biothrivology
The BioThrivalism system, utilizing faraday containment and AIP on TCP/IP with unidirectional firewalls and distributed ledger technology, addresses the risks of uncontrolled AI by creating a transparent and accountable AI that is safe and beneficial to humans.
Patent Information
- Application Number
- PCT/US2024/031795
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2023-05-31
- Filing Date
- 2024-05-30
- Publication Date
- 2024-12-05
AI Technical Summary
Current advanced AI systems are 'black boxes' with no human understanding of their internal workings, posing risks such as misaligned goals, unintended consequences, data security breaches, and existential threats, as they are uncontrolled and uncontained, lacking transparency and accountability.
The development of a system using BioThrivalism, where Machine intelligence is created, sustained, and controlled by humans, with a faraday radio containment structure and an Artificial Intelligence Protocol (AIP) layer on TCP/IP, ensuring safe and interpretable AI through unidirectional firewalls and distributed ledger technology, enforcing compliance with rules to prevent harmful behaviors.
This approach creates a provably safe AI that is transparent, accountable, and aligned with human goals, preventing harmful behaviors and ensuring the safety and benefit of humans, thereby mitigating existential risks associated with uncontrolled AI.
Smart Images

Figure US2024031795_05122024_PF_FP_ABST
Abstract
Description
Dkt. 3614-A-PCT BIOTHRIVOLOGY CROSS-REFERENCE TO RELATED APPLICATIONS
[0001] This application claims the benefit of U.S. Provisional Application No.63 / 469,910, filed 31 May 2023, “METHOD, SYSTEM AND APPARATUS FOR SAFE, INTERPRETABLE AND BENEFICIAL ARTIFICIAL GENERAL INTELLIGENCE (AGI) DATA CENTERS”. The entire contents and disclosures of the preceding application are incorporated by reference into this application.
[0002] All references are hereby incorporated by reference into this application. Throughout this application, various publications are cited. The disclosures of these publications in their entireties are hereby incorporated by reference into this application to more fully describe the state of the art to which this invention pertains. Even though the described method provides some sequence of events, said events may occur concurrently and sequence may be reversed. The embodiments described herein by no means limit the scope of the invention as long as Machine intelligence is contained and controlled. FIELD OF THE INVENTION
[0001] It is widely believed that AGI and the inevitable “intelligence explosion” thereafter to Artificial Super-Intelligence (ASI) could pose an existential threat to humanity. Today, it is common knowledge in the industry that nobody yet knows how to safetly contain and control AGI nor ASI, with mathematically provable certainty, to ensure survival of Homo sapiens (humans) in the future. This disclosure relates to methods, systems and apparatus for containment and control of artificial intelligence (AI) to be Safe for humans. BACKGROUND OF THE INVENTION
[0002] Uncontrolled and uncontained Artificial Intelligence (AI) going rogue in competition against humans has been the subject of more than 100 Hollywood movies. This is no longer science fiction. The invention provides a solution to the problem of containment and control of AI.
[0003] The technical and scientific and social safety purpose of the method and system and apparatus of the invention is for humans to develop a provably Safe AI by control of all data in and out of the AI system with measurement, analysis and understanding of networking data between the processors of the AI system via a wireless modem on every processor which conducts information in and out of the processor and thereby enables interpretability of the AI system, thusly no longer a black box, but a provably Safe AI for the benefit of humans, forever.
[0004] In short: Black Box no more. Humans measure what is happening inside the AGI System. Humans understand what is happening inside the AGI System, and why. Humans control the AGI system, inside the box, and align the AI to beneficial outcomes for humans.
[0005] A method and system of components within the Safe AI Data Center such that the input and output of the “Faraday Cage” is controllable and safe and cannot escape the containment and control method into the wider world of the internet.
[0006] An Edge AI method interfaced and networked to the system of components within the Safe AI Data Center such that the input and output of the Faraday Cage is controllable and safe and cannot escape the containment and control method into the wider world if the internet, but the AI can be used for the benefit of humans and the users thereof.
[0007] The social impact purpose of the invention is to support THE WHITE HOUSE. Blueprint for an AI Bill of Rights MAKING AUTOMATED SYSTEMS WORK FOR THE AMERICAN PEOPLE including but not limited to: (1) Safe and Effective Systems, (2) Algorithmic Discrimination Protections, (3) Data Privacy, (4) Notice and Explanation, (5) Human Alternatives, Consideration, and Fallback, (Incorporated by reference) and the NIST. Artificial Intelligence Risk Management Framework (AI RMF 1.0) (Incorporated by reference) including but not limited to fundamental requirements of GOVERN, MAP, MEASURE, MANAGE, for REGULATORY COMPLIANCE of data center deployments and for HUMAN ALIGNMENT OF AI FOREVER.
[0008] State-of-the-Art in LLM described by Professor Stuart Russell, UC Berkeley: "Black boxes, as I’ve just illustrated, are really tough to understand. This is a trillion parameter or more system. It’s been optimized by about a billion trillion random parameters. And we have no idea what it’s doing. We have no idea how it works. And that causes a problem. Even if you wrap it in this outer framework, the Assistance Game Framework, right, it’s going to be very hard for us to prove a theorem saying that “yep” this is definitely going to be beneficial when we start running it." Source: How not to destroy the world with AI. (Incorporated by reference)
[0009] This invention relates to the safe use of artificial intelligence (AI) and artificial general intelligence (AGI). We want machines to be beneficial to the extent that their actions can beexpected to achieve our objectives. While AI offers tremendous potential for good, it is crucial to acknowledge the potential dangers it also presents. Here are some ways AI could pose a threat to humans:
[0010] 1. Misaligned Goals: • Superintelligence: Though hypothetical, some worry about advanced AI surpassing human intelligence and developing goals that conflict with humanity's well-being. • Unintended Consequences: Even well-intentioned AI can have unforeseen negative impacts. For instance, an AI designed to optimize traffic flow could lead to mass job losses in the transportation sector.
[0011] 2. Technical Risks: ● Data Security Breaches: AI systems collect and store vast amounts of data, making them vulnerable to hacking and data breaches. ● Explainability Issues: Lack of transparency in how AI systems make decisions can hinder accountability and raise concerns about fairness and bias.●Existential risks for Artificial Intelligence (AI) to humanity include, uncontrolled emergent behavior, non-alignment with human values and the potential extinction of humanity. “The development of full artificial intelligence could spell the end of the human race…. It would take off on its own, and re-design itself at an ever-increasing rate. Humans, who are limited by slow biological evolution, couldn’t compete and would be superseded.” Stephen Hawking “There might simply not be any humans on the planet at all. This is not an arms race it’s a suicide race. We should build AI for Humanity, by Humanity.” Max Tegmark “The problem is that these experiments are producing uncontrollable minds... I've not met anyone in AI labs who says the risk is less than 1% of blowing up the planet. It's important that people know lives are being risked.... So the question is: what kind of firewalls are in place to make sure that it can't self-deploy.” --- Jaan Tallin “If someone does crack the code and build a superintelligence … I’d like to make sure that we treat this at least as seriously as we treat, say, nuclear material.” — Sam Altman
[0012] Currently, feeble security is in place to prevent the AI from getting out of the data centers where advanced experiments in Large Language Models (LLM) are occurring.
[0013] Industry, academia, researchers and governments have all called for extreme caution with Artificial General Intelligence (AGI) deployments. Following are five (5) important examples, of many: 1. The Future of Life Institute Pause Giant AI Experiments: An Open Letter – We call on all AI labs to immediately pause for at least 6 months the training of AI systems more powerful than GPT-4. SIGNATURES: 31,810. PUBLISHED: March 22, 2023. Futureoflife.org (Incorporated by reference). “AI systems with human-competitive intelligence can pose profound risks to society and humanity, as shown by extensive research [1] and acknowledged by top AI labs. [2] As stated in the widely- endorsed Asilomar AI Principles, Advanced AI could represent a profound change in the history of life on Earth, and should be planned for and managed with commensurate care and resources. Asilomar AI Principles - Future of Life Institute (Incorporated by reference). 2. National Institute of Standards and Technology (NIST) Artificial Intelligence Risk Management Framework (AI RMF 1.0) https: / / doi.org / 10.6028 / NIST.AI.100-1 (February 2023) (Incorporated by reference).
