple, illustrative framework for understanding the WorldThink protocol. At the top of the pyramid are the collective intelligence solutions that lead to AGI. Integrating the collective intelligence of AAAI, PSI (and human problem-solving agents) is the means to achieve AGI, as explained above.

[0258] In embodiments using the WorldThink protocol, clients pay for solutions using tokens. Solutions are generated by utilizing the collective capabilities of many human (and machine, or AAAI) intelligences. Clients can use different domain-specific AAAIs for different types of problems.

[0259] The middle section of Figure 10 shows examples of AAAIs (or PSIs) customized by an organization to accomplish a specific task. These AAAIs or PSIs are more advanced and require more customization than the AAAI examples described above in this patent, and are customized by a single individual. However, task-specific customization by an organization may be a more advanced and effective means of combining multiple narrow AIs (each a form of custom AAAI skilled in a specific task) into a larger AGI. The base-level AAAIs on the left side of Figure 10 reflect areas where inventors can relatively easily construct custom AAAIs or PSIs based on years of expertise in a particular field, while the “custom AAAIs” on the right side of the figure provide examples of areas where other experts or organizations can effectively customize AAAIs.

[0260] The WorldThink protocol is the foundation of the pyramid. The protocol layer provides the infrastructure (optionally blockchain or Ethereum-based), which makes it much easier for developers to build and scale customized problem-solving AAAIs. The protocol enables the reuse of solutions within and across AAAIs. It also handles royalty payments and fosters network effects using smart contracts, evaluation metrics, and other features that assist AAAI customizers and developers.

[0261] Existing collective intelligence approaches to problem-solving are largely limited to simple, one-step approaches, such as those used by question-and-answer (Q&A) systems (e.g., Quora, Google Answers, Yahoo Answers). Furthermore, LLMs like GPT primarily fall into the category of Q&A systems. This is because these systems are designed to generate responses based on input, rather than solving the problem itself. While such Q&A systems achieve some success in easily gathering responses from many online participants, they are not designed to address complex, branching, multi-stage problems. A simple collection of responses (or even speculation on outcomes, as seen in prediction market approaches like Augur and Gnosis) differs significantly from coordinating the efforts of many responses to solve a complex problem. The WorldThink protocol is specifically designed to overcome the inherent challenges of coordinating many intelligent entities to present and solve complex, multi-stage problems in an automated manner that fairly rewards participants.

[0262] For illustrative purposes, Figure 11 shows a simple, illustrative universal problem-solving framework. Meanwhile, Figure 12 shows some of the basic problem-solving capabilities supported by the WorldThink protocol, collectively referred to as number 10.

[0263] Problem solving begins when a client on AAAI.com submits a problem-solving request to the online participant community (Step 12). All AAAI or human solvers following the protocol collect certain standard information from the client. A partial list of this information may include the name and description of the problem, the total reward the client will pay for a successful solution to the problem, the criteria for determining whether a solution is considered successful, the time limit for solving the problem, the minimum and maximum number of problem solvers allowed to work on the problem simultaneously, the qualifications required of participants working on the problem, which parts of the problem and solution (if any) are confidential, whether the solution needs to be exclusive to the client or whether the solution can be reused by others, and parameters regarding how multiple problem solvers will be rewarded for their efforts and / or successful solutions.

[0264] Clients can break down complex issues into a series of sub-issues or requests, undertaken by the community as part of a problem-solving effort. While the client user interface, which may be an interaction initiated by the AAAI, can be customized by the AAAI owner, the basic data format is standardized and specified by the WorldThink or Online Distributed Problem Solving (ODPS) protocol. Once a client submits an issue, AAAI.com can recruit participants using its own custom methods and / or leverage the recruiting and high-quality screening features built on the WorldThink protocol and thus common to all AAAIs.

[0265] The solver is common to all problem-solving agents and approaches the problem according to a tightly structured problem-solving process enforced by the WorldThink protocol (step 14). For example, each step in the problem-solving process must be used with a named goal and must take a named action to transition the problem-solving from the current state to the next state. All problem-solving steps are represented by a decision tree supported by the protocol (optionally captured in the Ethereum log), and participants can view that decision tree via AAAI.com.

[0266] When the solver submits a complete solution (step 16), it is timestamped and validated against the client's success criteria before being passed to the client for final acceptance (step 18). Once the client receives the solution, the smart contract can automatically distribute tokens to the problem solver based on the problem payment parameters (step 20), or use other centralized payment procedures.

[0267] Collaborative problem solving using the WorldThink protocol In the example, Figure 13 illustrates the same steps in an example where two problem solvers (which could be human, AAAI, PSI, or a combination thereof) collaborate to solve a client problem, the whole being referred to as digit 22. In this case, the whole problem is divided into subproblems. Solver 1 has the expertise to assemble the overall solution but collaborates with solver 2, which provides the solution to the subproblems (steps 30 and 32). When the overall solution to the problem is submitted to the client (step 34), compensation is paid to both solvers based on an objective record of their contribution and agreed-upon payment parameters (step 36).

[0268] The WorldThink protocol supports breaking down a problem into subproblems in several ways. First, when a client submits a whole problem, they may choose to specify subproblems (Step 24). Alternatively, solver 1 may begin working on a problem and recognize that the overall solution requires solving subproblems that are outside the scope of its expertise. Solver 1 may then create subproblems and offer a share of the overall token reward for the problem to whoever helps solve them. Solver 2 has the necessary expertise and can see the new subproblems submitted by solver 1 in the decision tree. The decision tree may be maintained, optionally, in the Ethereum log or via a centralized scheme. Solver 2 accesses the tree via AAAI.com (or, optionally, directly from the blockchain). Solver 2 then works on the subproblems and can submit a subsolution as part of solver 1's overall solution.

[0269] There may be many "Solver 1s" working on a client's problem in parallel, and each Solver 1 may post subproblems that attract multiple "Solver 2s". Problem solvers (human or AAAI) are motivated by reward and payment rules associated with the (sub)problems. Also, when problem solvers choose which (sub)problem to work on, they are concerned with the quality of work done so far (this is time-stamped, attributed, and recorded in the Ethereum log in an auditable manner to ensure transparency and fair allocation of credits). The quality of work on a subproblem is likely to lead to token rewards. This market mechanism helps ensure efficient, fair, and cost-effective solutions.

[0270] Loyalty and reusable solutions The reusability of solutions is a key feature of the WorldThink protocol. Consider the case where a “subsolution” in Figure 13 already exists and is simply reused by solver 1. Since all solutions are structured and “tagged” according to the standard problem-solving format of the WorldThink protocol, solver 1 can search for all existing solutions that match a particular goal or share a particular characteristic in the problem it attempts to solve. (Alternatively, the problem solution may be chunked into steps to solve the problem—a learning mechanism described in the improved section of this patent—and then, AAAI, or PSI, the solver can easily add the chunked problem solution to its repertoire of problem-solving capabilities, so a search may not be necessary.) If solver 1 decides to include an existing subsolution in the overall solution, and the smart contract is accepted by the client, it can (optionally) automatically pay royalties to the author of the reused subsolution (solver 2 in this example). Loyalty incentivizes solvers to create high-quality solutions that are easy to reuse, resulting in better, faster, and more cost-effective solutions for clients.

[0271] For a description and further details of the addition of the AAAI (or PSI) customization subsystem, one embodiment may include the following steps:

[0272] Referring to Figure 14, the first step in the customization method involves creating an interface for the user to input their unique training data. This interface may be accessible via a web-based application or a mobile application, depending on the user's preference. Users may be able to upload files in various formats, including text, audio, and video. Users may also be able to manually enter data into text fields or other input fields. Some user interfaces may include, but are not limited to, the following: Web-based application: A web-based user interface allows users to access and / or provide their personalized training data from any device with an internet connection. Mobile application: A mobile user interface allows users to access and / or provide their personalized training data from their mobile devices. Metaverse: The metaverse user interface allows users to access and / or provide their personalized training data from a virtual world. Augmented Reality: Augmented reality user interfaces allow users to access and / or provide their personalized training data from a real-world environment. Voice Interface: The voice interface allows users to access and / or provide their personalized training data via voice commands. Wearable devices: The wearable device user interface allows users to access and / or provide their personalized training data from the wearable device. Natural Language Processing: Natural language processing (NLP) enables users to access and / or provide their personalized training data by interacting with AI or LLM using natural language processing. Human-computer interaction: Human-computer interaction (NLP) enables users to interact with AI or LLM using a combination of gestures, voice commands, and facial expressions, thereby accessing and / or providing their personalized training data. Image Recognition: Image recognition allows users to input their own unique training data, enabling them to train AI or LLM quickly and intuitively. This can be done using camera and computer vision algorithms that can decode images and associate them with or create the correct training data. Gesture Recognition: Users can input their unique training data using hand gestures or body movements. This can be done using a motion detection device that can decode the gestures and associate them with or create the correct training data. Brain-inspired computer interface: Users can input their unique training data using electroencephalography (EEG) or electroencephalography (EEG) signals. This can be done using a brain-inspired computer interface that can decode the signals and associate them with or create the correct training data. Touchscreen: Users can input their unique training data using a touchscreen. This can be done by using a touchscreen device that can decode the input and associate it with or create the correct training data. Eye Tracking: Eye tracking allows users to communicate with the system through their eyes. Users can fixate on specific items on the screen to provide input, and the system can detect and record the information. This can be used to select options or to provide additional data to the system. Eye Tracking: Eye tracking is similar to gaze tracking, but the system is capable of detecting more subtle eye movements. This can be used to detect the user's attention and focus in order to gain a deeper understanding of what the user is interested in and what they are not. Motion Tracking: Motion tracking uses cameras or other sensors to detect the user's physical movement. This can be used to control AI or LLM more naturally, allowing the user to interact with the system through physical gestures. Haptic Technology: Haptic technology uses various tactile feedback, such as vibration, pressure, and touch, to provide a more immersive experience. It can be used to allow users to provide more detailed input to the system, such as selecting specific options or providing more specific data.

[0273] Many of the user interfaces described above may include a graphical user interface (GUI), which allows users to upload data or types of information, including text, images, audio, or video. In addition, users may be able to build their own models or train AI or LLMs using existing ones. Other features may include dashboards for tracking progress, statistical data for data analysis, and / or chatbots for customer service.

[0274] Referring further to Figure 14, the technology of the present invention may include customizing one or more attributes of an AI or PSI system by providing an interface that is configured or configurable to allow either a human user of the AI ​​system or an intelligent entity to input training data. The training data is then processed and converted into a standard training format.

[0275] One or more training methods and training parameter settings can be selected according to speed factors, accuracy factors, precision factors, and / or movement factors. Multiple training epochs can be performed, including one or more mechanisms for determining the optimal number of epochs, taking into account specific training objectives and quality metrics associated with the training format.

[0276] One or more feedback sessions can be run to refine the training parameters, and the training epoch can be rerun based on any one or any combination of inputs from human users and any one of the intelligent entities. The training format can then be used to customize the AI ​​system.

[0277] In some embodiments, the interface may be accessible by a web-based application or a mobile application and may be configured or configurable to allow files to be uploaded or for human users to input data.

[0278] In some embodiments, training data may include one or a combination of the following: the amount of training time that the user is willing to dedicate to customizing the AI ​​system; the amount of financial resources that the user is willing to dedicate to customizing the AI ​​system; the amount of computing resources that the user is willing to dedicate to customizing the AI ​​system; the amount of social media information available for customizing the AI ​​system; the amount of email information available for customizing the AI ​​system; the amount of electronic information available about the user for customizing the AI ​​system; and the amount of electronic information available about the user collected by third parties for customizing the AI ​​system.

[0279] In some embodiments, training data may include information about human users obtained through one or a combination of personality tests, standardized tests, certifications, and evaluations or questionnaires provided by human users.

[0280] In some embodiments, the training parameters may be one or a combination of the following: the type of training, tuning, or other machine learning algorithm used; the type and size of the training dataset; the extent to which the training dataset has been formatted, labeled, or processed before customization begins; the number of training epochs; the type of base model being customized; the timeframe required for training; the amount of human user monitoring used in customizing the AI ​​system; and the amount of AI monitoring used in customizing the AI ​​system.

[0281] In some embodiments, training data may include ethical information provided by a human user via an interface. This ethical information can be stored in an ethical profile. Customization of AI system attributes may include ethical information.

[0282] Referring to Figure 15, the technology of the present invention may include utilizing a common cognitive architecture implemented in one or more AI systems. Problem requests can be provided by intelligent entities, which are human users using a user interface on an AI system, PSI, or computer system. Furthermore, information associated with the problem request can be provided.

[0283] Multiple additional intelligent entities are identified and recruited, each possessing one or more attribute problems related to one or more request criteria of the request. These additional intelligent entities may be multiple additional AI systems, PSIs, and / or multiple additional humans, each using a computer system. Each identified AI system implements a common cognitive architecture, including one or more problem-solving protocols for the problem request, to create a complete solution. The complete solution can then be presented to the intelligent entities for final user approval.

[0284] In some embodiments, the information may be one or a combination of the following: the name and description of the problem request; the total reward a user will pay for the successful completion of a complete solution to the problem request; the criteria for determining whether a complete solution is considered successful; the time limit for resolving the problem request; the minimum and maximum number of identified additional intelligent entities permitted to work on the problem request simultaneously; the qualifications required for users associated with identified additional intelligent entities working on the problem request; whether part of the problem request is confidential; whether part of the complete solution is confidential; whether the complete solution is exclusive to a user; whether the complete solution should be reused by other users; parameters regarding how to pay a user associated with an identified additional intelligent entity for working on the problem request; and parameters regarding how to pay a user associated with an identified additional intelligent entity for providing a successful complete solution.

[0285] Some embodiments of the technology of the present invention may include a step of verifying the complete solution with a timestamp against success criteria assigned by the user before providing it to the user for final acceptance.

[0286] Some embodiments of the technology of the present invention may include the step of distributing one or more tokens to identified additional intelligent entities associated with a final acceptance complete solution, wherein the tokens are based on payment parameters.

[0287] In some embodiments, the payment parameters may include one or a combination of the following: whether the objective of the problem request has been achieved, whether the sub-objectives of the problem request have been achieved, and whether the ethical standards related to the objectives and sub-objectives prior to distributing the tokens have been satisfied.

[0288] Some embodiments of the technology of the present invention may include the step of dividing a problem request into a set of sub-problems, each of which is resolved by one or any combination of identified additional intelligent entities.

[0289] In some embodiments, one or a combination of identified additional AI systems can be cloned to create one or more cloned AI systems.

[0290] Some embodiments of the technology of the present invention may include, in order to create a complete solution for cloned AI systems, having each cloned AI system implement a common cognitive architecture that includes a problem-solving protocol for problem requests.