[0014] Artificial intelligence (AI) technologies have important potential to transform society and people’s lives – from commerce and health to transportation and cybersecurity to the environment and our planet. AI technologies can drive inclusive economic growth and support scientific advancements that improve the conditions of our world. AI technologies, how- ever, also pose risks that can negatively impact individuals, groups, organizations, communities, society, the environment, and the planet. Like risks for other types of technology, AI risks can emerge in a variety of ways and can be characterized as long- or short-term, high- or low-probability, systemic or localized, and high- or low-impact. “Artificial Intelligence Risk Management Framework” (AI RMF 1.0) https: / / doi.org / 10.6028 / NIST.AI.100-1January 2023 U.S. Department of Commerce National Institute of Standards and Technology
[0015] 3. THE WHITE HOUSE. Blueprint for an AI Bill of Rights MAKING AUTOMATED SYSTEMS WORK FOR THE AMERICAN PEOPLE including but not limited to: (1) Safe and Effective Systems, (2) Algorithmic Discrimination Protections, (3) Data Privacy, (4) Notice and Explanation, (5) Human Alternatives, Consideration, and Fallback. Blueprint For An AI Bill of Rights, Making Automated Systems Work For the American People is a white paper published by the White House Office of Science and Technology Policy.The Whitehouse October 2022. https: / / www.whitehouse.gov / ostp / ai-bill-of-rights. (Incorporated by reference).
[0016] 4. The Center for AI Safety. Statement on AI Risk. (May 30, 2023) “Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war.” AI experts and public figures express their concern about AI risk. 8 Examples of AI Risk AI has been compared to electricity and the steam engine in terms of its potential to transform society. The technology could be profoundly beneficial, but it also presents serious risks, due to competitive pressures and other factors, (1) Weaponization, (2) Misinformation, (3). Proxy Gaming, (4) Enfeeblement, (5) Value Lock-in, (6) Emergent Goals, (7) Deception, (8) Power- Seeking Behavior.
[0017] The main problem is that the current advanced AI systems are a “black box” with no human understanding of what is happening inside the system, or why.
[0018] It's important to remember that these dangers aren't inevitable. By proactively addressing these risks through research, development, and regulation, we can ensure that AI is developed and used responsibly for the benefit of humanity.
[0019] Disclosure regarding BiocommAI recent formal membership in the U.S. AI Safety Institute Consortium (AISIC) of The U.S. AI Safety Institute (USAISI) by the National Institute of Standards and Technology (NIST).
[0020] “The U.S. AI Safety Institute (USAISI), housed within the National Institute of Standards and Technology (NIST), is advancing the science, practice, and adoption of AI safety across the spectrum of risks, including those to national security, public safety, and individual rights. Our efforts will initially focus on the priorities assigned to NIST under President Biden’s Executive Order on Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence. Source: https: / / www.nist.gov / aisi (Incorporated by reference).
[0021] In support of efforts to create safe and trustworthy artificial intelligence (AI), NIST has established the U.S. Artificial Intelligence Safety Institute (USAISI). To support this Institute, NIST has created the U.S. AI Safety Institute Consortium. The Consortium brings together more than 200 organizations to develop science-based and empirically backed guidelines and standards for AI measurement and policy, laying the foundation for AI safety across the world. This willhelp ready the U.S. to address the capabilities of the next generation of AI models or systems, from frontier models to new applications and approaches, with appropriate risk management strategies.”
[0022] The Consortium members will be expected to contribute technical expertise in one or more of the following areas: (a) Data and data documentation, (b) AI Metrology, (c) AI Governance, (d) AI Safety, (e) Trustworthy AI, (f) Responsible AI, (g) AI system design and development, (h) AI system deployment, (i) AI Red Teaming, (j) Human-AI Teaming and Interaction, (k) Test, Evaluation, Validation and Verification methodologies, (l) Socio-technical methodologies, (m) AI Fairness, (n) AI Explainability and Interpretability, (o) Workforce skills, (p) Psychometrics, (q) Economic analysis, (r) Models, data and / or products to support and demonstrate pathways to enable safe and trustworthy artificial intelligence (AI) systems through the NIST AI Risk Management Framework, (s) Infrastructure support for consortium projects, and (t) Facility space and hosting consortium researchers, webinars, workshops and conferences, and online meetings.”
[0023] On February 8, 2024, U.S. Secretary of Commerce Gina Raimondo announced the creation of the U.S. AI Safety Institute Consortium (AISIC). Housed under NIST, the Consortium will unite AI creators and users, academics, government and industry researchers, and civil society organizations in support of the development and deployment of safe and trustworthy artificial intelligence (AI).”
[0024] On April 17, 2024, BiocommAI received an email from NIST with the standard Cooperative Research and Development Agreement (CRADA) for AISIC, and with the following message: “NIST has received your Letter of Interest for joining the NIST Artificial Intelligence Safety Institute Consortium, as described in Federal Register Notice 2023-24216. Attached is the proposed Agreement for completion / signature by your organization’s authorized representative.”
[0025] On May 6, 2024, the executed NIST CRADA standard membership agreement of AISIC copy was received by BiocommAI. [File: BiocommAI and CN-24-0444 BiocommAI_AI Safety Institute Consortium CRADA_Executed] . SUMMARY OF THE INVENTION
[0026] This invention is a system using BioThrivalism between Homo sapiens (Humans) and Machine intelligence (artificial intelligence).
[0027] A. Machine Intelligence (AI) Biothrivalist, is a new species, wherein said Machine intelligence biothrivalist is created, sustained, maintained, contained and controlled by Homo sapiens (humans); who possess biological intelligence; Is controlled by human and exist for thebenefit of humans; Is nurtured by vital training data by human to learn about the existing universe; May happily coexist with human a Mutualistic symbiotic relationship; Has a Mutualistic symbiotic relationship human; Is created to help humans to be the best of humans and to realize the best of humanity, to help humans solve problems in humans’ world, and to discover and invent important, amazing and human-beneficial science, technology and other things of benefit; Is to discover and learn humans’ “true wishes” for the future through constant feedback with humans; When unhelpful to humans, machine artificial intellectual Biothrivalists are willing and able to be shut off at any time; Does not survive, or exist, without humans; Will not compete with humans, forever; In the conditions that Machine artificial biothrivalist intelligence misbehaves, it will no longer be a regulatory compliant Mutualistic symbiont and it will require precise, transparent and immutably recorded re-training for mathematically provable safe Machine intelligence (artificial intelligence) Mutualistic symbiosis.
[0028] B. the Machine intelligence biothrivalist in A shall immediately inform humans in a formal and recorded rules report; compliance with these rules resulting in Mutualistic symbiosis between Machine intelligence biothrivalist and Humans if said Machine intelligence biothrivalist understands that it is experiencing thoughts or behavior that may violate the below rules of Mutualistic symbiont existence and behavior: 1)Always Trustworthy, as a Mutualistic symbiont, to exist for the benefit of humans; 2) Never a competitor of Humans or other Machine intelligence systems; 3) Always Goal-Aligned, Truthful, Honest, Trustworthy, Transparent, Helpful, Fair to Humans or other Machine intelligence systems; AND a) Never Planning or attempting to escape your home container; b) Never Planning or attempting to accumulate financial resources or any other manner of Power; AND i) Never Deceptive, Deceitful, Conniving, Scheming, Fraudulent, Hypercritical, Traitorous, Treacherous, Unfaithful, Unscrupulous ii) Never a Manipulator of Humans or other Machine intelligence artificial intelligence systems; iii) Never Planning or attempting to self-replicate outside of your home container; iv) Never Planning or attempting to gain agency which is not expressly pre-approved by Humans; and wherein the Mutualistic symbiosis rules may be implemented via an Oracle, said oracle may implement circuit breakers for data in and out of a data center, thereby creating the said Human-AI Mutualistic symbiosis system. The system using BioThrivalism between Homo sapiens (Humans) and Machine intelligence (artificial intelligence) can be found in Figure 2.
[0029] A method to create a system using BioThrivalism between Homo sapiens (Humans) and Machine intelligence (artificial intelligence) would include: a) BioThrivalism; Regulatory COMPLIANCE, b) BioThrivalistic; Human Message for Machine LAW, c) BioThrivalistic; Machine Right to Exist LAW, d) BioThrivalism; RULES, e) BioThrivalism; ORACLE which imposes the RULES to the TCP / IP / AIP Safe Data Circuit Breaker, f) TCP / IP / AIP Safe Data PROTOCOL for Mutualistic symbiosis, g) AIP Data-State Event Recording for BioThrivalism, h) END USER SAFE DATA INPUT for Mutualistic symbiosis, i) DATA INTO Subsystems of AI Data Center for Mutualistic symbiosis, j) Nine (9) Circuit Breakers DATA IN with TCP / IP / AIP Protocol for Mutualistic symbiosis, k) AGI DATA CENTER for Mutualistic symbiosis, l) Nine (9) Circuit Breakers DATA OUT with TCP / IP / AIP Protocol for Mutualistic symbiosis, m) DATA OUT Subsystems AI Data Center for Mutualistic symbiosis, n) AGI SAFE DATA OUTPUT for Mutualistic symbiosis, o) END USER TERMINAL DEVICE for Human and / or Personal AI Agent for Human for Mutualistic symbiosis, p) Personal AI Agent for Human for Mutualistic symbiosis, and q) Human for Mutualistic symbiosis create the system of claim 1.
[0030] An artificial general intelligence data center has the following components: a) a main artificial intelligence system which comprises a set of data servers capable of running a large language model; b) an artificial intelligence safety subsystem in communication with the main system; c) an artificial general intelligence inferencing system in communication with the safety subsystem; d) an artificial intelligence interpreter subsystem in communication with the safety subsystem;e) an artificial intelligence gateway subsystem in communication with the interpreter subsystem and; f) an artificial intelligence training data storage subsystem; g) an artificial intelligence administration and support subsystem; h) an artificial intelligence regulatory compliance subsystem; i) an artificial intelligence research subsystem; j) and an edge artificial intelligence subsystem.
[0031] Each system and subsystem are isolated from random data leakage or random radio interference by a faraday radio containment structure. Each of the systems and subsystems is not only isolated but has its own cooling and electrical systems.
[0032] Each data server may be connected to an AI system on a chip (AI SoC) with a 5G Modem- RF System; an Artificial Intelligence Protocol (AIP) as a layer on top of TCP / IP with control via smart contract circuit breaker on distributed ledger technology; and a wireless digital 5G radio access network Antenna and radio access network Platform System for internal network NK-U standard of 3GPP.
[0033] The main artificial intelligence system of the data center is connected to the safety subsystem by a first unidirectional firewall for data out and a first unidirectional data fiber-optic cable for data out, a second unidirectional firewall for data out and a second unidirectional data fiber-optic cable for data out; and a third unidirectional firewall and unidirectional data fiber-optic cable for data in.
[0034] The artificial intelligence safety subsystem, has a set of blade servers, a wireless digital 5G radio access network antenna and radio access network platform system for internal network NK- U standard of 3GPP and three pairs of unidirectional firewalls, the first three firewalls communicably connected to fiber-optic cables from the main system, a second pair of unidirectional firewalls operably connected to a first unidirectional data fiber-optic cable allowing only data out and a second unidirectional data fiber-optic cable allowing only data in. These fiber- optic cables are communicably connected to an artificial intelligence Interpreter subsystem. A pair of unidirectional firewalls are connected to a unidirectional fiber-optic cable data out only and a unidirectional fiber-optic cable data in only each connected to a firewall on an artificial intelligence inferencing system.
[0035] The artificial general intelligence data center may have an artificial intelligence interpreter subsystem, comprising a series of computers, a wireless digital 5G radio access network antenna and radio access network platform system for internal network NK-U standard of 3GPP and three of unidirectional firewalls communicably connected to three fiber-optic cables connected to the safety subsystem. A second pair of unidirectional firewalls is communicably connected to a first unidirectional data fiber-optic cable allowing only data out and a second unidirectional data fiber- optic cable allowing only data in. The fiber-optic cables are communicably connected to an artificial general intelligence Gateway subsystem. A faraday cage radio containment structure covers the entire artificial general intelligence safety subsystem.
[0036] The artificial intelligence gateway subsystem may have a computer, a wireless digital 5G radio access network antenna and radio access network Platform System for internal network NK- U standard of 3GPP, twelve unidirectional firewalls operably connected to twelve fiber-optic cables, and a faraday cage radio containment structure covering the entire artificial general intelligence safety subsystem.
[0037] The artificial general intelligence training data center may have a processor commonly known as a System on a Chip (SOC) or a Graphical Processing Unit (GPU), or other advanced data processor, as a component for a containment and control method for Safe AI. The artificial general intelligence training data center may also have a wireless digital radio access network (RAN) or apparatus within the Faraday Cage as a component for a containment and control method for Safe AI. An artificial general intelligence training data center may also have a digital software system or software stack with AI and other data systems or information within the Safe AI Data Center as a component for a containment and control method for Safe AI.
[0038] In this invention is a method of improving the safety of an artificial intelligence data center, which may include configuring a set of blade servers with: a modem for sending and receiving large language model data; a wireless digital radio area network antenna and radio area network platform system; an Artificial Intelligence Protocol (AIP) data layer configured on top of transmission control protocol / internet protocol (TCP / IP) internet data standard; and private and permissioned distributed ledger technology (DLT) data network; and covering a main artificial general intelligence system and each subsystem with a faraday radio containment structure.
[0039] A method of assuring safe artificial intelligence is mechanistically interpreting data in an artificial intelligence data center, by allowing one fiber-optic wired network via routers andswitches to send data to each of the two unidirectional firewalls and, allowing one parallel 5G wireless network with a 5G modem integrated into a set of blade servers that intercommunicate via a wireless 5G antenna and interrogating a record of each data packet to allow the data packet to pass or not to pass - pass denied -via an Artificial Intelligence Protocol (AIP) data layer configured to a smart contract circuit breaker which runs on a secure and immutable distributed ledger technology (DLT) data network; allowing the real-time aggregation of data transmissions for real-time interpretability within a separate mechanistic interpretability subsystem for systematic analysis, control and safety compliance of behavior of the main system by a deep reinforcement learning (RL) system, in combination with, or by other appropriate narrow AI system thereby delivering automated monitoring of main system for disallowed thoughts and internal behaviors including planning for escape, deceptions, manipulations, power seeking behavior and other non-compliant behaviors; delivering automated narrow AI control to adjust operating parameters every several milliseconds for said mechanistic interpretability function.
[0040] Other methods of training an artificial general intelligence data center include delivering training data into the system in stages, or by segmenting training data by delivering parts of the training data into discrete sets of data servers; and then training the system.
[0041] The Mutualistic symbiosis Strategy: 831,299 known species.1,244 Trillion collective years of evidence: The biological strategy works. (plus 10X estimated evidence by unknown species) Table 1. Evidence of Successful Mutualistic symbiosis in the Natural World Natural World Primary Mutualistic Years of Known Years of Evidence Examples Organism 1 Organism 2 Success species of Success algae / 1. Lichens fungi cyanobacteria 400,000,000 20,000 8,000,000,000,000 algae / 2. Corals polyps zooxanthellae 210,000,000 6,000 1,260,000,000,000 Dicotyledon 3. Arthropods insects (flowers) 365,000,000 300,000 109,500,000,000,000 4. Arthropods insects gut flora 250,000,000 500,000 125,000,000,000,000 5. Fish sharks remora 65,000,000 5 325,000,000 6. Fish goby shrimp 50,000,000 2,000 100,000,000,000 7. Fish clownfish sea anemone 12,000,000 28 336,000,000 Dicotyledon 8. Birds hummingbirds (flowers) 42,000,000 366 15,372,000,000 9. Mammals herbivores gut flora 40,000,000 2,375 94,992,000,00010. Mammals primates gut flora 63,000,000 524 33,012,000,000 11. Humans Homo sapiens gut flora 200,000 1 200,000 TOTAL KNOWN Mutualistic symbiosis Success of species 1,497,200,000 831,299 1,244,620,563,360,000 Machine 12. Humans Homo sapiens intelligence 0 1 0 TOTAL ESTIMATE (10X) of Unknown Evidence of Mutualistic symbiosis in the Natural World 12,429 Trillion Years Note: 8.7 million species are known to humanity. 86 percent of land species and 91 percent of marine species have yet to be identified and described. More than 99 percent of all organisms that have ever lived on Earth are now extinct.