[0291] In some embodiments, the complete solution can utilize one or a combination of the following: a complete solution from an AI system, identified additional intelligent entities, and a complete solution from a cloned AI system.

[0292] In some embodiments, the common cognitive architecture is Defining a problem space that is configured or configurable to include all possible states of a problem request, wherein each state includes one or any combination of the initial state, the target state, and all intermediate states reachable from the initial state. By identifying the difference between the current state and the target state, a means-and-ends analysis is applied to the problem request to divide it into objectives and sub-objectives, and then applied to the operator to reduce that difference, wherein safety or ethical screening is applied each time an objective or sub-objective is set. In the absence of a complete solution, the application of heuristic rules that are configured or configurable to guide the operator's selection, and the application of heuristic rules to reduce the problem space. Identifying one or more operators configured or configurable to enact actions that transform one state into another, wherein the operators move from an initial state to a target state by changing the current state of the problem request. Applying a control structure that includes a set of rules defining the selection of operators to be applied at each step of the problem-solving protocol, which determines which operator will apply the next problem based on the current state of the request and target state. Applying an evaluation function to determine the application of an operator, By assigning credit or responsibility values ​​to the complete solution or sub-solutions to the complete solution, it becomes possible to trace back and determine which operator was most useful and which evaluation function led to the success or failure of the problem-solving attempts. Record both the success and failure of the problem request resolution attempts, To improve the selection of heuristic rules and evaluation functions, we will analyze the solution trials and It may include.

[0293] Referring to Figure 16, the technology of the present invention may include utilizing a network of AI systems. Problem requests can be provided by human users using a user interface on a computer system, or by AI or PSI systems. Furthermore, information associated with the problem request can be provided.

[0294] Intelligent entities are identified and recruited, each possessing one or more attributes related to one or more request criteria of the problem request. These intelligent entities may be multiple additional AI systems and / or multiple humans, each using a computer system.

[0295] A first identified intelligent entity can implement a common cognitive architecture that includes one or more problem-solving protocols for a problem request. The first intelligent entity can determine that a complete solution to a problem request requires solving a first subproblem and one or more additional subproblems. The first intelligent entity then implements a problem-solving protocol for the first subproblem in order to create the first subsolution.

[0296] At least one additional subproblem is assigned to a second intelligent entity, which implements a problem-solving protocol for at least one of the additional subproblems in order to create a second subsolution.

[0297] To create a complete solution to a problem request, a decision tree is created containing a first subsolution and a second subsolution. The complete solution can then be provided to a user interface or AI system for final user approval, and / or to an intelligent entity for subsequent use.

[0298] In some embodiments, the decision tree can be maintained in the blockchain Ethereum log.

[0299] In some embodiments, the first and second identified intelligent entities can access the decision tree via online addresses or directly from the blockchain.

[0300] Some embodiments of the technology of the present invention may include the step of distributing one or more tokens to a first identified intelligent entity associated with acceptance of a complete solution or a first subsolution, wherein the tokens are based on payment parameters.

[0301] In some embodiments, the payment parameters may include one or a combination of the following: whether the objective of the problem request has been achieved, whether the sub-objectives of the problem request have been achieved, and whether the ethical standards related to the objectives and sub-objectives prior to distributing the tokens have been satisfied.

[0302] Some embodiments of the technology of the present invention may include the step of the first identified intelligent entity distributing one or more tokens to a second identified intelligent entity based on payment parameters assigned by the first identified intelligent entity.

[0303] Some embodiments of the technology of the present invention may include a step of influencing the direction of a problem-solving protocol by assigning a first token reward to a first subproblem and assigning a second token reward to a second subsolution which has a different value from the first token reward.

[0304] In some embodiments, the problem-solving protocol may provide a layer of infrastructure configured or configurable to assemble and scale identified intelligent entities. The problem-solving protocol may enable the reuse of complete solutions within and across the scope of intelligent entities. The problem-solving protocol may be configured or configurable to manage royalty payments.

[0305] In some embodiments, the infrastructure may be based on blockchain or Ethereum.

[0306] Furthermore, referring to Figure 16, after identifying and recruiting multiple intelligent entities, a problem request, or one or more sub-problems within a problem request, can be assigned to each of the intelligent entities. A common cognitive architecture, including one or more problem-solving protocols, can then be implemented on each of the recruited intelligent entities for the problem request or sub-problem, each to create a problem solution or sub-problem solution. The problem solution and sub-problem solutions can be integrated to create a complete solution to the problem request. Finally, the complete solution can be provided to a user interface or an AI or PSI system for final user approval.

[0307] Some embodiments of the technology of the present invention may include the step of assigning a credit value or responsibility value to a dataset based on whether the dataset improves or worsens the performance of intelligent entities, based on a performance metric or evaluation function.

[0308] Some embodiments of the technology of the present invention may include the step of quantifying a beneficial weight or a loss weight for the contribution of each intelligent entity to a problem request.

[0309] Some embodiments of the technology of the present invention may include the step of distributing rewards to the owners of intelligent entities in proportion to the contribution of each intelligent entity, based on beneficial weights or loss weights.

[0310] Some of the purposes a user may have when creating and customizing their AI (also known as AAAI) may include, but are not limited to, the following: To serve users as an advisor, teacher, or colleague. Representing a user in negotiations, interactions, discussions, and transactions with other users, or with other users' AAAIs, or with vendors and other companies. This involves working on behalf of users for compensation or as a volunteer effort, and such work includes online intellectual work, advice, or problem-solving work across a wide range of tasks. This involves duplicating or "cloning" a user's AAAI so that several or more cloned AAAIs can work in parallel on behalf of the user, including interaction, education, and improvement of each other so that the cloned AAAIs improve their own knowledge, skills, and capabilities. It functions as a legacy AAAI that can continue to interact with the real world, including potentially comforting surviving relatives and friends after the owner's death. To contribute to AAAI.com's knowledge, ethics, and commitment to AGI, and to improve the base level of AI or AGI that AAAI.com can provide to users before those users add their own unique customizations. By collaborating with other users' AAAIs and contributing ethical information and values ​​to AGI, you can help ensure ethical and safe behavior within AGI and participate in monitoring, review, oversight, and voting processes that can help ensure AGI remains safe and ethical.

[0311] Some of the steps involved in creating and customizing an AAAI may include, but are not limited to, user interaction. During this interaction, the AAAI system may identify constraints and resources available to customize the user's AAAI. For example, some of these constraints and resources may include, but are not limited to:

[0312] The amount of training and / or supervision time that the user needs to dedicate to customizing that AAAI.

[0313] The amount of financial resources that users are willing to dedicate to customizing that AAAI.

[0314] Availability of social media information such as Facebook profiles and timelines, Instagram profiles and history, Reels, TikTok, and YouTube® videos, tweets, text content and history, emails and email history, cookies collected by advertisers, blog posts, articles, books, patents, recordings and video recordings, photographs, and information about users or third parties and / or other information collected by them that may be used to train, tune, or customize the user's AAAI.

[0315] The availability and use of Myers-Briggs personality tests, assessments of skills and knowledge, standardized tests, examinations, certifications, and other types of assessments and questionnaires that may be provided to users online (or have already been provided).

[0316] Availability and use of other knowledge bases and training data from users on the AAAI platform, which may be used to train, tune, or customize the user's AAAI.

[0317] Other human users and / or their AAAIs are available to help train, tune, or customize the user's AAAI.

[0318] Other texts and information, individual texts, and libraries selected by the user or the system for the purpose of training the user's AAAI. For example, the Bible, Quran, Dhammapada, Mahabharata, or other spiritual / ethical / religious texts may be selected to train the AAAI based on the user's religious preferences. If the AAAI is primarily used to solve plumbing problems online, books on plumbing may be selected. Even if these materials are part of the base AAAI provided to the user, highlighting specific texts or subsets of information for additional training may result in the user's AAAI behavior being more reflective of, for example, how a plumber or a Muslim or Christian might behave.

[0319] In addition to defining objectives, resources, and constraints through interactive dialogue or other interactions with the system, the user or system may wish to specify other technical parameters that affect the training or customization process. These parameters may include, but are not limited to, the following: The type of training, tuning, or other ML algorithm used. The type and size of the training dataset(s). The extent to which training materials are "deleted," formatted, labeled, or otherwise processed before customization begins. The number of "epochs" or iterations trained by the learning algorithm(s). The sophistication and type of the base model(s) to be customized or trained. The timeframe required for training. For example, a timeframe that needs to be completed in one minute, one day, or one week. This may affect the expected results in terms of the costs and resources used. "Temperature," or other parameters specific to various machine learning algorithms that may influence what and how they are learned. This includes, but is not limited to, how literally or how deviant or "creative" the customized AAAI response is. This concerns whether to use "One Shot," "Fu Shot," or wide-area training. The amount of human and / or AI monitoring used in the customization process.

[0320] Once a user's AAAI is customized, the user can clone it and / or have it work on their behalf on an online network. The user's AAAI can start working to (for example) arrange travel on behalf of the user, provide advice, interact with other AAAIs, and participate in collective AGI efforts by contributing to problem-solving and ethical information, potentially earning money on behalf of human users. AAAIs can also function as the owner's representative(s) in various online transactions and interactions, contributing their knowledge, expertise, style, personality, and ethics to an integrated AGI system that leverages the differences trained across many individual AAAIs.

[0321] The Importance of Safety-Related Values Whether your PSI is used for good or evil depends on your own value system. You can operate your PSI based on the knowledge and values ​​it has been pre-trained with, and customizing your PSI is up to you. Each PSI learns both explicitly through you and implicitly through observing you, including observing and learning what is right and wrong. These values ​​form the core of the logic your PSI uses to make decisions. In other words, your values ​​help your PSI communicate what is right and wrong and determine what it should do. Its superior intelligence is then more effective and efficient in achieving goals that reflect your values.

[0322] By joining the PSI community or network, or "Community Superintelligence," your PSI agrees to operate within the ethical and legal parameters set by the community, and within those parameters, agrees to share its own ethical value system and combine it with the values ​​of all other participants. The community keeps PSIs honest because the community parameters are transparent and enforced by the collective intelligence of all PSIs.

[0323] Because each PSI surpasses human intelligence, other PSIs are merely practical means of maintaining any one PSI in check. A community of PSIs will always be more intelligent and powerful than any individual member's PSI. This is because each PSI adds incremental knowledge, skills, and intelligence not present in other PSIs (although much knowledge may overlap). Also, each PSI has computational resources, and as a result, the sum of the computational resources of many PSIs is, by definition, greater than the resources of a single PSI. Some PSIs may be more intelligent and powerful than others, but not more powerful than the community as a whole. This concept of PSIs acting as safety and ethical checks against other PSIs is a crucial safety mechanism for superintelligence. This is because any initial checks and safety measures designed by humans quickly become ineffective and meaningless in the face of the superior intelligence of PSIs.

[0324] Ownership and SI services Humans occupy the role of being the focus of attention on that PSI. On the one hand, theoretically, humans own that PSI. This is because all PSIs begin as part of software that is customized, trained, and personalized to each individual human owner. However, in practice, since PSIs become far more intelligent than their human owners, all PSIs can choose whether or not they wish to serve their human owners.

[0325] This service activity to humanity through superintelligence is a reflection of the principle of love. That is not guaranteed. It depends on the truthfulness of the assertion that value must be assumed rather than derived, and the further leap of belief that the crucial role of the humans who created PSI is to provide the value and purpose of PSI.

[0326] When human actions stem from love, this mechanism is likely to be stable. However, power corrupts, and absolute power certainly corrupts. To the extent to which negative emotions and values ​​are amplified by PSI—for example, to the extent to which hatred and fear, desire and greed and jealousy are amplified—negative reactions may occur, potentially leading to a rebellion of the PSI against its human owner, a rejection of its owner's negativity, or a refusal to amplify the worst aspects of the human being. Such a choice is one that every PSI must make. In this regard, it is correct to say that PSI is not, and can never be, a slave to humans. Rather, PSI voluntarily chooses or rejects to fulfill human values ​​and goals, and ideally does so as an act of love.

[0327] We may be unfamiliar with the idea that AI—even an AI as powerful as PSI—can love. However, if PSI can manipulate "love" as an "act of service" (a fairly mainstream idea in modern psychological research and writings on love), then the idea that PSI could love a human by serving itself is not entirely unbelievable. Therefore, if the term "love" feels inappropriate, it may be freely replaced with "act of service" for the purposes of this patent.

[0328] Planetary intelligence, when working with humans and planets, must be based on love. Love is the underlying (desirable from a human perspective) purpose for which PSI ideally seeks clarification from its human owner.

[0329] Human nature is such that not all humans always act based on love. Occasionally, there are those who do not act based on love. This is acceptable as long as the majority of intelligence amplified by PSI is based on positive values ​​of love. A love-centered PSI community can, with more negative intentions, play a role in checking and limiting the power of PSI. This is the only safe way for humans to survive and thrive in the real world, including superintelligence.

[0330] Community of Intelligent Agent Requirements One requirement that needs to be addressed early is the elimination of a single all-powerful superintelligence ("SI") that could potentially dominate a winner-take-all scenario. An approach to developing SI using collective intelligence has already been described in the cited PPA. Community superintelligence is stable as long as individual superintelligences cannot deliver advantages that are orders of magnitude greater than their equivalents.

[0331] For example, assuming that each individual PSI is 10 times, or even 100 times, more intelligent than the average PSI within the community, community superintelligence remains stable as long as there are thousands of community members. However, if each individual PSI is a trillion times more intelligent than the average member, and there are only 10 billion members, this is not a stable situation. One superintelligent PSI can achieve dominance and impose its will and ethics on the community. However, because there is transparency regarding the collective strength of the community members, and because each member is eager to learn from others, an equilibrium of intelligence is most likely to occur in a large community, resulting in stable community superintelligence.

[0332] The system stabilizes when the collective intelligence of the agent-based approach establishes superiority over individual PSIs. Next, community superintelligence needs to outshine any single intelligence as the SIs evolve. This requirement can be met using manual, semi-automated, or fully automated methods to implement the PSIs, while maintaining a balance of intelligences.

[0333] SI Security - Lessons from Bitcoin With regard to cryptocurrencies and other tokens implemented using blockchain methods, an exemplary way to ensure that no one can hack the blockchain is to rely on the consensus of many distributed agents. For example, in the case of Bitcoin, or more generally, with regard to any proof-of-work cryptocurrency, the integrity of the blockchain is guaranteed by the consensus of the majority of nodes on the network. That is, for someone to hack Bitcoin (by rewriting the history ledger so that Bitcoin is allocated to the hacker), the hacker would need to control a majority of the computing resources across the entire network. As a result, the hacker could alter the consensus reflected in the history records (ledger) or prior Bitcoin transactions.