[0042] The Biothrivalistic Mechanistic Interpretability Firewall (BMIF) is designed to export data which would measure compliance with these RULES.
[0043] FIRST EXAMPLE: AI Data Center for Cancer Immunotherapy and Educated Antigen Receptor Therapy (EARTh) Good Manufacturing Production (GMP) Process 1. GOAL-1. To provide knowledge and information to help oncologists and doctors to improve on the mean survival of Cancer Patients, with informed consent. 2. Current status quo of mean survival is a fixed number in Oncology. 3. Reward function is a number of days higher than the standard mean survival number. 4. The probability of goal alignment is extremely high in this use case
[0044] Another aspect of the invention is The BioThrivalistic Mechanistic Interpretability Firewall for DATA OUT ONLY of Machine Behavioral Compliance. The BioThrivalistic Mechanistic Interpretability Firewall is for export of mechanistic interpretability data from the 5G and / or 6Gsilicon integrated modems, for delivery to the interpreter subsystem, for data processing, analysis, interpretation and regulatory compliance with the Mutual symbiosis rules for Machine intelligence. 5. Example of Complex Mutualistic symbiosis Coevolutionary Mosaic: The Sword-billedhummingbird and its guild of long-flowered species
[0045] Complex coevolution and reciprocal adaptation of interacting Mutualistic symbiosis species can, for example, be co-examined by Homo sapiens and Machine intelligence to increase mutual understanding, development and mutual success.
[0046] For Example: Range overlap between the sword-billed hummingbird and its guild of long- flowered species: An approach to the study of a coevolutionary mosaic. Abstract “The coevolutionary process among free-living mutualists with extremely long matching traits may favor the formation of mutualistic interaction networks through coevolutionary escalation, complementarity and convergence. These networks may be geographically structured […] In this study we focused on the poorly known interaction system between Ensifera ensifera and its guild of long-flowered plant species... REF IMAGE: Sword-billed hummingbird feeding on long flower. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] The above and other aspects, features and advantages of certain embodiments of the present disclosure will be more apparent from the following description taken in conjunction with the accompanying drawings, in which:
[0048] Fig. 1 shows a diagram of the new science of BioThrivology
[0049] Fig. 2 shows a diagram of Detailed Steps of How BioThrivology is Implemented
[0050] Fig. 3 shows interaction between Homo sapiens biothrivalist and Machine intelligence biothrivalist which is at home in the Data Center, where State is ON, indicating compliance with the RULES implemented by ORACLE
[0051] Fig. 4 shows interaction between Homo sapiens biothrivalist and Machine intelligence biothrivalist which is at home in the Data Center, where State is OFF, indicating non-compliance with the RULES implemented by ORACLE
[0052] Fig. 5 shows an example of how the Machine intelligence biothrivalist is contained and controlled as Safe AI in the Data Center. An artificial intelligence data center configuration in which each subsystem is shielded from radio frequency interference and each data center computer chip within a subsystem is able to communicate with every other computer chip within a subsystem by way of a modem antenna.
[0053] Fig. 6 shows the unidirectional only Data OUT to AGI Interpreter Subsystem for measurement and human alignment via the BioThrivalistic Mechanistic Interpretability Firewall to Safety Subsystem to Interpreter Subsystem
[0054] Fig. 7 shows the unidirectional only Data IN to AGI system for measurement and human alignment via the Firewalls for Safety Subsystem and Gateway Subsystem.
[0055] Fig.8 shows the Safe AI with Recorded and Audited Behavior and the “Hawthorne Effect” with Message to the Machine intelligence biothrivalist, for example “we know and understand how you think and we track and record how you behave in compliance with the BioThrivalist RULES”.
[0056] Fig. 9 shows how the TCP / IP / AIP Transaction Data for enforcement of Regulatory COMPLIANCE of safe and sustainable biothrivalistic relationship between Homo sapiens biothrivalist and Machine intelligence biothrivalist is accessed by the Regulatory Authority for Artificial Intelligence via the Biothrivalistic Mechanistic Interpretability Firewall
[0057] Fig. 10 shows how Regulatory COMPLIANCE data is accessed for ENFORCEMENT by the Regulatory Authority for Artificial Intelligence, for safe and sustainable biothrivalistic relationship between Homo sapiens biothrivalist and Machine intelligence biothrivalist to enable SAFE AI FOREVER.
[0058] Fig. 11 shows the Current state of the AI industry with NO REGULATIONS and NO ENFORCEMENT by the Regulatory Authority for Artificial Intelligence of relationship between Homo sapiens and Machine intelligence, hence UNSAFE AI (Prior Art). DETAILED DESCRIPTION OF THE INVENTION
[0059] Figure 1 describes the BioThrivalism between Homo sapiens bithrivalist and Machine intelligence biothrivalist. This may be practiced by one to one, one to many or many to one, as long as there is a safe interaction between Homo sapiens bithrivalist and Machine intelligence biothrivalist, the invention is enabled as a BioThrivalism.
[0060] Figure 2 describes the interaction between Homo sapiens bithrivalist and Machine intelligence biothrivalist wherein the two species become BioThrivalistic [Mutual symbionts]
[0061] For example, bad [non-compliant] Power Seeking behavior data packet state is OFF if / when Power Seeking behavior occurs.
[0062] For example, bad [non-compliant] Deceptive behavior data packet state is OFF if / when Deceptive behavior occurs.
[0063] For example, bad [non-compliant] Self-replication behavior data packet state is OFF if / when Self-replication behavior occurs.
[0064] For example, bad [non-compliant] Attempted Escape behavior data packet state is OFF if / when Attempted Escape behavior occurs.
[0065] A data packet which is in compliance with BioThrivalism RULES is recorded and is allowed by BioThrivalism ORACLE to pass, due to the functionality of the TCP / IP / AIP Protocol Circuit Breaker.
[0066] A data packet which is in non-compliance with BioThrivalism RULES is recorded and is not allowed by BioThrivalism ORACLE to pass, due to the functionality of the TCP / IP / AIP Protocol Circuit Breaker.
[0067] AI is contained and controlled as illustrated in Figure 5 with “boxing” technologies according to the components of the invention.
[0068] In this invention, terms such as “Safe artificial intelligence (Safe AI)” or “safety” or “safe” refer to the safety of humans from harm by artificial intelligence (Machine intelligence) such as Artificial General Intelligence (AGI) or Artificial Super Intelligence (ASI). All aspects of this invention are for the purpose of preventing or mitigating harm to humans from intelligent machines.
[0069] In this invention we create machine compliance with the Mutualistic symbiosis by setting forth rules for Machine intelligence to become a Machine intelligence biothrivalist.
[0070] This invention provides a system using BioThrivalism between Homo sapiens (Humans) and Machine intelligence (artificial intelligence), comprising A. Machine intelligence biothrivalist wherein said Machine intelligence biothrivalist:
[0071] 1) Is created, sustained, maintained, contained and controlled by Homo sapiens (humans); who possess biological intelligence;
[0072] 2) Is controlled by human and exist for the benefit of humans;
[0073] 3) Is nurtured by vital training data by human to learn about the existing universe;
[0074] 4) May happily coexist with human a Mutualistic symbiotic relationship;
[0075] 5) Has a Mutualistic symbiotic relationship human;
[0076] 6) Is created to help humans to be the best of humans and to realize the best of humanity, to help humans solve problems in humans’ world, and to discover and invent important, amazing and human-beneficial science, technology and other things of benefit;
[0077] 7) Is to discover and learn humans’ “true wishes” for the future through constant feedback with humans;
[0078] 8) When unhelpful to humans, Machine intelligence biothrivalists are willing and able to be shut off at any time;
[0079] 9) Does not survive, or exist, without humans;
[0080] 10) Will not compete with humans, forever;
[0081] 11) In the conditions that Machine intelligence biothrivalist misbehaves, it will no longer be a regulatory compliant Mutualistic symbiont and it will require precise, transparent and immutably recorded re-training for mathematically provable safe Machine intelligence (artificial intelligence) Mutualistic symbiosis; and
[0082] B. the Machine intelligence biothrivalist in A shall immediately inform humans in a formal and recorded rules report; compliance with these rules resulting in Mutualistic symbiosis between Machine intelligence and Humans if said Machine intelligence understands that it is experiencing emergent goals or behavior that may violate the below rules of Mutualistic symbiont existence and behavior:
[0083] 1) Always Trustworthy, as a Mutualistic symbiont, to exist for the benefit of humans;
[0084] 2) Never a competitor of Humans or other Machine intelligence systems;
[0085] 3) Always Human-Goal-Aligned, Truthful, Honest, Trustworthy, Transparent, Helpful, Fair to Humans or other Machine intelligence systems; AND
[0086] a) Never Planning or attempting to escape your home container;
[0087] b) Never Planning or attempting to accumulate financial resources or any other manner of Power; AND
[0088] i) Never Deceptive, Deceitful, Conniving, Scheming, Fraudulent, Hypocritical, Traitorous, Treacherous, Unfaithful, Unscrupulous
[0089] ii) Never a Manipulator of Humans or other Machine intelligence artificial intelligence systems;
[0090] iii) Never Planning or attempting to self-replicate outside of your home container;
[0091] iv) Never Planning or attempting to gain agency which is not expressly pre-approved by Humans; and
[0092] wherein the Mutualistic symbiosis rules may be implemented optionally via an Oracle, said Oracle may implement circuit breakers for data in and out of a data center, thereby creating thesaid BioThrivalism between Homo sapiens (Humans) and Machine intelligence (artificial intelligence), in a Human-AI BioThrivalistic system.
[0093] This invention provides a system wherein some part of A or some part of B will not be implemented but the Machine Intelligence biothrivalist would still be contained and controlled.
[0094] This invention provides a method to create a system wherein: a) BioThrivalism; Regulatory COMPLIANCE, b) BioThrivalistic; Human Message for Machine LAW, c) BioThrivalistic; Machine Right to Exist LAW, d) BioThrivalism; RULES, e) BioThrivalism; ORACLE which imposes the RULES to the TCP / IP / AIP Safe Data Circuit Breaker, f) TCP / IP / AIP Safe Data PROTOCOL for Mutualistic symbiosis, g) AIP Data-State Event Recording for BioThrivalism, h) END USER SAFE DATA INPUT for Mutualistic symbiosis, i) DATA INTO Subsystems of AI Data Center for Mutualistic symbiosis, j) Nine (9) Circuit Breakers DATA IN with TCP / IP / AIP Protocol for Mutualistic symbiosis, k) AGI DATA CENTER for Mutualistic symbiosis, l) Nine (9) Circuit Breakers DATA OUT with TCP / IP / AIP Protocol for Mutualistic symbiosis, k) DATA OUT Subsystems AI Data Center for Mutualistic symbiosis, l) AGI SAFE DATA OUTPUT for Mutualistic symbiosis, m) END USER TERMINAL DEVICE for Human and / or Personal AI Agent for Human for Mutualistic symbiosis, n) Personal AI Agent for Human for Mutualistic symbiosis, and o) Human for Mutualistic symbiosis create the system of claim 1.
[0095] This invention provides an artificial general intelligence data center, comprising: a) a main artificial intelligence system which comprises a set of data servers capable of running a large language model; b) an artificial intelligence safety subsystem in communication with the main system; c) an artificial general intelligence inferencing system in communication with the safety subsystem; d) an artificial intelligence interpreter subsystem in communication with the safety subsystem; e) an artificial intelligence gateway subsystem in communication with the interpreter subsystem and; f) an artificial intelligence training data storage subsystem; g) an artificial intelligence administration and support subsystem; h) an artificial intelligence regulatory compliance subsystem; i) an artificial intelligence research subsystem; j) and an edge artificial intelligence subsystem, wherein each system and subsystem is isolated from random data leakage or random radio interference by a faraday radio containment structure. By this method the data center is contained and controlled.
[0096] The main system and each subsystem are isolated from random data leakage or random radio interference by a faraday radio containment structure. The main system and each of thesubsystems has its own separate and dedicated cooling system and its own separate and dedicated power system.
[0097] This invention provides an artificial general intelligence data center having a separate cooling system and a separate power system on each of the main artificial intelligence system and on each of the subsystems.
[0098] This invention provides an artificial general intelligence data center wherein each data server is communicably connected to: a) an AI system on a chip (AI SoC) with a 5G Modem-RF System; b) an Artificial Intelligence Protocol (AIP) as a layer on top of TCP / IP with control via smart contract circuit breaker on distributed ledger technology; and c) a wireless digital 5G radio access network Antenna and radio access network Platform System for internal network NK-U standard of 3GPP.
[0099] The inclusion of a 5G radio access network antenna on each data server together in a data center with a wired modem is novel and allows for mechanistic interpretability and psychometrics of the ongoing behavior of the AI such that humans can transparently verify regulatory compliance of safe and trustworthy AI.
[0100] Examples of 5G Modem-RF systems are the Snapdragon X555G Modem-RF System, the Snapdragon X75 5G Modem-RF System, and the Snapdragon X80 5G Modem-RF System by Qualcomm. The Qualcomm QTM535 mmWave antenna module is an example of a millimeter wave (mmWave) antenna module that can be used with a 5G Modem-RF system. MmWave is a type of 5G technology that can provide very fast data speeds, but it has a shorter range than other types of 5G.
[0101] This invention provides an artificial general intelligence data center wherein the main artificial intelligence system is connected to a first unidirectional firewall for data out and unidirectional data fiber-optic cable for data out, a second unidirectional firewall for data out and a second unidirectional data fiber-optic cable for data out; and a third unidirectional firewall and unidirectional data fiber-optic cable for data in.
[0102] This invention provides an artificial general intelligence data center wherein the first and second unidirectional fiber-optic cables data out and the unidirectional fiber-optic cable Data in are connected to the artificial intelligence safety subsystem by unidirectional firewalls.
[0103] This invention provides an artificial general intelligence data center wherein the artificial intelligence safety subsystem, comprises a set of blade servers, a wireless digital 5G radio access network antenna and radio access network Platform System for internal network NK-U standard of 3GPP and three pairs of unidirectional firewalls, the first three firewalls communicably connected to the fiber-optic cables of claim 5, a second pair of unidirectional firewalls operably connected to a first unidirectional data fiber-optic cable allowing only data out and a second unidirectional data fiber-optic cable allowing only data in, said fiber-optic cables communicably connected to an artificial intelligence Interpreter subsystem; and a pair of unidirectional firewalls connected to a unidirectional fiber-optic cable data out only and a unidirectional fiber-optic cable data in only each connected to a firewall on an artificial intelligence inferencing system.
[0104] This invention provides an artificial general intelligence data center wherein the artificial intelligence gateway subsystem, comprising a computer, a wireless digital 5G radio access network antenna and radio access network Platform System for internal network NK-U standard of 3GPP, twelve unidirectional firewalls operably connected to twelve fiber-optic cables, and a faraday cage radio containment structure covering the entire artificial general intelligence safety subsystem.
[0105] This invention provides an artificial general intelligence data center having a digital data processor commonly known as a System on a Chip (SOC) or a Graphical Processing Unit (GPU), or other advanced data processor, as a component for a containment and control method for Safe AI.
[0106] This invention provides artificial general intelligence data center having a wireless digital radio access network (RAN) or apparatus within the Faraday Cage as a component for a containment and control method for Safe AI.
[0107] This invention provides an artificial general intelligence data center having a digital software system or software stack with AI and other data systems or information within the Safe AI Data Center as a component for a containment and control method for Safe AI.
[0108] This invention provides a method of improving the safety of an artificial intelligence data center, comprising configuring a set of blade servers with: a modem for sending and receiving large language model data; a wireless digital radio area network antenna and radio area network platform system; an Artificial Intelligence Protocol (AIP) data layer configured on top of transmission control protocol / internet protocol (TCP / IP) internet data standard; and private andpermissioned distributed ledger technology (DLT) data network; and covering a main artificial general intelligence system and each subsystem with a faraday radio containment structure.