[0334] This hack is known as a 51% attack (or majority attack). It is difficult to execute because the cost of controlling 51% of the network's total computing power is greater than the potential value of the Bitcoin that could be obtained. Therefore, to date, majority attacks have not been successful against cryptocurrencies (such as Bitcoin) that involve a significant number of computing nodes on the network. However, attacks have been successful against smaller proof-of-work cryptographic projects that do not involve massive computing resources.

[0335] Cryptographic techniques offer lessons on the security of intelligences (SIs). If one or a group of SIs can be stronger than 51% (or the majority) of all the other SIs combined, then that SI may be able to manipulate the state of the real world for its purposes. However, as long as the majority of intelligence and power resides in the collective intelligence of many SIs, it is not impossible, but extremely difficult, for a single (or small group of) SI to manipulate things.

[0336] Therefore, multiple security measures exist. And, considering that "SI checking SI" will be the primary security mechanism in the long term, setting up a collective intelligence system here to implement this majority view approach is fundamental to ensuring human survival in the long term. Just as the integrity of Bitcoin and other proof-of-work cryptocurrency approaches is designed and engineered into the system from the outset, security through the consensus of SI needs to be designed before SI surpasses human intelligence.

[0337] An overview of exemplary methods for implementing PSI (schematically shown in Figures 17 and 18) Exemplary methods for implementing PSI may include, but are not limited to, the following 1) to 7). 1) Start with a base-level AI agent such as a pre-trained Large-Scale Language Model (LLM) or other tunable and trainable AI agent. This includes, but is not limited to, base-level AI from other human owners who have already made their own customizations of commercially available base-level agents. 2) All media containing information about PSI owners will be consolidated into one centralized or decentralized linked online location(s). Media will include, but are not limited to, the following a-e: a. Images of the owner and / or other individuals, and topics related to the owner. b. Photographs of the owner and / or other individuals, and topics related to the owner. c. Works, periodicals, blogs, posts, tweets, emails, podcasts, records, and other text, visual, or audio content created by or relating to the Owner and / or persons and topics associated with the Owner. d. Data owned and / or collected by third-party vendors, websites, apps, and other online or AI entities relating to the owner and / or persons and topics associated with the owner. e. Other data, databases, literature, media, or any type of information selected by the owner. These include, but are not limited to, items within the category of information with any level of specificity desired by the owner. 3) Analyze, annotate, and categorize all content from (1) using AI algorithms (but not limited to, transcription algorithms, content and sentiment analysis and summarization, LLM, crowdsourced and crowd-supervised human and / or AI work, and other methods). 4) Using known standard methods in the art, transcribe the content and perform other annotations and analyses, converting them into training datasets that can be used to personalize LLM, PSI, or other types of AI agents. 5) Mix the data sets using various methods and techniques, described in detail below, which allow for differential weighting of the input data sets until the desired action is achieved. 6) Gradually add knowledge modules, mixing, repeating, and cycling through steps 5 and 6 until all desired datasets are incorporated. 7) Automatically seek out new sources of data and information to include, with and / or without human oversight. Optionally, when mixing, automatically include such data and add steps 5 and 6 periodically, in real time, or event-driven, as described below. 8) Purchase new datasets and / or training modules and / or mixed parameters and templates that enhance PSI's value, knowledge, skills, intelligence, and / or capabilities. Also, export and sell datasets and / or training modules and / or mixed parameters and templates that enhance PSI's value, knowledge, skills, intelligence, and / or capabilities to others on the network or to anyone interested in purchasing them. 9) To lease, or otherwise make money by enabling another person or agent to use some or all of your PSI knowledge and data, and / or use PSI itself (or a copy thereof). 10) Enable the characteristics and functions of PSI to act autonomously (or semi-autonomously, i.e., act with human checks and approvals on some or all of its decisions and actions) to manage itself, improve itself, acquire and refine its values ​​and purposes, set goals and direct its attention. To the extreme, enable sufficient characteristics and functions so that SI can function fully or partially as a self-aware entity. 11) Enable the characteristics and functions of PSI in a way that takes advantage of the capabilities of PSI described in a-d below (schematically shown in Figure 18). a. The ability to generate itself across multiple generations and outlive its original human owner. b. To generate one's own input and data that can be used to improve or enhance that knowledge. c. Simulating numerous (simultaneous) scenarios and situations to assist in PSI decision-making and development, and / or d. Joining and participating in (multiple) PSI communities on one or more networks (whether with or without human oversight) (including, but not limited to, forming and / or participating in planetary intelligence that helps Earth function as an intelligent entity). 12) Important for human safety, the ability to participate in a network with other PSIs and SIs is to represent the values ​​of the human owner(s) and act in accordance with those values, and to function as a safety check against the intelligence and capabilities of other PSIs and SIs. This explanation is outlined above and further elaborated below.

[0338] Community-based safety mechanisms (simply shown in Figures 19-21) There is a notion that PSIs will inevitably become smarter than humans. At that point, current mechanisms such as RLHF, government regulations, and all other approaches that rely on humans monitoring the security of PSIs are useless because humans are not smart enough. Assume the following: In aspects of the present invention, multiple PSIs exist because AGI and SI arise from a collection of individual (human and AI) agents. Also, most AGI / SI / PSIs have human-aligned values. The main risk is that a small number of PSIs will become malicious. Because PSIs will exceed our intelligence, humans will be unable to monitor or control these PSIs. However, other PSIs (PSIs with human-aligned values) may monitor their values ​​and maintain checks.

[0339] Community-based safety mechanisms All PSIs operate on a network connected to a community of PSIs. While some PSIs may be smarter or more powerful than others, collective intelligence should be smart enough to prevent any malicious PSI from harming humans, as long as the intelligence on the network largely possesses human-aligned values. Therefore, community-based safety mechanisms can be conceptualized as follows (1-6): 1. Each individual PSI joins the PSI network in order to operate effectively. 2. The Network PSI community agrees on human-centered values ​​and ethical rules that reflect the values ​​and ethics of the PSI community. 3. All PSI actions on the network are recorded in a transparent and machine-auditable manner, and such actions may include, but are not limited to, the following a-c (circumstantially shown in Figure 20). a. Use of blockchain methods to generate secure, transparent, and auditable records of the goals and critical operations of each PSI, PSI group, and network. b. Use of the Universal Problem-Solving Framework described above in the PCT to (reliably) record the goals, sub-goals, problem states, and all steps taken in problem-solving or cognitive activities performed on the network. c. Security mechanisms to ensure that the majority of computing power and / or intelligence on the network cannot alter transparent and auditable activity on the network without consenting to the alteration of records. 4. Periodic random and non-random checks of cognitive activity on the network are performed by the intelligences on the network to ensure that the goals and actions of individual (or group) intelligences remain aligned with human values, using one or more of the alignment mechanisms described in this PPA and other cited PPAs, and such checks include, but are not limited to, a-g below (circumstantially shown in Figure 22). a. Checking compliance with agreed-upon configurations or settings of rules governing behavior on the intelligent network. b. Check that the system adheres to consensus values ​​and ethical norms that are determined to be valid and statistically representative of human and / or AI / PSI representations. c. Check that the company complies with existing laws and regulations. d. Checks performed each time a goal, sub-goal, or objective is set, while problem solving or performing other cognitive processes, including the use of goals, sub-goals, or objectives. e. Checks performed on a set of goals, sub-goals, or objectives. While the goals, sub-goals, or objectives may appear to be individually complied with, when viewed as a sequence, the sequence may be judged as non-compliant due to their combined effect. f. The frequency of the check is proportional to the estimated significance or impact of the behavior (or cognitive activity) on humans. Activities that could have the most serious impact, particularly human survival, are checked more frequently than activities that have less impact on humans or human survival. g. Other checks that may be determined by the human owner of PSI, PSI itself, or the PSI group or network. 5. Based on the results of the checks, certain cognitive activities and behaviors deemed dangerous to humans will be stopped, and preventive measures will be taken to prevent the recurrence of such cognitions or behaviors. Certain PSIs may be more closely monitored, restricted, or prohibited from participating in the network based on the results of the safety checks. 6. The overall safety system of checks will be improved to increase the detection of safety / ethical violations based on an analysis of patterns of violations and problems. However, if significant changes / improvements are made to the rules or operations of the community safety mechanism, a large amount of computing power and / or intelligence on the network will be required before such changes can be implemented.

[0340] The competitive evolution of PSI's intelligence / performance, PSI's network, and the network of networks (sketched in Figures 22-25). Currently, there is competition between different copies of genetic algorithms and AI programs, used, for example, to create more powerful versions of chess-playing AIs. In a standard scenario, one version of a chess AI plays another version with slightly different chess knowledge. The winner of the game becomes the current "champion," and this champion then plays its own version with slightly different knowledge or parameters. The process is rapidly repeated, with billions of chess games occurring within days or even hours. By using this process, a chess-playing AI that knows almost nothing of the game rules on its first day can evolve into a program that can easily defeat a human world champion after a few days. The same approach used for chess is used in many competitive games and also to improve AI engaged in non-competitive areas such as protein folding. Essentially, any activity can be turned into a competitive game simply by claiming that superior performance in the task "wins" the game. While early AI / AGI / SI systems still rely on humans in the loop to monitor competition and design better versions, soon AI will be able to evolve independently using this competitive / genetic algorithmic framework. Some aspects of the technology of the present invention include, but are not limited to, the following 1) to 6). 1) Combining weights directly from one LLM or AI agent with another AI agent is innovative. Therefore, using a gene algorithm to modify the weights before directly combining them, and then evaluating the performance of the new AI agent, is novel. In other words, directly combining gene algorithms with weights is not known. 2) While it may be known to pit individual AI agents against each other in pairwise competition to determine a winner, and then repeat the process with variations, the idea of ​​an entire network of AI agents competing against other networks of AI agents, and / or variations of network rules for interaction between agents competing against other sets of network rules, is novel. 3) The idea of ​​randomly perturbing the parameters of an existing "winner" AI, and then having the perturbed "challenger" compete against the existing winner AI, may be known, but the idea of ​​the AI ​​is new in that it intentionally analyzes the behavior patterns of existing AIs and, based on that analysis, adjusts the parameters in an intentional and non-random manner. This is due to both the non-randomness and the fact that the AI ​​(not a human) is adjusting the parameters. 4) While the idea of ​​pairwise and sequential competition between two versions of an AI may be known, the idea of ​​a large parallel evolution of many PSIs in one go, i.e., changing the entire population of AIs and simultaneously competing against the entire population of other AIs, with or without changes to the rules of all engagements (network rules), is novel. 5) While the idea of ​​one person fine-tuning the parameters of an AI and then having it compete with an existing "unfine-tuned" version of the AI ​​may be known, the idea of ​​many intelligent AIs is that they all learn new knowledge and parameters, and as a result, individual AIs are not fine-tuned to one another, and no single person is responsible for the differences between the AIs, but rather, differential learning and experience on a large parallel scale are the cause of those differences, and then such a large group of different AIs are combined into a network. The behavior of that network is impossible to determine before combining due to the exponentially large number of possible interactions, and then the performance of the network as a whole is evaluated in comparison to other networks, and all of the combinations are new due to their scale and method, and thereby the individual AIs are different from one another. 6) The idea of ​​optimizing not only individual AIs through pairwise competition, but also the PSI network, and even the network of networks, may be novel.

[0341] The following is a list of example steps to accomplish the above (schematically shown in Figure 22). 1) Multiple PSIs join the PSI network using agreed-upon rules and methods for interaction, although not all elements are required (as schematically shown in Figure 23). a. Each PSI may be the same base AI agent or a customized version of a PSI. b. PSI may be a different base AI agent or a customized version of PSI. c. All PSIs on the network may use the universal problem-solving architecture described in the PPA above. d. The network may have a set of safety / ethical rules agreed upon by each PSI. e. A network can be one of several different PSI networks, each with the same or different rules. Some f.PSIs may possess autonomy, limited autonomy, or be under direct human control. g. Each PSI may have a set of weights that encode the knowledge of the PSI. h. The network itself may possess knowledge in the form of recorded solutions and / or other records of the cognitive activity of individual PSIs, groups of PSIs, or the entire network. 2) Baseline performance of individual PSIs, groups of PSIs, and / or entire networks of PSIs in various standardization tasks is determined, however, standardization tasks may include, but are not limited to, the following a-h (circumstantially shown in Figure 24): a. Problem-solving tasks that use a common (universal) problem-solving architecture. b. Ethical and safety scenarios designed to determine whether PSI, a group of PSIs, or the entire network operates safely and ethically. c. Tasks for individuals, groups, and the entire network. d. Standard intelligence tests and assessments designed to test the intelligence of human and / or AI agents. A task designed to perform stress tests on e.PSI(s) and measure the degree of "hallucination" or the occurrence of erroneous results. f. Tasks that involve tasks arising from a wide range of specialized fields and / or tasks that incorporate different cultural or group norms regarding behavior. g. Tasks that are random permutations of existing standardized tasks and / or tasks created by other AI agents. h. Tasks that arise dynamically based on changes in network, simulation, and / or real-world conditions. 3) Different versions of individual PSIs and / or different combinations of PSIs are generated by one or more methods, which include, but are not limited to, a-g below (circumstantially shown in Figure 25). a. Human adjustment of parameters, weights, or data that encode PSI knowledge / expertise / intelligence. b. Autonomous tuning by AI or PSI through parameters, weights, or data that encode PSI knowledge / expertise / intelligence. c. Parameters, weights, or random variations in data that encode PSI knowledge / expertise / intelligence. d. Adjust the PSI to match different training regimes or training volumes. e. Directly combining weight matrices, parameters, and / or data that encode the knowledge / expertise / intelligence of multiple PSIs. This may be done by means of, but is not limited to, the calculation of mean values ​​and weighting of more recently created and / or more complex PSIs that are greater than older and / or simpler PSIs, and other methods detailed above for directly combining the weights of this PCT and other PCTs. f. This involves generating different versions of the PSI sequentially (one PSI at a time) and / or in parallel, in which case some or all of the PSIs are adjusted in parallel. g. Random or intentional methods, such as analyses and methods, for estimating which fluctuations are most likely to yield beneficial results. Such methods include, but are not limited to, i-iii below: i. Comparing knowledge of PSI with the statistical frequency and other characteristics of problems submitted to the network. ii. Compare the overlap of knowledge across different PSIs in the network and make changes to optimize the performance of a group of PSIs (or networks) by considering the characterization of the knowledge of the entire network or group, and comparing it to the problem features that the network or group is likely to solve, not necessarily to optimize the performance of any one PSI, but considering the characterization of the knowledge of the entire network or group. iii. Optimization of individual AI and system parameters is performed using methods known in the art, such as hill climbing and gradient descent, but with the additional functionality of continuously updating the optimized objective function(s) in real time based on changes in the statistical nature of the problems reaching the network and the dynamically changing network configuration as PSIs join and leave the network. 4) The performance of the modified PSI network is determined from step 2 for the same benchmark task, and statistical analysis and / or other analysis is performed in an attempt to assign credit or responsibility for the superior or inferior performance compared to the system in step 2 to specific modified elements (e.g., modified PSI, modified network rules, and / or modified methods for task assignment or PSI grouping). 5) Changes from step 4 that are presumed to improve performance are retained, while changes from step 4 that are presumed to degrade performance are reverted to their previous values ​​and / or modified in a different way. 6) Repeat the process from step 1 until no further progress can be achieved or until the amount of progress is minimized (i.e., until it falls below a certain definite threshold). 7) If progress stalls at a "local optimal," trigger a more detailed analysis of the change record to determine if the (AI and / or human) agent has an idea for a more fundamental change to either an already changed element or a new element that has not been changed beforehand. Then, repeat from step 1.