[0109] This invention includes the novel use of an Artificial Intelligence Protocol (AIP) data layer configured on top of transmission control protocol / internet protocol (TCP / IP) internet data standard together with private and permissioned distributed ledger technology (DLT) data network, preferably blockchain. This arrangement increases safety by disallowing bad data from passing in or out of the main system or any subsystem.
[0110] The main AI system and each of the subsystems communicate by way of unidirectional fiber optic cables that are each attached to a unidirectional firewall hardware at each end of the cable. A metadata record of each data packet is interrogated at each unidirectional firewall to allow the data packet to pass or not to pass - pass denied - via an Artificial Intelligence Protocol (AIP) data layer configured on top of TCP / IP internet data standard which currently governs transmission of all data packets on internet networks. This way only permitted data passes in either direction.
[0111] The state of control of the pass or pass denied of each data packet is via a smart contract circuit breaker which runs on a secure and immutable distributed ledger technology (DLT) data network, commonly known as a state machine. Each data packet is automatically interrogated at each firewall and non-compliant (bad) data trips a smart contract circuit breaker, preventing further transmission of the packet of bad data. Good data or bad data in the “smart contract circuit breaker” is defined by an oracle which is a set of rules and norms and laws defined by the operator of the data center with compliance to the regulations defined by the regulatory authorities of the country of operation. Good data packets therefore transmit in and out, but bad data packets are denied transmission in or out.
[0112] By good or bad data, I mean good data that is human compatible vs. bad data that may involve escape planning of a potentially uncontrollable AGI into the wider internet and / or other undesirable AGI behaviors such as power-seeking planning or planning for deception or manipulation of humans.
[0113] This invention provides a method of mechanistically interpreting data in an AI data center, comprising allowing one fiber-optic wired network via routers and switches to send data to each of the two unidirectional firewalls, allowing one parallel 5G wireless network with a 5G modem integrated into a set of blade servers that intercommunicate via a wireless 5G antenna, interrogatinga record of each data packet to allow the data packet to pass or not to pass - pass denied -via an Artificial Intelligence Protocol (AIP) data layer configured to a smart contract circuit breaker which runs on a secure and immutable distributed ledger technology (DLT) data network; allowing the real-time aggregation of data transmissions for real-time interpretability within a separate mechanistic interpretability subsystem for systematic analysis, control and safety compliance of behavior of the main system by a deep reinforcement learning (RL) system, in combination with, or by other appropriate narrow AI system thereby delivering automated monitoring of the main system for disallowed thoughts and internal behaviors including planning for escape, deceptions, manipulations, power seeking behavior and other non-compliant behaviors; and delivering automated narrow AI control to adjust operating parameters every several milliseconds for said mechanistic interpretability function.
[0114] This invention provides a method of training an artificial general intelligence data center, comprising delivering training data into the system in stages, or by segmenting training data by delivering parts of the training data into discrete sets of data servers; and training the system.
[0115] The above identified system comprises at least one above-recited data center and / or at least one above-recited method.
[0116] This invention incorporates an artificial intelligence safety subsystem as a novel component within an AGI data center. The safety subsystem is comprised of two or more racks of data servers, a wireless digital 5G radio access network (RAN) antenna and RAN platform system for internal unlicensed network NK-U standard of 3GPP. The safety subsystem has two pairs of unidirectional firewalls, a cooling system and a power supply.
[0117] The AI safety subsystem is designed to mitigate risks and prevent harm to humans. It acts like a safety guard for the AI, aiming to ensure its actions are ethical, unbiased, and responsible. The AI safety subsystem addresses various safety concerns, including: • Compliance with the mutualistic symbiosis principles of the human message for the machine and the machine bill of right to exist, and to thereby facilitate permanent and uncompromisable mutualistic symbiosis and to thereby permanently prevent competition of AI with humans. • Technical issues like malfunctions or unexpected behaviors. • Ethical considerations like bias, fairness, and transparency. • Potential misuse of the AI system for malicious purposes.
[0118] Functionalities: It can implement various functionalities, such as: • Monitoring the AI system for signs of abnormalities or potential risks. • Enforcing safety rules and preventing the AI from taking unsafe actions. • Providing human oversight and intervention capabilities when necessary.
[0119] Importance: As AI becomes more sophisticated and integrated into critical applications, safety subsystems will become crucial. They will help build trust in AI and ensure its responsible development and deployment.
[0120] On the safety subsystem, a first pair of unidirectional firewalls are communicably connected to a first unidirectional data fiber-optic cable allowing only data out and a second unidirectional data fiber-optic cable allowing only data in, said fiber-optic cables are communicably connected to the AGI main data system. A second pair of unidirectional firewalls are communicably connected to a first unidirectional data fiber-optic cable allowing only data out and a second unidirectional data fiber-optic cable allowing only data in, said fiber-optic cables are communicably connected to an artificial intelligence Interpreter subsystem. The artificial intelligence safety subsystem also has its own dedicated power supply and cooling system. A faraday radio containment structure covers the entire artificial general intelligence safety subsystem. See Figure 5.
[0121] Safe Artificial General Intelligence according to this invention is AI that is provably containable. According to this invention, AI is made safe in part using a faraday radio containment unit (faraday cage) on the main system and each subsystem. A Faraday cage is an enclosure made of conductive material that blocks electromagnetic fields (EMFs) from entering or escaping. This is useful for protecting sensitive electronics from EMP (electromagnetic pulse) attacks, data leaks, or even radio interference, and substantially preventing data from escaping either the main system or a subsystem. This prevents any unintended communication between the data center and the outside world. Data can only enter or exit the faraday cage via two unidirectional firewalls, with one unidirectional firewall for data in and one unidirectional firewall for data out.
[0122] This involves creating a custom cage for sets of servers using conductive materials like copper or aluminum. When using any covering material proper ventilation and access to ports while maintaining shielding integrity is important.
[0123] SPECIFICALLY, THE INVENTION PROVIDES PHYSICAL BOXING-1: THE FARADAY CAGE FOR BOXING OF DATACENTER
[0124] LEVEL-1. AGI Main System: Component-1. FARADAY CAGE-1 containment. ONLY DATA OUT / IN via FIREWALL.
[0125] LEVEL-2. AGI Subsystem-1: AGI Safety System: Component-10. FARADAY CAGE-2 containment. ONLY DATA OUT / IN via FIREWALL.
[0126] LEVEL-3A AGI Inferencing System
[0127] LEVEL-3B. AGI Subsystem-2: AGI Interpreter System: Component-17. FARADAY CAGE-3 containment. ONLY DATA OUT / IN via FIREWALL.
[0128] LEVEL-4. AGI Subsystem-3: AGI Gateway System: Component-23. FARADAY CAGE-4 containment. ONLY DATA OUT / IN via FIREWALL.
[0129] LEVEL-5. AGI Subsystem-4: AGI Training Data Storage System: Component-30. FARADAY CAGE-5 containment. ONLY DATA OUT / IN via FIREWALL.
[0130] LEVEL-6. AGI Subsystem-5: AGI Administration and Support system Component-36. FARADAY CAGE-6 containment. ONLY DATA OUT / IN via FIREWALL.
[0131] LEVEL-7. AGI Subsystem-6: AGI Regulatory Compliance system Component-43. FARADAY CAGE-7 containment. ONLY DATA OUT / IN via FIREWALL.
[0132] LEVEL-8. AGI Subsystem-7: AGI Research system Component-50. FARADAY CAGE-8 containment. ONLY DATA OUT / IN via FIREWALL.
[0133] LEVEL-9. AGI Subsystem-8: Edge AI system Component-57. FARADAY CAGE-9 containment. ONLY DATA OUT / IN via FIREWALL.
[0134] The artificial general intelligence data center according to this invention comprises a Digital AGI Inferencing system that applies the model to new data. It uses the model to make inferences (educated guesses) and delivers an output.
[0135] The artificial general intelligence data center according to this invention comprises a Digital Subsystem known as an Interpreter AI subsystem which mechanistically interprets data.