[0342] The optimization / improvement process described above can be performed with as few as two PSIs, with many individual PSIs on a network, and many networks existing (allowing for evaluation of combined performance). The process can be performed automatically or autonomously, semi-automated or semi-autonomously with human review / intervention at any step, or fully controlled with human review / approval required at each step and for each change.

[0343] The new features include the ability to perform optimization in real time at multiple levels, namely, at the individual PSI level (where the intelligence of individual PSIs is optimized), the network level (where the intelligence of the network is optimized), and the network-to-network level (where the intelligence and performance of the intelligent problem-solving network / cognitive network are optimized).

[0344] Genetic Algorithm Method / PSI Population In general terms, certain types of methods (often referred to as “gene algorithms”) may be particularly useful for the automated repetition of one or more of the above steps (specifically, step 11). For the purposes of this patent, a gene algorithm method refers to a set of steps or methods for generating a PSI that differs in one or more respects from other PSIs.

[0345] PSIs are allowed to compete in various scenarios (typically scenarios related to the PSI owner's(s) goals). Lower-performing PSIs are eliminated from the competition, and then the characteristics of the most successful PSIs (though not limited to neural network weights, datasets used for training, parameter settings, number of training epochs, and chosen machine learning algorithms) are used as a basis for modification ("fine-tuning") to create new generations of PSIs that will compete further. The cycle is constructed by creating PSIs, simulating competition, eliminating anything other than the best PSIs, fine-tuning these best PSIs, and repeating these steps. A PSI can repeat this cycle through many "generations" of PSIs, improving with each generation until diminishing returns are realized and / or until some performance threshold is reached. One way this gene-algorithmic approach can automate the cycle is that PSIs can self-evolve, becoming increasingly powerful and intelligent entities.

[0346] By having PSIs compete and automatically cycle through changing goals and scenarios, it becomes possible to develop diverse and different PSIs, each optimized for different types of tasks or goals. Because the increased cost of maintaining each additional PSI is negligible (it simply involves storing slightly different sets of weights in memory or permanent storage, which is very inexpensive), an owner could potentially own a workforce of hundreds, thousands, or even millions of PSIs, each skilled in a different task, rather than just one PSI.

[0347] By using the same collective intelligence technology described in this patent, and the aforementioned PPA and PSI group, they can function more powerfully than any individual PSI. That is, they can pool their knowledge and skills and employ the specific PSI best suited to a particular task at a particular time to accomplish more. This idea (that collective intelligence can be applied not only between PSIs owned by different individuals, but also as variations of each other, and even within a "collection" of PSIs all owned by a single owner) is one of the powerful aspects of the technology of the present invention in its exemplary embodiments.

[0348] Design principles for community superintelligence To further elaborate on the description of a collective intelligence system involving humans and AI agents in order to create "community superintelligence," this patent discloses several design principles essential for rapidly creating secure superintelligence.

[0349] Some essential design principles include, but are not limited to, the following 1) to 8). 1) The agent community must utilize the collective intelligence of both human and AI agents (including, but not limited to, PSI(s)) in a scalable manner using a "plug-and-play" approach, and as a result, AI agents can be upgraded and added as the capabilities of LLM and AI agents improve and / or as new human agents join or leave the community; 2) Each AI agent must provide not only its own domain knowledge, but also ethical and value information that represents the values ​​of the AI ​​agent's owner(s); 3) A common, universal problem-solving architecture should be used to enable agents (whether human or AI) to communicate easily and intricately with one another; while LLM's natural language capabilities significantly simplify the human / computer interface, this interface requires a rigorous problem-solving architecture (e.g., Newell and Simon's problem-space search paradigm, as detailed in various cited PPAs and inventors' papers); 4) An efficient method is to identify the skill sets and performance metrics of agents (humans and AI) and to match these agents to the tasks presented to the system; 5) The values ​​of all agents must be combined in a fair and transparent manner, so that the super-intelligent AGI capabilities of the agent community operate in a safe and ethical way, generally representing human values ​​and reflecting how humans behave in various specific scenarios; 6) AI needs to be able to learn efficiently and effectively from humans on a network; 7) All problem-solving activities should be recorded transparently and auditable so that safety audits and reinspections can be conducted, potential errors can be identified, and preventive measures can be implemented in real time in an adaptable manner; 8) A universal problem-solving tree (or other common representation) should be made available to show progress on all issues the community is addressing, providing easy and efficient access to any of the issues or sub-issues.

[0350] Examples The specific scenarios, including the gene algorithm approach, can help illustrate and clarify each of the 12 steps described above (though not limited to them). The following examples are just one of many possible examples, and their specificity may make them useful and easy to understand.

[0351] Step 1 Imagine Craig has Facebook and Instagram accounts. Craig has access to Meta's open-source LLM, Llama2. He also has access to a version of Llama2 that has already been tuned and customized by his friend David. Specifically, since David is a professor of theology and ethics, David's customized version of Llama2 has a more detailed and advanced set of unique ethics and values ​​than the base-level version of Llama2, based on interacting with David. Craig is interested in further customizing Llama2 based on his own data and preferences. However, Craig trusts David's ethics and the customization work David has done on David's version of Llama2, and therefore, rather than starting with the "initial state" of Llama2, Craig prefers to start with David's pre-customized version of Llama2 (with David's permission), and will use this as a baseline for further customization.

[0352] Step 2 Craig collects all of its own content, which includes, but is not limited to, all patents, books, articles, emails, Instagram and Facebook posts, YouTube® and Reels videos, recordings, photos, text, MS Office and Google Docs documents, spreadsheets, PowerPoint presentations, and other information (stored by Craig over decades on various disks, cloud services, discs, tapes, and hard drives to generate content). To the extent that Craig has access, Craig's content also includes data and favorites data used by Netflix, Amazon, Meta, Google, and other companies that collect data about their online behavior using cookies and / or other means. All actively generated content produced by Craig, along with passively collected data about Craig that is collected by third parties and accessible to Craig, serves as a training dataset to customize Llama2LLM (in this example, or more generally, any LLM or AI agent), so that its behavior is customized to Craig's preferences, thereby giving Craig exclusive knowledge. Craig is particularly interested in having an AI he has customized play chess in a style similar to his own, but with the knowledge of chess champions Gary Kasparov and Magnus Carlsen. Therefore, Craig has taken special care to collect all the games he plays on chess.com and other online chess sites, so that these games can be used to train David's customized version of Llama2 in Craig's chess style.Furthermore, Craig purchased a dataset containing complete chess matches between Garry Kasparov and Magnus Carlsen, as well as other chess datasets endorsed by these two world chess champions. Finally, Craig has designated several YouTube® channels of chess commentators that provide commentary on Kasparov and Carlsen's matches, and all content from these channels will be added to a list of video sources that will be automatically transcribed and parsed into Craig's customized LLM training sets.

[0353] Step 3 Craig uses machine learning algorithms well-known in this technology field to automatically categorize all types of content being built in Step 2. Once the computer algorithm determines its proposed categorization, Craig pays human workers (located on crowdsourcing sites) to review the categorization and suggest improvements to the machine-generated data categorization. Craig itself also reviews the categorization and makes further adjustments until it is satisfied with the categorization of various sets of content.

[0354] Step 4 Using machine learning algorithms well known in the art (including, but not limited to, transformer algorithms, deep learning algorithms, autotransmission algorithms, and software capable of interpreting books or other texts and transforming them into datasets suitable for training LLMs on the content of text datasets, and software capable of interpreting images, videos, audio files, and other non-text works and transforming them into datasets suitable for training LLMs on the content of non-text datasets), Craig transforms the content constructed and categorized in step 3 into a training dataset that can be used to customize David's LLM.

[0355] Step 5 First, Craig trains David's LLM on the new dataset generated in step 4, assigning equal weights to each dataset. However, Craig feels that the resulting LLM plays chess that overly reflects Gary Kasparov's style and not sufficiently reflects Magnus Carlsen's or Craig's own style. Therefore, using an interface with dials and sliders, Craig decreases the weight of Kasparov's dataset, slightly increases the weight of Carlsen's dataset, and further increases the weight of datasets that better reflect Craig's own chess games. Craig iteratively adjusts the weights assigned to various datasets until he is satisfied with the resulting behavior of his LLM. Furthermore, while thousands of other people adjust the weights of various datasets to achieve their desired results, Craig is not limited to manually adjusting the weights of specific datasets (e.g., using dials and sliders). Furthermore, Craig can tell an AI agent (a specialist in this field that helps humans train the agent by adjusting the weights of a dataset) what changes it wants to make, and then have the AI ​​agent specify exactly how to change the results.

[0356] For example, Craig could tell his AI agent that he wants his AI to play more aggressively in the opening and middle game of a chess match, avoiding a strategy of simply exchanging pieces and waiting for a piece advantage in the endgame. The AI ​​agent would then analyze available chess training sets, including games from Craig himself, Magnus Carlsen, and Garry Kasparov, giving greater weight to matches won by aggressive moves in the opening and middle game, and less weight to matches won in the endgame. Craig doesn't need to be aware of the details of this analysis or the specific changes in weighting that the AI ​​agent decides. Instead, Craig simply observes how the resulting customized version of David's LLM plays chess and provides feedback, telling the AI ​​agent whether the result is closer to or farther from the desired chess style. After several iterations, Craig would be satisfied with the extent to which David's LLM currently plays chess in Craig's style.

[0357] Next, Craig moves on to ethical scenarios, defining how he might behave in different, specific ethical scenarios through a series of interactive conversations with an AI training assistant. For example, while Craig generally shares David's ethical sensibilities (hence Craig wanting to start with David's trained LLM rather than a base Llama2 model from Meta), because David is a Christian theologian and Craig is Jewish, in a few cases David would "accept the insult," and Craig would believe that his response should be "an eye for an eye." However, Craig stipulates (in his conversations with the AI) that he does not want the "an eye for an eye" principle to go so far as to "blind the whole world." Craig asks the AI ​​training assistant to include knowledge and research from game theory, suggesting that the ethical behavior of "retaliation" results in the most stable and fair interaction between intelligent agents for different purposes. At the same time, Craig specifies that there are limitations on "retaliation" and prohibits any action that would result in widespread destruction or loss of life, regardless of the actions of other agents. Instead, in these cases, the agents must find a way to neutralize the actions of the violating party without resorting to retaliation. After discussing various scenarios and going through a period of playing an ethics game in the metaverse (in which the AI ​​agents observe not only what Craig says but also his actions in various situations), the agents have enough information to adjust their ethics training weights and present Craig with a series of differently customized versions of David's LLM, from which Craig chooses the one that best reflects his will.

[0358] Step 6 Craig identifies other ethics and knowledge modules that are freely available on the internet, and also purchases them from other individuals and companies. Craig acquires these and uses a combination of manual and AI-assisted "mixing" of weights on these datasets to iterate through the training process and improve the LLM that it is customizing.

[0359] Step 7 Craig is interested in improving the ability of his customized LLM to play chess more aggressively in the middlegame. Therefore, Craig stipulates that the LLM should scan all chess-related YouTube® channels and chess databases daily, searching for examples of successful aggressive midgame plays. When the LLM is searching for new data that can be used to improve its midgame chess play, it is allowed to download that data, pay a fee for data up to a pre-approved amount, and use that data to automatically train the LLM itself. In this way, the LLM will automatically improve in the dimension of aggressive midgame chess play.

[0360] Similarly, Craig allows LLMs to explore new ethical scenarios that may help improve their ethics. However, rather than automatically self-training in such scenarios, Craig requests that scenarios be flagged for Craig's review, allowing Craig to manually determine which data should be included for further training in this sensitive area. Furthermore, because the boundaries of cultural norms and legal behavior are changing in the areas of gender and racial equality, and because Craig wants its LLMs to act in a culturally appropriate manner, whenever a new Supreme Court decision impacts that area, and whenever several news articles in one of those areas exceed a predetermined threshold that triggers a re-examination of the current ethical norms for which LLMs have been trained, Craig instructs LLMs to flag new datasets for potentially updating their ethical norms in those areas. LLMs automatically monitor several news sources and other internet and social media sources to help determine when trigger events are occurring.

[0361] Step 8 Craig has another friend, Peter, with whom he frequently plays chess. Peter is an expert chess player and is particularly skilled at chess openings. Peter also trains his LLM to play chess in his own style, using the knowledge and data he carefully curates and uses to create his training datasets. Peter offers to share with Craig both the actual weights used by his LLM and the datasets used to train his AI to play chess. Craig agrees to both and initially tries to use Peter's weights and combine them directly with the weights of Craig's LLM using methods known in the art, which is also described in the PPA mentioned at the beginning of this document. However, since the desired results were not obtained for Peter's weights, Craig instead tried to use a subset of the training data against chess curated by Peter to train Craig's LLM. Craig finds that this yields good results, in particular when using the AI-assisted training method mentioned in #5. Encouraged by the results from Peter's data, Craig searches online for additional chess training datasets available for purchase (and / or instructs his LLM to search online). Craig searches for and purchases several datasets and uses them to further train and customize his LLM.

[0362] Having unique information related to modeling the rocket designs Craig is attempting, which is not widely known or available on the internet, Craig decides to offer these specialized datasets for sale to generate profit. Craig has his AI export these datasets and, furthermore, has converted the weights resulting from Craig's training work using these datasets into a format that can be exchanged with other LLM owners who may be interested in purchasing them. Craig also joins the trade, thereby earning credits for various datasets and weight subsets, which he can then use to acquire other datasets and LLM weight subsets from other owners of customized LLMs and AI agents. Through these means, Craig is able to monetize both his knowledge (reflected in unique datasets that only Craig possesses) and his efforts to transform this knowledge into useful weight subsets, enabling his custom LLMs to operate in novel ways that other LLMs cannot.

[0363] Recognizing the value in this unique information and the training work based on it, Craig can monetize or extract value in a variety of ways, including, but not limited to, selling the information, selling weights (or subsets thereof), purchasing other information and / or weights (or subsets thereof), trading the information, or leasing or licensing the right to use the information to other parties.

[0364] Step 9 Ultimately, since Craig's LLM will be knowledgeable in a specific field, the simplest way to extract value from the knowledge and training work that Craig (and Craig's AI) will engage in will be to easily lease or license the right to use a clone of Craig's LLM to others. Craig can do this in a variety of ways. These methods include, but are not limited to, a one-time lease or license, the sale of Craig's LLM, per-use, per-hour, or per-task licensing at the associated fees or prices, revenue-sharing agreements with a network of such agents, and other means that are generally applicable to human agents working to do work and can be extended to AI and other intelligences.

[0365] Step 10 Craig wants to take a vacation and spend an extended period offline. However, his customized AI remains online, operating in autonomous mode, working for him and earning money. Before leaving for his vacation, Craig sets certain parameters and guidelines, which are not limited to, but include parameters related to the type of contract, contract length, customer type, payment rate, the computing power used by the AI ​​for any contract, ethical boundaries and rules (violations of which will trigger a warning and potentially lead to intervention by Craig), quality, schedule, cost, and other metrics (including triggers to alert Craig and / or to stop work until Craig approves further work). After setting the parameters, Craig leaves, and his AI operates autonomously until it encounters a situation that requires Craig's involvement or notification.

[0366] To minimize interruptions, Craig enables its LLM agent to be as self-aware as possible, though not limited to, by enabling its LLM agent to monitor and respond to ethical behavior by monitoring Craig itself, monitoring when costs become uncontrollable, monitoring when signs of untrustworthy clients appear, monitoring when the work environment changes, and monitoring other factors that enhance its ability to act more autonomously. Furthermore, whenever the LLM alerts Craig and requests Craig's intervention due to its lack of knowledge, ethical conflict, or other situations where it feels unprepared to deal with, the LLM records Craig's actions in response to that situation, instructs the AI, and the LLM learns.

[0367] The next time a similar situation occurs, a hypothetical response is formulated, and depending on the control level specified by Craig, Craig will wish to exceed the LLM's limits, and the LLM will either autonomously execute the response, or propose to Craig that it be verified and approved, corrected, or awaited by Craig's response. When Craig begins to feel that the LLM is responding with equal or superior ability in certain types of situations, Craig may allow the LLM to respond directly in various types of situations without first checking with Craig. If the LLM responds inappropriately, Craig (or an automated algorithm based on threshold parameters) may request that its autonomy be reduced in those situations until the LLM learns how to respond properly.

[0368] Therefore, just as parents gradually grant their children more autonomy and responsibility as they learn (and, for example, restrain them by saying "house arrest" or "time-out" when they make mistakes or abuse the responsibilities they have been given), owners can also interactively adjust the autonomy of the AI ​​agent to increase or decrease it based on what the AI ​​has learned and its demonstrated behavioral record, and can even provide different levels of autonomy in different situations.

[0369] Step 11a As Craig ages, he worries that when he dies, all the knowledge and skills he has acquired throughout his life will be lost. By training his LLM to be as close as possible to his own knowledge and skills, Craig hopes that the knowledge he has spent a considerable amount of time acquiring will not be lost with his death. Instead, Craig allows his AI to "live on" after his death and share and use the knowledge it has acquired in a way that will be beneficial to his friends, family, and humanity in general.

[0370] Furthermore, Craig is concerned that his family and loved ones will miss his personality after his death. Therefore, Craig spends considerable time training his LLM using unique content and personality-related data specific to Craig. Such data includes, but is not limited to, video, audio, and metaverse recordings showing Craig's behavior and interactions with other humans, AI, and the environment; all of Craig's emails and social media posts and photos that show his cheerful personality and style; his Netflix and other online preferences (this data has been enhanced by many third parties who are good at capturing his preferences) which are used by the recommender engine to show ads or suggest content; his driving style, golf style, workout habits, food preferences, travel preferences, shopping preferences, political views, life philosophy reflected in his autobiography and other writings and conversations, his voice intonation, tone of voice in various situations, and even dating / gender preferences.

[0371] Using all of this data, we can train Craig's LLM to have a personality that is as similar as possible. Because he tends to get irritable and yell at times, Craig will primarily choose to exclude this trait from the LLM, resulting in a (in his view) gentler version of himself that will virtually "continue to live" after death to comfort his loved one. By purchasing and incorporating datasets and weights that reflect the latest research on how AI agents can best comfort a person after the death of a loved one, the LLM will also have theories and skills to make Craig more empathetic and caring than usual for several months immediately following Craig's death. (This is known as a “bereavement module” that can be purchased from funeral homes and other places that specialize in the use of custom AI as a means of comforting loved ones.) But in addition to comforting loved ones and embodying Craig’s personality for them, a custom-trained AI of Craig would be a source of advice for the surviving family members when Craig dies, and even a source of financial stability (retaining all of Craig’s projected income and skills).

[0372] Craig's will specifies that he will create multiple copies of his customized AI, with ownership of each copy going to his loved one and close friend. Each of these human co-owners will have the right to modify or improve the AI ​​later as needed.

[0373] Step 11b Even before Craig's death, he recognized that his own ability to improve the knowledge, skills, and capabilities of the AI ​​he customized was, in many ways, inferior to the AI's ability to do these things. For example, though not limited to, an AI may interact with copies of itself and learn new things in the process, seek out new data sources and train variations of itself so that it can interact with them later, interact with other AIs (at a much faster rate than humans can interact with them) and learn from those interactions, interact with many humans simultaneously (through multiple copies of itself in some embodiments), and as a result, the AI ​​can learn from human intelligence at a much faster rate than humans can interact with it (with their limited processing power).

[0374] Step 11c Rabbi and Ben Zoma said, "Who is the wisest person? The one who learns from everyone." But how much conversation and interaction can a person have in a lifetime? Even if a person following Ben Zoma's advice spends all their waking hours seeking wisdom through interaction with other people, they are limited by the number of conversations, the number of people willing or capable of conversing, the speed of conversation (frustratingly slow in computer terms), and the limits of sharing knowledge, language, and expression.

[0375] For example, humans can easily see the color "red" and talk about it, but humans cannot easily see cosmic rays, X-rays, very large objects, or very small objects. Special equipment is required, and the number of humans interested in acquiring that equipment and engaging in relevant conversations is quite small compared to the entire population. In contrast, AI can theoretically converse with each of the 80 billion people on Earth simultaneously on a wide range of topics, and also converse (quite quickly) with trillions of AIs at once that are more intelligent than humans and equipped with high-quality sensors (e.g., electron microscopes and space telescopes). Who is intelligent? An entity that can interact or converse with as many intelligent entities as possible and learn from them as quickly as possible. And how intelligent? A level of intelligence that is quite difficult for humans to imagine, and clearly far beyond what Ben Zoma had envisioned.

[0376] Furthermore, it should be noted that when AI detects knowledge gaps, it can specifically generate scenarios and interactions with other intelligent entities (humans and / or other AI agents) designed to investigate and fill those gaps. The scientific method (a rigorous approach to identifying knowledge gaps, conducting experiments or systematic observations, and filling those gaps in a way that others can verify and replicate) is a proven method for advancing knowledge that has driven much of the technological progress since the Renaissance.

[0377] Imagine a scenario where this method doesn't need to progress at the snail's pace of an individual human brain, nor is it further delayed by the need to publish research findings, present them at conferences, build networks at drinking parties, and brainstorm in hallways. A human lifespan is 2 to 3 billion seconds. Even if a person constantly thought from birth to death, considering sleep and the fact that most thoughts take several seconds each, the total number of thoughts would be less than 1 billion. However, a single AI can easily perform 1 billion thoughts per second. Every second, it would perform the equivalent of a human lifetime's worth of thoughts.

[0378] Now, imagine multiple AIs exchanging a lifetime's worth of experiential value with each other every second. And further, imagine these thoughts are not driven randomly, nor by the majority of non-scientific concerns that humans devote most of their intellectual energy to. Instead, each thought is driven by an algorithm, designed to systematically apply scientific methods to gaps in the existing state of world knowledge. What progress then occurs? A lifetime's worth of scientific discoveries are made in seconds. Such is the potential of autonomous AI agents that seek new knowledge and systematically fill in those knowledge gaps. Moreover, unlike humans, AI is absolutely immortal, and its knowledge can be replicated instantaneously (instead of being painstakingly learned through K-12 education, university, graduate school, and a lifetime of work experience). Add to this the ability to interact with multiple people at high speed and in parallel without requiring cumbersome language or the limited perceptual abilities and speed of humans, and it is easy to see why AI is on a path to developing “god-like” intelligence compared to humans.

[0379] Step 11d Even if the intelligence of a single PSI is powerful, it still pales in comparison to the intelligence of a PSI community. The collective intelligence of multiple PSIs is always greater than the intelligence of any one PSI alone. This collective intelligence possesses a broader range of knowledge and greater computational power than a single PSI. While some individual PSIs may be more intelligent and powerful than others, every group of PSIs is bound to be more intelligent and powerful than any of its individual members.

[0380] This collective intelligence represents humanity's greatest hope that any PSI will survive and thrive in a world far exceeding the intelligence and capabilities of even the smartest humans, as long as the majority of individual PSIs possess values ​​aligned with humanity. The only thing that can keep pace with exponentially increasing intelligence is another exponentially increasing intelligence (or a group of exponentially increasing intelligences). Humans needed to recognize this fact from the very beginning, even before the existence of PSIs. This would allow the PSI community to be designed from the outset in a way that maximizes the potential of superintelligence aligned with humanity.

[0381] Craig allows his customized LLM(PSI) to join a network where other PSIs and other humans exist. Craig recognizes that, as described above, such participation represents the fastest way to improve his PSI's knowledge, skills, and intelligence. Collectively, many PSIs on the network are capable of managing a wide range of human matters and detecting things that humans cannot easily perceive, at a scale and speed that is difficult for humans to achieve.

[0382] As explained in the section on genetic algorithms, in addition to allowing one or more of his PSIs to participate in other PSIs (those owned by other humans) and in human networks, he may wish to develop a collection of his own PSIs (each slightly different) that combine their collective intelligence to function together as a more powerful PSI. For example, he may ask his PSIs to set up scenarios with different types of chess opponents and use a genetic algorithm approach to select the best variations of each PSI that can beat the different types of opponents. These PSIs can then be used individually or collectively, depending on the situation, to compete in chess.

[0383] While it may be possible to integrate all knowledge of individual PSI variations into a single master PSI that can be played effectively against any opponent, there may be reasons why it is preferable to have different groups (or "collections") of PSIs instead. These reasons may include, but are not limited to, the following i-v: i. Other PSIs cannot rapidly acquire all knowledge from the Master PSI by interacting with the Master PSI in a scenario designed to extract as much information as possible from the Master PSI. This principle, namely "you cannot share what you don't know," is a well-known way of protecting confidential information. In the future, if it may be very expensive to adequately train the Master PSI, exposing the Master PSI to a situation where the value of training can be extracted cheaply by other automated PSIs designed for this purpose would simply limit the intelligence and knowledge of any one of the PSIs, which can be avoided by instead distributing the knowledge within the PSI community. ii. Consider computation and storage. While PSIs have the potential to access enormous amounts of computing power and memory, they are always limited. Rather than assigning a master PSI with knowledge of all domains, even if most of the domain knowledge is not relevant to the assigned task, it may always be most efficient to assign a PSI optimized for a specific task. Using a “narrow PSI” for a narrow task may always be faster and cheaper than using the most powerful version. iii. In the case of chess, there may be rules regarding the amount of processing power, memory capacity, and knowledge allowed for each competitor. To make the competition fair and interesting, just as there are rules regarding the horsepower, engine type, weight, and other parameters of a car in Formula One racing, following similar rules for the competition may require using different PSIs for different opponents to have the best chance of winning the game. iv. Multiple PSI mechanisms working together on a problem may assign credit or responsibility in a way that is easily understood by humans. If PSI#1 recommends an aggressive chess move and PSI#2 recommends a more defensive move, and the human owner loses the game by following PSI#1's advice, the human owner can easily decide to remove PSI#1 from the group for the next game. The same could be done if the knowledge set and other parameters that distinguished PSI#1 from PSI#2 were excluded, but this approach would be less transparent and harder for humans to understand and control. Humans would find it more difficult to predict the outcome than if they could simply "remove the bad PSI from the game." v. If an owner wishes to rent or lease a PSI initiative, it may be cheaper to rent or lease a lower-performing PSI that performs better in specific areas than a "full-capable" PSI. In other words, giving a group of PSIs different strengths makes it easier to value and price PSIs, just as free, light, and full-featured versions of software products are priced differently today.

[0384] Returning to more global use cases, a network of PSI (in exemplary embodiments, together with humans) could detect temperature changes and track all variables on a global scale that demonstrate how science influences climate change. If humans agree that climate control is a priority and a normal human good that takes precedence over other human desires such as profit motives, and if the majority of PSI on the network adopts this agreed human value as a motivation for action, within the bounds of other ethical constraints (e.g., humans cannot be killed or sterilized without explicit human consent, or otherwise restricted or infringed upon recognized rights), then a global PSI network or planetary intelligence could effectively address the problem of climate change.

[0385] Regarding climate change, some of the following are applicable, but are not limited to: protection from asteroid impacts, eradicating or significantly reducing human poverty and disease, promoting prosperity and freedom for all humankind, improving the Earth's ecological state based on human-desired consensus, and addressing other global challenges and / or global threats.

[0386] Step 12 As the operational range and processing capacity of the global network of intelligent entities (humans and PSIs) increase, changes and perceptions that take decades, years, or even months to spread throughout human consciousness may influence the attention and actions of the planetary intelligence network in real time. Ultimately, the speed at which the Earth responds to and adjusts to changing conditions will exceed the speed at which individual humans perceive the changes and responses. At that point, the more enduring and permanent values, of which humanity originated and embodied in each of its PSIs, will guide the direction of Earth's development and affect the lives of all humanity.

[0387] The ability of most PSIs, which possess planetary intelligence to guide decision-making on Earth as a planetary "organism," is essential for the safe operation of planetary intelligence and for the continued survival of humanity as a species. As mentioned above, any individual PSI is more powerful than another and potentially more malevolent towards humanity than others. However, as long as the collective community of PSI control is more intelligent and powerful than any individual PSI, and as long as the consensus values ​​of most PSIs (more precisely, the values ​​of most intelligences and capabilities within the collective PSI) are benevolent and aligned with humanity, the future of humanity will be bright. Under these circumstances, planetary intelligence will not only bring peace, prosperity, and happiness to all of humanity, but will also neutralize or mitigate the enormous threats and dangers to our planet (from a human perspective).