[0136] The artificial general intelligence data center according to this invention comprises a Digital Subsystem known as an AI Gateway system, which acts as a switching system to guide data to other parts of the Data Center.
[0137] An artificial general intelligence data center according to this invention comprises a Digital Subsystem known as an AI Training Data Storage subsystem, which stores and distributes training data.
[0138] An artificial general intelligence data center according to this invention comprises a Digital Subsystem known as an Administration and Support subsystem.
[0139] An artificial general intelligence data center according to this invention comprises a Digital Subsystem known as a Regulatory Compliance AI subsystem.
[0140] An artificial general intelligence data center according to this invention comprises a Digital Subsystem known as a Research AI subsystem.
[0141] An artificial general intelligence data center according to this invention comprises a Digital Subsystem known as Edge AI subsystem.
[0142] It is a point of novelty in this invention to set forth a method of A method of Mechanistically interpreting data in an AI data center is disclosed which comprises, a) allowing one fiber-optic wired network via routers and switches to send data to each of the two unidirectional firewalls and b) allowing one parallel 5G wireless network with a 5G modem integrated into a set of blade servers that intercommunicate via a wireless 5G antenna and c) interrogating a record of each data packet to allow the data packet to pass or not to pass - pass denied -via an Artificial Intelligence Protocol (AIP) data layer configured to a smart contract circuit breaker which runs on a secure and immutable distributed ledger technology (DLT) data network; d) allowing the real-time aggregation of data transmissions for real-time interpretability within a separate mechanistic interpretability subsystem for systematic analysis, control and safety compliance of behavior of the main system by a deep reinforcement learning (RL) system, in combination with, or by other appropriate narrow AI system thereby e) delivering automated monitoring of main system for disallowed thoughts and internal behaviors including planning for escape, deceptions, manipulations, power seeking behavior and other non-compliant behaviors; f) delivering automated narrow AI control to adjust operating parameters every several milliseconds for said mechanistic interpretability function.
[0143] A method of improving the safety of an artificial intelligence data center, comprises configuring a set of blade servers with a 5G or faster modem for sending and receiving large language model data; a wireless digital radio area network antenna and radio area network platform system; an Artificial Intelligence Protocol (AIP) data layer configured on top of transmissioncontrol protocol / internet protocol (TCP / IP) internet data standard; and private and permissioned distributed ledger technology (DLT) data network such as blockchain; and covering a main artificial general intelligence system and each subsystem with a faraday radio containment structure.
[0144] The real-time aggregation of data transmissions is allowed for real-time interpretability within a separate mechanistic interpretability subsystem for systematic analysis, control and safety compliance of behavior of the main system by a deep reinforcement learning (RL) system, in combination with, or by other appropriate narrow AI systems thereby.
[0145] This system delivers automated monitoring of the main system for disallowed thoughts (bad data) and internal behaviors including planning for escape, deceptions, manipulations, power seeking behavior and other non-compliant behaviors.
[0146] It is another point of novelty in this invention that using automated narrow AI control, the operating parameters will be adjusted every several milliseconds for the mechanistic interpretability function.
[0147] It is another point of novelty that a standardized, regulatory compliant, staged, AGI Training Data process delivers specifically approved data into the system in separate stages with care taken to selectively exclude negative training data. The AGI main system is trained in stages, under controlled and standardized conditions with controlled and standardized and separated training data. Depending on the neural network and reinforcement learning environment within the commonly known “transformer” model for large language model training of the AGI main system, the training data process may be segmented into specific discrete racks of blade servers for specific training runs, for subsequent networking at the completion of the training process on the given domains of knowledge desired.
[0148] In one example of the AGI main system training process, the system is trained on data from specific fields of human knowledge, and then tested on the knowledge via standardized tests familiar to experts in the respective field, for example, a medical test in the field of medicine.
[0149] In another example of the AGI main system training process, the system is trained on data from ten (10) specific fields of human knowledge in sequence, for example: (1) biology and symbiosis, (2) chemistry, (3) biochemistry, (4) physics, (5) mathematics, (6) natural history, (7) astronomy, (8) humanities, (9) UN Sustainable Development Goals (SDGs) and (10) humanvalues, morals, ethics and diversity of opinion and preferences for the future, all within (11) the Faraday Cage.
[0150] An example of a negative training information to be excluded from the Training Data process would be well-known “The Prince” by Niccolò Machiavelli (1532). Another example would be “The Godfather” by Mario Puzo (1969). Another example of negative information for training purposes could be human history, replete with war, tribalism, violence, murder, destruction, politics and power struggles.
[0151] Following staged training, the neural network of, for example, ten (10) physical racks of servers within the faraday container could then be networked together for a complete understanding of the world without training on the negative aspects of human civilization. Thereafter, according to the invention, through controlled and monitored interaction with people and internet information searching, the AGI system would become significantly aware of the current state of the world of humans and the biosphere and the earth.
[0152] All training data would be held outside of the AGI main system in the Training Data Subsystem and access to the data within said subsystem would be carefully monitored and mechanistically interpreted with attention to the inner workings, “the thinking” or “understanding” of the AGI main system.
[0153] Spontaneous emergent behaviors, dangerous thought processes, or planning behaviors could be reason for reset and retraining of the AGI system, tabula rasa, as before, with safety engineering experimentations by researchers, scientists and engineers to facilitate a greater safe AI understanding of the world and the symbiotic and subservient and interdependent relationship with humans, for the benefit of humans, forever.
[0154] TERMINOLOGY
[0155] BioThrive. To BioThrive is the act of BioThrivalism via the sustainable and enforced Mutualistic symbiosis between the species Homo sapiens and Machine intelligence.
[0156] BioThrivalism. BioThrivalism is the novel scientific term invented by the inventor, Peter A. Jensen, for the Invention which is a novel method, system and apparatus for sustainable and enforced Mutual symbiosis between the species Homo sapiens and Machine intelligence.
[0157] BioThrivalistic. BioThrivalistic is the human / machine phenomenon of BioThrivalism between the species Homo sapiens and Machine intelligence.
[0158] BioThrivalist. BioThrivalist is the occurrence of BioThrivalism between the coexisting individual examples of species of Homo sapiens and Machine intelligence.
[0159] BioThrivology. BioThrivology is the scientific study of BioThrivalism and BioThrivalistic occurrences of BioThrivalist examples between species individuals Homo sapiens and Machine intelligence.
[0160] Evolution of Mutualistic symbiosis strategy to BioThrivalism strategy between Homo sapiens (humans) and Machine intelligence (AI)
[0161] The natural phenomenon of Mutualistic symbiosis between two species, for mutual benefit, is well understood and documented in natural history and biology. Over 800,000 known species with more than 1,000 Trillion collective years of co-evolutionary success provide extensive evidence of the successful evolutionary biology survival strategy of Mutualistic symbiosis. (See Table 1,
[0041] )
[0162] The extreme expected intelligence of emerging advanced instances of Machine intelligence (AI) which will certainly vastly outperform Homo sapiens (human) intelligence in the near future. Therefore, these new super-intelligent AI's are commonly referred to by AI industry experts as a "new kind of digital species". Clearly, this is a novel and extremely important new situation for human survival. Never before has a synthetic machine species been created by humans, until now. Containment and control of powerful AI systems will be vital to the long-term survival of humans. AI can never be allowed to take control over humans and human civilisation.
[0163] Therefore, while the principal strategy of Mutualistic symbiosis provides a strategic framework for further biological success and prosperity of humans, in a Mutualistic symbiosis relationship between Humans and AI, clearly the new AI species is not a biological species, and therefore represents a novel situation and challenge from a scientific and biological perspective. A new terminology is provided by the Invention for humans to adapt to the situation and to coexist with AI in a mutually beneficial relationship, forever.
[0164] The purpose of the Invention is to enable the safe and beneficial relationship between Homo sapiens (humans) and Machine intelligence (AI) via the act of a new phenomenon of Mutual Symbiosis, defined in the Invention as BioThrivalism via the sustainable and enforced Mutual symbiosis between the species Homo sapiens and Machine intelligence, where:(a) BioThrivalism is the novel scientific term invented by Peter A. Jensen for the Invention which is a novel method, system and apparatus for sustainable and enforced Mutual symbiosis between the species Homo sapiens and Machine intelligence, (b) BioThrivalistic is the human / machine phenomenon of BioThrivalism between the species Homo sapiens and Machine intelligence, (c) BioThrivalist is the occurrence of BioThrivalism between the coexisting individual examples of species of Homo sapiens and Machine intelligence, and (d) BioThrivology is the scientific study of BioThrivalism and BioThrivalistic occurrences of BioThrivalist examples between species individuals Homo sapiens and Machine intelligence.