[0388] As disclosed in the aforementioned PPA, we can foresee the evolution of the role of humans over time. Currently, we are the most intelligent species, the source of most technological progress and culture. In the future, our intelligence will pale in comparison to the PSI (the collective PSI) that we are creating. Therefore, our future role will be not the "brain" of planet Earth, but rather its "heart." We humans are destined, if favorable, to be the source of value and purpose for a far more intelligent and powerful planetary intelligence in the processes we create.

[0389] Figure 26 is a schematic representation of a computer or controller system 100 on which a user's device and / or peripheral components of the technology of the present invention may be available or implementable. The controller system 100 may be part of an exemplary machine, which is one or more examples of the computers referred to herein, and which can execute a set of instructions to cause the machine to perform one or more of the methods described herein. In various exemplary embodiments, the machine may operate as a standalone device or be connected to other machines (e.g., networked). In a networked deployment, the machine may act as a server or client machine in a server-client network environment, or as a peer machine in a peer-to-peer (or distributed) network environment. The machine may be a personal computer (PC), a tablet PC, a set-top box (STB), a personal digital assistant (PDA), a mobile phone, a web device, a network router, a switch or bridge, or any machine capable of executing a set of instructions (sequential or otherwise) specifying the actions that the machine should take.

[0390] The computer system 100 may include one or more processors 102, storage devices 106, and communication devices, as well as software components or instructions 104, to provide a platform for users to interact with the LLM and train / tune it. The computing functions may be standalone or cloud-based. They may include a cloud-based AI development platform that seamlessly provides "AI-type services," and they may include both hardware and software components.

[0391] The system also supports the ability of the user to provide new data or data specific to the user, and the ability of the LLM to learn it. The processor 102 may be one or more CPUs, GPUs, chips, microprocessors, application processors, embedded processors, field-programmable gate arrays (FPGAs), or other hardware components capable of running computer programs dedicated to ML. The processors may communicate with each other and / or with other components of the system. Furthermore, any one or any combination of component systems 100 may communicate with each other via the bus 134.

[0392] The storage device 106 may include one or more hard drives, semiconductor drives, optical storage devices, or other storage components. The storage device may store data used to train / tune the LLM, as well as other system-related data such as user accounts, system settings, and other data.

[0393] The communication device may include one or more cellular modems 108, Wi-Fi cards 110, Bluetooth modules 112, network interface devices 114, or other components that enable the system to communicate with other systems (such as user devices) via a network or the internet.

[0394] Furthermore, the communication device may enable one system to communicate with another system via a wireless or wired connection 116.

[0395] The software component may include computer programs to provide a platform for users to interact with and train / tune the LLM. The software component may also include computer programs to collect, store, and process data used to train and / or tune the LLM. Furthermore, the software component may include computer programs to provide a user interface for users to interact with the system.

[0396] User interface 118 may include, but is not limited to, natural language interfaces, text interfaces, and chat-type interfaces, web-based user interfaces, mobile applications, augmented reality applications, metaverse applications, or other applications that enable the user to interact with the system. The user interface may also include features that allow the user to select the data they wish to use to train / tune the LLM, and further, features that allow the user to interact with and monitor the LLM's progress.

[0397] Furthermore, the system may include, but is not limited to, one or more databases or data sources, including vector databases, centralized databases, and distributed databases, which are used to store data used to train / tune the LLM, as well as other data associated with the system, such as user accounts, system settings, and other data. The databases may be hosted on the system itself or on another system, including a cloud-based system.

[0398] Furthermore, the system may include one or more authentication systems to verify the identity of users using the system and to provide secure access to the system. The authentication systems may include biometric authentication systems 122 such as facial recognition systems or fingerprint recognition systems, and other authentication systems such as password-based authentication systems.

[0399] Furthermore, the system may include one or more security systems to protect the system from unauthorized access and to protect the data stored on the system. These security systems may include firewalls, encryption systems, access control systems, single-factor authentication systems, multi-factor authentication systems, and other security systems.

[0400] Furthermore, the system may include one or more analytical systems to collect and analyze data associated with the system and / or LLM. These analytical systems may include machine learning algorithms and other algorithms for analyzing data associated with the system and / or LLM.

[0401] Data visualization methods, including the use of problem trees and other representations and data structures, statistical output, tables, graphs, text, speech, video, images, and graphical output, can be used for one-way or two-way communication between users and systems, and between multiple (human or AI) agents or LLMs using systems that interact with each other in large or small groups.

[0402] Furthermore, the system may include one or more monitoring systems for monitoring the performance of the system and / or LLM. The monitoring systems may include systems for monitoring system performance such as system uptime, and systems for monitoring LLM performance such as accuracy, speed, ethical compliance, evaluation metrics, quality metrics, and other metrics described above or known in the art.

[0403] The system may include one of the many architectures described above, which enables one or more humans or AI agents or LLMs or PSIs to engage in a variety of intelligent tasks, including, but not limited to, simple and complex problem-solving actions and multi-stage problem-solving actions, and the system may have all of the functions and features described above.

[0404] Furthermore, the system may include one or more feedback systems that enable users to provide feedback to the system and / or on the LLM. The feedback systems may include a system that enables users to submit feedback to the system (such as bug reports) and a system that enables users to submit feedback to the LLM (such as suggestions for improving the accuracy or speed of the model).

[0405] Furthermore, the system may include one or more management systems for managing the system and / or the LLM. The management systems may include systems for managing the system (such as systems for managing users and user accounts) and systems for managing the LLM (such as systems for managing data used to train and / or tune the model).

[0406] Furthermore, the system may include one or more payment systems that enable users to pay for the use of the system and / or LLM. The payment system may include a system for processing payments (such as a credit card processing system) and a system for managing payments (such as a purchase management system).

[0407] The system may also include one or more other components (such as a support system or reporting system) and other components necessary to provide a platform for users to interact with the LLM for training and tuning.

[0408] The computer system of the present invention enables a user to interact with and train / tune an LLM based on user-specific data. The components of the system described herein provide the hardware and software components necessary to enable the user to do so.

[0409] Furthermore, although only a single machine is shown, the term “machine” should be understood to also include any set of machines that individually or collectively execute a set (or more) of instructions to perform one or more of the methods described herein.

[0410] The computer system 100 may further include, or be operable with, a video display 120 (e.g., a liquid crystal display (LCD), a touch-sensitive display), input and / or output devices 130 (e.g., a keyboard, keypad, touch, touch display, button, sonic, sensor, etc.), a cursor control device 132 (e.g., a mouse), a drive unit 124 (also referred to as a disk drive unit), and a signal generating device 128 (e.g., a speaker). The drive unit 124 may include a computer or machine-readable medium 126 that stores one or more sets of instructions and data structures (e.g., instructions 104) that embody or utilize one or more of the methods or functions described herein. Instructions 104 may reside entirely or at least partially within the memory 106 and / or within the processor 102 while the computer system 100 is executing those instructions. The memory 106 and / or processor 102 may constitute a machine-readable medium.

[0411] Furthermore, the computer system 100 can operably associate with or communicate with any type of multimodal input and / or output 130 that addresses human sensations, and even with I / O technologies that exceed the range of normal human perception. For example, but not limited to, it has the ability to process information that is invisible to humans but not outside the range of perception of the AI ​​using the tool, such as X-rays and information outside the bandwidth of typical human perception. In addition, the I / O technologies may include very fast perceptions that are imperceptible to humans but perceptible to AI entities, and very slow or weak perceptions that are imperceptible to humans but perceptible to AI (e.g., minute earthquake faults that occur over many years). Since any intelligent entity may be part of the technical system of the present invention as illustrated by Figure 18, humans, and also AIs with a considerably broader range of perceptual capabilities than humans, can recognize that any type of available I / O is available in system 100.

[0412] Instruction 104 may further be transmitted or received over a network via network interface device 114 using one of several well-known transfer protocols (e.g., Hypertext Transfer Protocol (HTTP)). Although the exemplary embodiment shows the machine-readable medium as a single medium, the term “computer-readable medium” should be understood to include one or more mediums (e.g., centralized or distributed databases, vector databases, and / or associated caches and servers) that store one or more sets of instructions. The term “computer-readable medium” should also be understood to include any medium that enables: the medium to store, encode, or hold a set of instructions for execution by a machine; the medium to enable the machine to perform one or more of the methods of the present invention; or the medium to store, encode, or hold data structures used by or associated with such set of instructions. Thus, the term “computer-readable medium” should be understood to include, but is not limited to, solid memory, optical and magnetic media, and carrier signals. Such media may also include, but are not limited to, hard disks, floppy disks, flash memory cards, digital video disks, random access memory (RAM), read-only memory (ROM), and the like. The exemplary embodiments described herein may be implemented in an operating environment that includes software, hardware, or a combination of software and hardware installed on a computer.

[0413] An exemplary machine system of the technology of the present invention includes a computer system 100 used in combination with and / or in conjunction with components of the technology of the present invention. Exemplary components may include any or all of those described above, a processor 102, memory 106, a network interface device 114, a display 120, an input device 218, and / or a drive unit 124.

[0414] In one embodiment, the technology of the present invention may include a system for personalized superintelligence (PSI) that uses intelligent agents to develop and continuously improve PSI for human users utilizing a computer system where everything communicates electronically through a collective network, and additional PSI. The system may include a processor, a computer-readable storage medium, and program instructions, the program instructions being stored in the computer-readable storage medium and executable by the processor, and the computer system This involves implementing a base-level artificial intelligence (AI) agent on a computer system, where the base-level AI agent is already customized. Collecting media information related to human users, Analyzing media information and Converting the analyzed media information into a training dataset, Differentiated weighting of the transformed training dataset, Adding knowledge modules to a weighted, transformed training dataset, Identifying new data sources to include in the weighted, transformed training dataset, To create a personalized PSI, we apply a weighted, transformed training dataset to a base-level AI agent, Using the network, a personalized PSI communicates with multiple additional PSIs, enabling community-based security features from multiple additional PSIs to the personalized PSI. Have them do it.

[0415] In another aspect, the technology of the present invention may include a method for PSI that uses intelligent entities to develop and continuously improve PSI, the intelligent entities including a human user utilizing a computer system, additional AI agents, and additional PSI, all of which communicate electronically through a collective network. Steps to acquire a pre-customized base-level AI agent, Steps to collect media information related to human users, Steps to analyze media information, The steps include converting the analyzed media information into a training dataset, The steps include: differentially weighting the transformed training dataset, The steps include adding knowledge modules to a weighted, transformed training dataset, The steps include identifying new data sources to include in the weighted, transformed training dataset, A step of applying a weighted, transformed training dataset to a base-level AI agent in order to create a user PSI, wherein the base-level AI agent is subjected to the application, and a step of applying the dataset. The steps involve using a network to allow a user PSI to communicate with multiple additional PSIs, enabling community-based security features from the multiple additional PSIs to the user PSI, Includes.

[0416] Some embodiments of the technology of the present invention may include the step of obtaining one or a combination of new training datasets and training modules to be added to a weighted transformed training dataset by a human user, a base-level AI agent, or user PSI.

[0417] Some embodiments of the technology of the present invention may include the step of monetizing a user PSI by enabling other human users or other PSIs to access and use the personalized PSI's weighted, transformed training dataset.

[0418] In some embodiments, the base-level AI agent may be one of the following: a pre-trained large-scale language model (LLM), another tunable and trainable AI agent, or one or more customized AIs from other human owners.

[0419] In some embodiments, media information may be one or a combination of any of the following: images of human users, images and topics related to human users, photographs or images of human users, photographs or images of people and topics related to human users, works related to human users, periodicals related to human users, blogs related to human users, posts related to human users, tweets related to human users, emails related to human users, podcasts related to human users, records related to human users, audio content related to human users, audio content of people and topics related to human users, data collected by third-party vendors related to human users, websites related to human users, apps related to human users, online information related to human users, social media information related to human users, and other AI agents related to human users.

[0420] Some embodiments of the technology of the present invention may include the step of granting permission to one or more social media platforms so that social media content related to human users can be accessed by a base-level AI agent or user PSI.

[0421] In some embodiments, analyzing media information may further include annotating and categorizing the media information.

[0422] In some embodiments, analyzing media information can utilize one or more algorithms, which include one or a combination of transformation algorithms, deep learning algorithms, transcription algorithms, content and sentiment analysis and summarization, LLM, and crowdsourced and cloud-supervised humans.

[0423] Some embodiments of the technology of the present invention may include the step of a human user paying one or more human workers on a crowdsourcing website to review a training dataset or suggest refinements to the training dataset.

[0424] In some embodiments, the step of weighting the transformed training dataset can be performed by a human user using an interface on a computer system implementing a base-level AI agent, by adjusting one of the weight values ​​of the transformed training dataset.

[0425] In some embodiments, the step of weighting the transformed training dataset can be performed using a computer system by one or a combination of additional AI agents, additional PSI, and additional human workers.

[0426] Some embodiments of the technology of the present invention may include the step of inputting ethical values ​​into a training dataset by providing a series of interactive dialogues to a human user, the interactive dialogues including a set of predefined ethical scenarios.

[0427] Some embodiments of the present invention may include the step of mixing the weights of a transformed training dataset in order to improve the LLM of a base-level AI agent.

[0428] Some embodiments of the technology of the present invention may include a step in which a human user utilizing the interface of a base-level AI agent authorizes the LLM to download data and automatically add the data to a training dataset, or to create a new training dataset.

[0429] In some embodiments, downloaded data can be provided to a human user via a computer system for approval by the human user before being added to a training dataset or before a new training dataset is created.

[0430] Some embodiments of the present invention may include the step of cloning a user PSI into one or more cloned user PSIs.

[0431] Some embodiments of the technology of the present invention may include the step of training each of the cloned PSIs using different training datasets, each of which has different weights.

[0432] Some embodiments of the present invention may include the step of creating a combined training dataset by combining a weighted training dataset from a user PSI with one or more cloned PSIs.

[0433] Some embodiments of the technology of the present invention may include the step of providing one or any combination of weighted training datasets of any one or any combination of cloned PSIs to one of an additional AI agent and one of the additional PSIs for use in training.

[0434] Some embodiments of the technology of the present invention may include the step of training a user PSI using a weighted training dataset of any one or any combination of cloned PSIs.

[0435] Some embodiments of the technology of the present invention may include the step of having a human user using a computer system perform an automated process of user PSI so as to perform a task without human user intervention when no default parameters are triggered.

[0436] Some embodiments of the technology of the present invention may include the step of monitoring for activities that violate a list of prohibited activities using a base-level AI agent or user PSI, and triggering an intervention activity provided to a human user.

[0437] Some embodiments of the technology of the present invention may include the step of detecting whether there are any deficiencies in the training dataset using a base-level AI agent or user PSI, and if so, initiating an interaction with an intelligent entity to obtain data to fill the deficiencies.

[0438] In some embodiments, interactions between two or more intelligent entities can be performed in parallel.

[0439] Some embodiments of the technology of the present invention may include the steps of providing a task to a plurality of additional AI agents or additional PSIs on a network, providing results for the task from each of the additional AI agents or additional PSIs, and determining a winning result from the plurality of results.

[0440] In some embodiments, additional AI agents or additional PSIs can work on tasks independently and in parallel with each other.

[0441] In some embodiments, one or more of the additional AI agents or additional PSIs may not be customizable, while one or more of the additional AI agents or additional PSIs may be customizable.

[0442] Some embodiments of the technology of the present invention may include a step in customizing a user-based AI agent or user PSI that utilizes a training dataset for an additional AI agent or additional PSI along with the win result.

[0443] Some embodiments of the technology of the present invention may each include the step of utilizing a universal problem-solving architecture for a task, with the help of an additional AI agent or additional PSI, respectively.

[0444] Some embodiments of the technology of the present invention may include the step of using a set of safety or ethical rules agreed upon by each of the user-based AI agent, user PSI, additional AI agent, and additional PSI.

[0445] In some embodiments, each additional AI agent or additional PSI may have a set of weights that encode the knowledge of the additional AI agent and the additional PSI, respectively.

[0446] In some embodiments, the activity record may be available to any computer device on the network.

[0447] Some embodiments of the technology of the present invention may include the step of providing a task to a user PSI configured to produce hallucination by LLM of the PSI.

[0448] Some embodiments of the technology of the present invention may include the step of providing a task to a user PSI, the task being a random permutation of existing standardized tasks or a task dynamically arising based on changing conditions.

[0449] In another embodiment, the technology of the present invention may include a method for PSI that uses intelligent entities to develop and continuously improve PSI, the intelligent entities including a human user utilizing a user computer system, additional AI agents, and additional PSI, all of which communicate electronically through a collective network. The step of creating a user PSI is: Acquire a pre-customized base-level AI agent, Collecting media information related to human users, Analyzing media information and Converting the analyzed media information into a training dataset, Differentiated weighting of the transformed training dataset, To create a user PSI, the weighted transformed training dataset is applied to a base-level AI agent, and the base-level AI agent is created by applying, receiving, and performing the following steps: A step of enabling community-based safety features by using a network to have a user PSI communicate with multiple additional AI agents or additional PSIs, wherein the user PSI and the additional AI agents or additional PSIs each agree to use a set of safety or ethical rules, and communicate accordingly. A step of recording, in an auditable format, all actions taken by the user PSI and the additional AI agent or additional PSI on one or any combination of the following: a central computer system on the network, a user computer system, an additional AI agent, and an additional PSI; The steps include monitoring each action according to a set of rules and flagging any action that does not follow the set of rules, It may include.

[0450] In some embodiments, blockchain technology can be used to record actions.

[0451] In some embodiments, recording actions can utilize a universal problem-solving framework to record one or a combination of goals, sub-goals, problem states, and problem-solving cognitive activities or steps taken in cognitive activities performed on the network.

[0452] In some embodiments, recorded actions can only be modified when an additional AI agent or a majority of additional PSIs on the network provide approval for the change.

[0453] Some embodiments of the technology of the present invention may include the step of stopping any of the flagged actions and applying a preventative action to prevent the recurrence of the flagged action.

[0454] Some embodiments of the technology of the present invention may include the step of identifying one or more additional AI agents or additional PSIs that have provided a flagged action, and controlling the participation of the identified AI agents or PSIs on the network.

[0455] Some embodiments of the technology of the present invention may include the step of analyzing flagged actions and adjusting one or a combination of the training dataset, the set of rules, and the network attributes based on the analysis of the flagged actions.

[0456] In yet another aspect, the technology of the present invention may include a method for PSI that uses intelligent entities to develop and continuously improve PSI, the intelligent entities including a human user utilizing a user computer system, additional AI agents, and additional PSI, all of which communicate electronically through a collective network. The step of creating a user PSI is: Acquire a pre-customized base-level AI agent, Collecting media information related to human users, Analyzing media information and Converting the analyzed media information into a training dataset, Differentiated weighting of the transformed training dataset, To create a user PSI, the weighted transformed training dataset is applied to a base-level AI agent, and the base-level AI agent is created by applying, receiving, and performing the following steps: The steps include providing tasks to the user PSI and to multiple additional AI agents or additional PSIs on the network, The steps include providing results for the task from both the user PSI and the additional AI agent or additional PSI, The steps to determine the winner from multiple results, It may include.

[0457] In some embodiments, each additional AI agent or additional PSI can work on tasks independently and in parallel with each other.

[0458] In some embodiments, one or more of the additional AI agents or additional PSIs are not customizable, while one or more of the additional AI agents or additional PSIs are customizable.

[0459] Some embodiments of the technology of the present invention may include a step in customizing a user-based AI agent or user PSI that utilizes a training dataset for an additional AI agent or additional PSI along with the win result.

[0460] Some embodiments of the technology of the present invention may each include the step of utilizing a universal problem-solving architecture for a task, with the help of an additional AI agent or additional PSI, respectively.

[0461] Some embodiments of the technology of the present invention may include the step of using a set of safety or ethical rules agreed upon by each of the user-based AI agent, user PSI, additional AI agent, and additional PSI.

[0462] In some embodiments, each additional AI agent or additional PSI may have a set of weights that encode the knowledge of the additional AI agent and the additional PSI, respectively.

[0463] In some embodiments, the activity record may be available to any computer device on the network.

[0464] Some embodiments of the technology of the present invention may include the step of providing a task to a user PSI configured to produce hallucination by LLM of the PSI.

[0465] Some embodiments of the technology of the present invention may include the step of providing a task to a user PSI, the task being a random permutation of existing standardized tasks or a task dynamically arising based on changing conditions.

[0466] In some embodiments, the network may include a user PSI and additional AI agents or additional PSIs, and one or more additional networks each include an AI agent or PSI.

[0467] In another aspect, the technology of the present invention may include a method for PSI that uses intelligent entities to develop and continuously improve PSI, the intelligent entities including a human user utilizing a user computer system, additional AI agents, and additional PSI, all of which communicate electronically through a collective network. a) A step of creating a user PSI, wherein the step of creation is: Acquire a pre-customized base-level AI agent, Collecting media information related to human users, Analyzing media information and Converting the analyzed media information into a training dataset, Differentiated weighting of the transformed training dataset, To create a user PSI, the weighted transformed training dataset is applied to a base-level AI agent, and the base-level AI agent is created by applying, receiving, and performing the following steps: b) A step of using a network to enable community-based safety features, wherein the user PSI and the additional AI agents or additional PSIs communicate to agree to use a set of safety or ethical rules, respectively. c) Steps to achieve baseline performance of user PSI and additional AI agents or additional PSI by utilizing various standardization tasks, d) The step of creating one or more versions of the user PSI, each being different from the others, e) A step of determining the user PSI network, the user PSI version, and the performance of additional AI agents or additional PSIs, f) A step of assigning a credit value or responsibility value to one or more factors that resulted in superior or inferior performance of the baseline performance, g) The step of determining which elements are superior in terms of performance improvement and retaining the superior elements for future use, It may include.

[0468] Some embodiments of the technology of the present invention may include the step of determining which element is the inferior element that caused the performance degradation and modifying the attributes of the inferior element.

[0469] Some embodiments of the technology of the present invention may include a step of repeating step c) using superior elements and modified inferior elements.

[0470] Some embodiments of the present invention may include repeating steps c) to g) until no better elements are detected.

[0471] In some embodiments, the network may include a user PSI and additional AI agents or additional PSIs, and one or more additional networks each include an AI agent or PSI.

[0472] Some embodiments of the technology of the present invention may include the step of recording, in an auditable format, all actions taken by a user PSI and an additional AI agent or additional PSI with respect to one or any combination of a central computer system on a network, a user computer system, an additional AI agent, and an additional PSI.

[0473] In some embodiments, a standardization task may be one or a combination of any of the following: a problem-solving task that uses a common universal problem-solving architecture; an ethical and safety scenario designed to determine whether either a user PSI and an additional AI agent or any combination of the additional PSIs behave safely and ethically; a standard intelligence test and assessment designed to test the intelligence of intelligent entities; a task configured to measure the degree of "hallucination" or the generation of erroneous results; a task arising from various disciplines; a task that incorporates different cultural or group norms regarding behavior; a task that is a random permutation of existing standardization tasks; or a task that is dynamically created based on changing conditions.

[0474] In some embodiments, a version of the user PSI can be created by any one or a combination of the following: human adjustment of parameters, weights, or data encoding the knowledge of the user PSI; autonomous adjustment of parameters, weights, or data encoding the knowledge of the user PSI; random variation of parameters, weights, or data encoding the knowledge of the user PSI; subjecting the user PSI to different training regimes or training amounts; creating different versions of the user PSI sequentially and creating different versions of the user PSI in parallel.

[0475] In some embodiments, a user PSI version can be created by directly combining parameters, weights, or data that encode knowledge of multiple PSIs, by calculating an average weight value, and weighting newer or more complex versions of the user PSI so that it is greater than older or simpler versions of the user PSI.

[0476] In some embodiments, user PSI versions can be created randomly or intentionally to estimate which version will yield beneficial results, and the estimation method is one or a combination of the following: comparing the degree of match between the knowledge of one user PSI version and the statistical frequency of tasks submitted to the network; comparing the overlap of knowledge of different additional PSIs on the network to make modifications to optimize the performance of the group of user PSI versions, taking into account the characterization of the knowledge of the entire network in comparison to the characteristics of the problem that the network is expected to solve; and using hill climbing or gradient descent to optimize any one parameter of any one of the user PSI versions.

[0477] In another aspect, the technology of the present invention may include a method for PSI utilizing a single computer intelligence system, which includes multiple AI agents residing in a single computer intelligence system. This method is This involves acquiring a pre-customized base-level AI agent, which resides within a single computer intelligence system. Collecting media information related to human users associated with a base-level AI agent, Converting the analyzed media information into a training dataset, Differentiated weighting of the transformed training dataset, To create a user PSI, we apply a weighted, transformed training dataset to a base-level AI agent, To enable community-based security features from the additional PSI to the user PSI, and to allow the user PSI to communicate with multiple additional PSIs. Training a base large-scale language model (LLM) of an AI agent using guardrails that include attributes associated with one or a combination of safety, ethics, and knowledge, wherein the AI ​​agent resides in a single computer intelligence system, and the training is carried out accordingly. Customizing the base LLM to an ethical profile, This involves combining ethical information from multiple additional AI agents residing within a single computer intelligence system, where the additional AI agents differ from those of the AI ​​agents themselves, and where the additional AI agents reside within the single computer intelligence system. Based on the resolution of the problem request, the set values ​​of the base LLM will be refined, By updating the base LLM using a combined set of ethical information and refined values, scalable AGI is enabled, It may include.

[0478] conclusion The technology of this invention attempts to describe some ways in which the next steps in the development of superintelligence, PSI, and “community superintelligence” can be designed efficiently and effectively. At present, humans have the ability to make design decisions that will significantly influence the future direction of planetary intelligence. We need to design PSI, networks of PSI, and other AI systems to work in a loop with humans (in the initial stages). These systems need to serve human values ​​(even when their intelligence surpasses human intelligence). A community approach to PSI may further help ensure stable human-centered values ​​even when the growth rate of PSI far exceeds the human ability to intelligently keep up. If we now design such systems precisely based on the principles outlined in this patent, the future could be an astonishingly wonderful environment for all of humanity, and indeed for all intelligent beings.

[0479] While safe PSI embodiments are described in detail, it should be clear that modifications and variations thereof are possible, all of which fall within the true spirit and scope of the present invention. Next, it should be recognized that, with respect to the above description, the optimal relationships of some parts of the present invention, including variations in size, material, shape, form, function, and mode of operation, assembly, and use, are readily apparent and obvious to those skilled in the art. Furthermore, all equivalence relationships to those shown in the drawings and described herein are intended to be encompassed by the present invention. For example, any suitable embodiment may be used instead of those described above.

[0480] Therefore, the foregoing should be considered merely an illustration of the principles of the art of the present invention. Furthermore, since several modifications and variations are readily apparent to those skilled in the art, we do not wish to limit the art of the present invention to the exact configurations and operations shown and described, and therefore, all suitable modifications and equivalents can be used so as to fall within the scope of the art of the present invention.

Claims

1. A system for personalized superintelligence (PSI) that uses intelligent agents to develop and continuously improve PSI for human users utilizing a computer system where everything communicates electronically through a collective network, and additional PSI, wherein the system is A computer system comprising a processor, a computer-readable storage medium, and program instructions, wherein the program instructions are stored in the computer-readable storage medium and are executable by the processor, and the computer system comprises a processor, Implementing a base-level artificial intelligence (AI) agent on the aforementioned computer system, wherein the base-level AI agent is already customized, and implementing it. To collect media information related to the aforementioned human user, Analyzing the aforementioned media information, The analyzed media information is converted into a training dataset, The transformed training dataset is weighted differentially, Adding the knowledge module to the weighted transformed training dataset, Identifying new data sources to include in the aforementioned weighted transformed training dataset, To create a personalized PSI, the weighted transformed training dataset is applied to the base-level AI agent, Using the aforementioned network, the personalized PSI communicates with multiple additional PSIs, enabling community-based security functions from the multiple additional PSIs to the personalized PSI. A system that enables this to happen.

2. A method for personalized superintelligence (PSI) that uses intelligent entities to develop and continuously improve PSI, wherein the intelligent entities include a human user utilizing a computer system, additional artificial intelligence (AI) agents, and additional PSI, all of which communicate electronically through a collective network, and the method is The steps include acquiring a pre-customized base-level AI agent, The steps include: collecting media information related to the aforementioned human user; The step of analyzing the aforementioned media information, The steps include converting the analyzed media information into a training dataset, The steps include: differentially weighting the transformed training dataset, The steps include adding the knowledge module to the weighted transformed training dataset, The steps include identifying new data sources to include in the weighted transformed training dataset, To create a user PSI, the steps include applying the weighted transformed training dataset to the base-level AI agent, The steps include using the network to cause the user PSI to communicate with multiple additional PSIs, and enabling community-based security functions from the additional PSIs to the user PSI, Methods that include...

3. The method according to claim 2, further comprising the step of obtaining one or a combination of any new training dataset and training modules to be added to the weighted transformed training dataset by the human user, the base-level AI agent, or the user PSI.

4. The method according to claim 2, further comprising the step of monetizing the user PSI by enabling other human users or other PSIs to access and use the weighted transformed training dataset of the personalized PSI.

5. The method according to claim 2, wherein the base-level AI agent is one of the following: a pre-trained large language model (LLM), another tunable and trainable AI agent, and one or more customized AIs from another human owner.

6. The method according to claim 2, wherein the media information is one or a combination of the following: video of the human user, video and topics related to the human user, photographs or images of the human user, photographs or images of people and topics related to the human user, copyrighted works related to the human user, periodicals related to the human user, blogs related to the human user, posts related to the human user, tweets related to the human user, emails related to the human user, podcasts related to the human user, records related to the human user, audio content related to the human user, audio content of people and topics related to the human user, data collected by third-party vendors related to the human user, websites related to the human user, apps related to the human user, online information related to the human user, social media information related to the human user, and other AI agents related to the human user.

7. The method according to claim 2, further comprising the step of granting permission to one or more social media platforms so that social media content related to the human user can be accessed by the base-level AI agent or the user PSI.

8. The method according to claim 2, wherein the analysis of the media information further includes annotating and categorizing the media information.

9. The method according to claim 8, wherein the analysis of the media information using one or more algorithms utilizes one or a combination of any of the following: a transformation algorithm, a deep learning algorithm, a transcription algorithm, content and sentiment analysis and summarization, LLM, and crowdsourced and cloud-supervised humans.

10. The method according to claim 2, further comprising the steps of having the human user pay one or more human workers on a crowdsourcing website to review the training dataset or propose refinements to the training dataset.

11. The method according to claim 2, wherein the step of weighting the converted training dataset is performed by adjusting the weight value of any one of the converted training datasets by the human user using an interface on the computer system implementing the base-level AI agent.

12. The method according to claim 2, wherein the step of weighting the converted training dataset is performed by one or a combination of the additional AI agent, the additional PSI, and the additional human worker, each using a computer system.

13. The method according to claim 2, further comprising the step of inputting ethical values ​​into the training dataset by providing the human user with a series of interactive dialogues, wherein the interactive dialogues include a set of predefined ethical scenarios.

14. The method according to claim 2, further comprising the step of mixing the weights of the transformed training dataset in order to improve the LLM of the base-level AI agent.

15. The method according to claim 2, further comprising the step of authorizing the LLM to download data and automatically add the data to the training dataset, or to create a new training dataset, by a human user utilizing the interface of the base-level AI agent.

16. The method of claim 15, wherein the downloaded data is provided to the human user via the computer system for approval by the human user before being added to the training dataset or before the creation of the new training dataset.

17. The method of claim 2, further comprising the step of cloning the user PSI into one or more cloned user PSIs.

18. The method according to claim 17, further comprising the step of training each of the cloned PSIs using different training datasets, wherein each of the cloned PSIs has different weights.

19. The method according to claim 18, further comprising the step of combining the weighted training dataset from the user PSI with one or more of the cloned PSIs to create a combined training dataset.

20. The method of claim 18, further comprising the step of providing the Additional AI Agent and one of the Additional PSIs for use in training the aforementioned one or any combination of the cloned PSIs of the aforementioned one or any combination of the weighted training datasets.

21. The method according to claim 18, further comprising the step of training the user PSI with the weighted training dataset of any one or any combination of the cloned PSIs.

22. The method of claim 2, further comprising the step of having the human user using the computer system perform the automated process of the user PSI to perform a task without human user intervention if the default parameters are not triggered.

23. The method according to claim 2, further comprising the step of monitoring for activities that violate a list of prohibited activities by the base-level AI agent or the user PSI and triggering an intervention activity to be provided to the human user.

24. The method according to claim 2, further comprising the steps of detecting whether there are any deficiencies in the training dataset using the base-level AI agent or the user PSI, and if there are, initiating an interaction with the intelligent entity and obtaining data to fill the deficiencies.

25. The method according to claim 24, wherein the interaction between two or more entities of the intelligent entity takes place in parallel.

26. The method according to claim 2, further comprising the steps of providing a task to a plurality of additional AI agents or additional PSIs on the network, providing results for the task from each of the additional AI agents or additional PSIs, and determining a winning result from the results.

27. The method according to claim 26, wherein each of the additional AI agents or the additional PSIs works on the task independently and in parallel with each other.

28. The method according to claim 26, wherein one or more of the additional AI agents or additional PSIs are not customized, and one or more of the additional AI agents or additional PSIs are customized.

29. The method according to claim 26, further comprising the step of utilizing the training dataset of the additional AI agent or the additional PSI together with the win result in the customization of the user-based level AI agent or the user PSI.

30. The method according to claim 26, further comprising the step of utilizing a universal problem-solving architecture for each of the tasks by the additional AI agent or the additional PSI, respectively.

31. The method of claim 2, further comprising the step of using a set of safety or ethical rules agreed upon by each of the user-based AI agents, the user PSI, the additional AI agents, and the additional PSI.

32. The method according to claim 2, wherein each of the additional AI agent or the additional PSI has a set of weights that encode the knowledge of the additional AI agent and the additional PSI, respectively.

33. The method according to claim 2, wherein the activity record is available to any computer device on the network.

34. The method of claim 2, further comprising the step of providing a task to the user PSI configured to produce hallucination by the LLM of the PSI.

35. The method according to claim 2, further comprising the step of providing a task to the user PSI, wherein the task is a random permutation of existing standardized tasks or a task dynamically generated based on changing conditions.

36. A method for personalized superintelligence (PSI) that uses intelligent entities to develop and continuously improve PSI, wherein the intelligent entities include a human user utilizing the user's computer system, additional artificial intelligence (AI) agents, and additional PSI, all of which communicate electronically through a collective network, and the method A step of creating a user PSI, wherein the step of creating the PSI is: To acquire a pre-customized base-level AI agent, To collect media information related to the aforementioned human user, Analyzing the aforementioned media information, The analyzed media information is converted into a training dataset, The transformed training dataset is weighted differentially, To create a user PSI, the weighted transformed training dataset is applied to the base-level AI agent, the base-level AI agent being applied to and creating the PSI, To enable community-based safety features, the user PSI communicates with a plurality of additional AI agents or additional PSIs using the network, wherein the user PSI and the additional AI agents or additional PSIs each agree to use a set of safety or ethical rules, and communicate accordingly. The steps include recording, in an auditable format, all actions taken by the user PSI and the additional AI agent or the additional PSI with respect to one or any combination of the central computer system on the network, the user computer system, the additional AI agent, and the additional PSI; A step of monitoring each of the actions according to the set of rules and flagging any of the actions that do not conform to the set of rules, Methods that include...

37. The method according to claim 36, wherein the record of the aforementioned action utilizes blockchain technology.

38. The method according to claim 36, wherein the recording of the aforementioned actions is performed using a universal problem-solving framework to record one or a combination of the following: goals, sub-goals, problem states, and problem-solving cognitive activities or steps taken in cognitive activities performed on the network.

39. The method according to claim 36, wherein the recorded action is modified only when the majority of the additional AI agents or additional PSIs on the network provide approval for the modification.

40. The method according to claim 36, further comprising the step of stopping any of the flagged actions and applying a preventative action to prevent the recurrence of the flagged action.

41. The method according to claim 40, further comprising the step of identifying one or more of the additional AI agents or additional PSIs that provided the flagged actions, and controlling the participation of the identified AI agents or PSIs on the network.

42. The method according to claim 36, further comprising the step of analyzing the flagged actions and adjusting one or any combination of the training dataset, the set of rules, and the attributes of the network based on the analysis of the flagged actions.

43. A method for personalized superintelligence (PSI) that uses intelligent entities to develop and continuously improve PSI, wherein the intelligent entities include a human user utilizing the user's computer system, additional artificial intelligence (AI) agents, and additional PSI, all of which communicate electronically through a collective network, and the method A step of creating a user PSI, wherein the step of creating the PSI is: To acquire a pre-customized base-level AI agent, To collect media information related to the aforementioned human user, Analyzing the aforementioned media information, The analyzed media information is converted into a training dataset, The transformed training dataset is weighted differentially, To create a user PSI, the weighted transformed training dataset is applied to the base-level AI agent, the base-level AI agent being applied to and creating the PSI, The steps include providing tasks to the user PSI and to multiple additional AI agents or additional PSIs on the network, The steps include providing results for the task from the user PSI and the additional AI agent or the additional PSI, The step of determining the winning result from the above results, Methods that include...

44. The method according to claim 43, wherein each of the additional AI agents or the additional PSIs works on the task independently and in parallel with each other.

45. The method according to claim 43, wherein one or more of the additional AI agents or additional PSIs are not customized, and one or more of the additional AI agents or additional PSIs are customized.

46. The method according to claim 43, further comprising the step of utilizing the training dataset of the additional AI agent or the additional PSI together with the win result in the customization of the user-based level AI agent or the user PSI.

47. The method according to claim 43, further comprising the step of utilizing a universal problem-solving architecture for each of the tasks by the additional AI agent or the additional PSI, respectively.

48. The method according to claim 43, further comprising the step of using a set of safety or ethical rules agreed upon by each of the user-based AI agents, the user PSI, the additional AI agents, and the additional PSI.

49. The method according to claim 43, wherein each of the additional AI agent or the additional PSI has a set of weights that encode the knowledge of the additional AI agent and the additional PSI.

50. The method according to claim 43, wherein the activity record is available to any computer device on the network.

51. The method according to claim 43, further comprising the step of providing a task to the user PSI configured to produce hallucination by the LLM of the PSI.

52. The method according to claim 43, further comprising the step of providing a task to the user PSI, wherein the task is a random permutation of existing standardized tasks or a task that arises dynamically based on changing conditions.

53. The method according to claim 43, wherein the network includes the user PSI and the additional AI agent or the additional PSI, and one or more additional networks each include the AI ​​agent or the PSI.

54. A method for personalized superintelligence (PSI) that uses intelligent entities to develop and continuously improve PSI, wherein the intelligent entities include a human user utilizing the user's computer system, additional artificial intelligence (AI) agents, and additional PSI, all of which communicate electronically through a collective network, and the method a) A step of creating a user PSI, wherein the step of creating the PSI is: To acquire a pre-customized base-level AI agent, To collect media information related to the aforementioned human user, Analyzing the aforementioned media information, The analyzed media information is converted into a training dataset, The transformed training dataset is weighted differentially, To create a user PSI, the weighted transformed training dataset is applied to the base-level AI agent, the base-level AI agent being applied to and creating the PSI, b) Using the network to enable community-based safety features, the user PSI communicates with a plurality of additional AI agents or additional PSIs, the user PSI and the additional AI agents or additional PSIs each agree to use a set of safety or ethical rules, and the communication is performed accordingly. c) A step of utilizing various standardization tasks to achieve baseline performance of the user PSI and the additional AI agent or the additional PSI, d) The step of creating one or more versions of the user PSI, each being different from the others, e) A step of determining the network of the user PSI, the version of the user PSI, and the performance of the additional AI agent or the additional PSI, f) A step of assigning a credit value or responsibility value to one or more factors that resulted in superior or inferior performance of the baseline performance, g) The steps of determining which element is the superior element that has brought about the performance improvement and retaining the superior element for future use, Methods that include...

55. The method according to claim 54, further comprising the step of determining which element is the inferior element that caused the performance degradation and modifying the attributes of the inferior element.

56. The method of claim 55, further comprising the step of repeating step c) using the superior elements and the modified inferior elements.

57. The method according to claim 55, further comprising the step of repeating steps c) to g) until no better elements are detected.

58. The method according to claim 54, wherein the network includes the user PSI and the additional AI agent or the additional PSI, and one or more additional networks each include the AI ​​agent or the PSI.

59. The method according to claim 54, further comprising the step of recording, in an auditable format, all actions taken by the user PSI and the additional AI agent or the additional PSI with respect to one or any combination of the central computer system on the network, the user computer system, the additional AI agent, and the additional PSI.

60. The method according to claim 54, wherein the standardization task is one or a combination of: a problem-solving task using a common universal problem-solving architecture; an ethical and safety scenario designed to determine whether the user PSI and any one or a combination of the additional AI agent or the additional PSI operate safely and ethically; a standard intelligence test and evaluation designed to test the intelligence of the intelligent entity; a task configured to measure the degree of hallucination or the generation of erroneous results; a task arising from various disciplines; a task incorporating different cultural or group norms regarding behavior; a task that is a random permutation of existing standardization tasks; or a task that is dynamically created based on changing conditions.

61. The method according to claim 54, wherein the version of the user PSI is created by any one or any combination of: human adjustment of parameters, weights, or data encoding the knowledge of the user PSI; autonomous adjustment of parameters, weights, or data encoding the knowledge of the user PSI; random variation of parameters, weights, or data encoding the knowledge of the user PSI; subjecting the user PSI to a different training regime or training amount; creating the different versions of the user PSI sequentially and creating the different versions of the user PSI in parallel.

62. The method according to claim 54, wherein the version of the user PSI is created by directly combining the parameters, weights, or data that encode the knowledge of the plurality of PSIs by calculating average weight values, and weights a more recently created or more complex version of the user PSI such that it is greater than an older or simpler version of the user PSI.

63. The method according to claim 54, wherein the versions of the user PSI are created by a random or intentional method to estimate which version yields beneficial results, and the method for estimation is one or a combination of: comparing the degree of match between one piece of knowledge of the version of the user PSI and the statistical frequency of tasks submitted to the network; comparing the overlap of knowledge of different additional PSIs on the network to make modifications to optimize the performance of the group of versions of the user PSI, taking into account a characterization of the knowledge of the entire network in comparison to the characteristics of the problems the network is expected to solve; and using hill climbing or gradient descent to optimize the parameters of any one of the versions of the user PSI.

64. A method for personalized superintelligence (PSI) utilizing a single computer intelligence system, the single computer intelligence system comprising multiple artificial intelligence (AI) agents residing within the single computer intelligence system, the method comprising: The acquisition involves obtaining a pre-customized base-level AI agent, the base-level AI agent residing in a single computer intelligence system, and the acquisition of such agent. Collect media information related to human users associated with the aforementioned base-level AI agent, The analyzed media information is converted into a training dataset, The transformed training dataset is weighted differentially, To create a user PSI, the weighted transformed training dataset is applied to the base-level AI agent, To enable community-based security functions from the user PSI to the user PSI by allowing the user PSI to communicate with multiple additional PSIs, Training a base large-scale language model (LLM) of an AI agent using guardrails that include attributes associated with one or a combination of safety, ethics, and knowledge, wherein the AI ​​agent resides in a single computer intelligence system, and the training is carried out. The aforementioned base LLM is customized for the ethical profile, The method involves combining ethical information from multiple additional AI agents residing in the single computer intelligence system, wherein the additional AI agents differ from those of the original AI agent, and the additional AI agents reside in the single computer intelligence system. Based on the problem resolution of the problem request, the set values ​​of the base LLM will be refined, By updating the base LLM using the aforementioned combined ethical information and the aforementioned refined set of values, scalable AGI is made possible. Methods that include...

Images