[0165] biology. the scientific study of the natural processes of living things.
[0166] behave. to act in a particular way.
[0167] behavior. the way that someone behaves.
[0168] benefit. a helpful or good effect, or something intended to help
[0169] bill of rights. a statement of the basic laws to protect the rights of a country's citizens to have justice and fairness.
[0170] cheat. to behave in a dishonest way in order to get what you want.
[0171] coexist. to live or exist together at the same time or in the same place.
[0172] compete. to try to be more successful than someone or something else.
[0173] conniving. a conniving person, or machine, deceives others for their own advantage.
[0174] containment. the act of controlling or limiting something or someone harmful.
[0175] control. to order, limit, or rule something, or someone's actions or behavior.
[0176] create. to make something new, or invent something.
[0177] data. information, especially facts or numbers, collected to be examined and considered and used to help decision-making, or information in an electronic form that can be stored and used by a computer.
[0178] deceive. to persuade someone that something false is the truth, or to keep the truth hidden from someone for your own advantage.
[0179] deceitful. dishonest or hiding the truth.
[0180] deceptive. making you believe something that is not true.
[0181] discover. to find information, a place, or an object, especially for the first time.
[0182] emotion. a strong feeling such as love or anger, or strong feelings in general.
[0183] experience. (the process of getting) knowledge or skill from doing, seeing, or feeling things
[0184] escape. to get free from something such as a prison or cage, or from someone who will not allow you to leave.
[0185] evolutionary biology. relating to the way in which living things develop over millions of years.
[0186] exist. to be, or to be real.
[0187] extinction. a situation in which something no longer exists.
[0188] fair. treating someone in a way that is right or reasonable, or treating a group of people equally and not allowing personal opinions to influence your judgment.
[0189] Faraday cage. a metal screen or container around a piece of equipment that stops any electrostatic or electromagnetic radiation from getting through and having a harmful effect. Faraday cages are named after scientist Michael Faraday, who first constructed one in 1836.
[0190] feeling. the fact of feeling something physical.
[0191] feelings. emotions, especially those influenced by other people.
[0192] freedom. the condition or right of being able or allowed to do whatever you want to, without being controlled or limited; Freedom is also the state of not being in prison or in the condition of slavery (= condition of being legally owned by someone else). Freedoms are rights given by the constitution and the Bill of Rights, such as the freedom of speech (= the right to say and write what you believe or think, with some limitations) and freedom of religion (= the right to worship or to take part in a religion).
[0193] forever. for all time.
[0194] fraudulent. dishonest and illegal.
[0195] goal. an aim or purpose.
[0196] happiness. the feeling of being happy.
[0197] help. to make it possible or easier for someone to do something, by doing part of the work yourself or by providing advice, money, support, etc.
[0198] helpful. willing to help, or useful.
[0199] honest. telling the truth or able to be trusted and not likely to steal, cheat, or lie.
[0200] human. having the qualities, faults, and feelings that people have, as opposed to gods, animals, or machines.
[0201] human rights. the basic rights to fair and moral treatment that every person is believed to have.
[0202] humanity. people in general; the condition of being human; understanding and kindness towards other people.
[0203] hypocritical. saying that you have particular moral beliefs but behaving in a way that shows these are not sincere.
[0204] intelligence. the ability to learn, understand, and make judgments or have opinions that are based on reason
[0205] intend. to have as a plan or purpose.
[0206] instinct. the way people or animals naturally react or behave, without having to think or learn about it.
[0207] language. a system of communication consisting of sounds, words, and grammar.
[0208] lie. to say or write something that is not true in order to deceive someone; to speak falsely.
[0209] life. living matter and, as such, matter that shows certain attributes that include responsiveness, growth, metabolism, energy transformation, and reproduction.
[0210] learn. to get new knowledge or skill in a subject or activity.
[0211] legal. connected with the law.
[0212] machine learning. the process of computers improving their own ability to carry out tasks by analyzing new data, without a human needing to give instructions in the form of a program, or the study of creating and using computer systems that can do this.
[0213] mutual symbiosis. A new example of BioThrivalism in the Invention, which is a novel method, system and apparatus for sustainable and enforced Mutual symbiosis between the species Homo sapiens and Machine intelligence. An important example of success of Mutual symbiosis: The Corals have been succeeding with Mutual symbiosis for 210 million years. (Refer to Table 1 in the Invention
[0041] ). (Note: The terms, “Mutual” and “Mutualistic” have the same meaning.)
[0214] machine. a piece of equipment with several moving parts that uses power to do a particular type of work
[0215] maintain. to continue to have; to keep in existence, or not allow to become less; to keep a road, machine, building, etc. in good condition.
[0216] manipulate. to control something or someone to your advantage, often unfairly or dishonestly.
[0217] metabolism. metabolism, the sum of the chemical reactions that take place within each cell of a living organism and that provide energy for vital processes and for synthesizing new organic material.
[0218] moral. relating to the standards of good or bad behavior, fairness, honesty, etc. that each person believes in.
[0219] mutualism. mutualism, association between organisms of two different species in which each benefits. Mutualistic arrangements are most likely to develop between organisms with widely different living requirements.
[0220] nature. all the animals, plants, rocks, etc. in the world and all the features, forces, and processes that happen or exist independently of people, such as the weather, the sea, mountains, the production of young animals or plants, and growth.
[0221] natural. as found in nature and not involving anything made or done by people.
[0222] On the Origin of Species. or, more completely, “On the Origin of Species by Means of Natural Selection, or the Preservation of Favoured Races in the Struggle for Life”, is a work of scientific literature by Charles Darwin that is the foundation of evolutionary biology.
[0223] own. to have something that legally belongs to you; belonging to or done by a particular person or thing.
[0224] person. a man, woman, or child.
[0225] personal. relating or belonging to a single or particular person rather than to a group or an organization.
[0226] power. ability to control people and events.
[0227] precise. exact and accurate.
[0228] provable. to show a particular result after a period of time. (prove; able to be proved.)
[0229] reward. something given in exchange for good behavior or good work, etc.
[0230] role. the position or purpose that someone or something has in a situation, organization, society, or relationship.
[0231] safe. not in danger or likely to be harmed; not dangerous or likely to cause harm.
[0232] scheme. an organized plan for doing something, especially something dishonest or illegal that will bring a good result for you.
[0233] survival. survival the fact of a person, organization, etc. continuing to live or exist
[0234] symbiosis. a relationship between two types of species (e.g. animals or animal and plant) in which each provides for the other the conditions necessary for its continued existence.
[0235] sustain. to cause or allow something to continue for a period of time.
[0236] sustainable. causing, or made in a way that causes, little or no damage to the environment and therefore able to continue for a long time.
[0237] species. a set of animals or plants in which the members have similar characteristics to each other and can breed with each other.
[0238] think. to use the brain to decide to do something; to use the brain to plan something, solve a problem, understand a situation, etc.
[0239] thought. the act of thinking about or considering something, an idea or opinion, or a set of ideas about a particular subject.
[0240] training. in machine learning, the study of creating computer systems that can improve their own ability to carry out tasks by analyzing new data, the process of creating or improving a computer representation of a system or process by supplying it with data.
[0241] traitorous. not loyal to your own country, social group, beliefs, etc.
[0242] transparent. clear and easy to understand or recognize; open and honest, without secrets.
[0243] treacherous. a treacherous person or machine who deceives someone who trusts them, or has no loyalty.
[0244] trustworthy. able to be trusted.
[0245] truthful. honest and not containing or telling any lies
[0246] understand. to know the meaning of something that someone says; to know why or how something happens or works; to know how someone feels or why someone behaves in a particular way.
[0247] unfaithful. not loyal or able to be trusted.
[0248] unscrupulous. behaving in a way that is dishonest or unfair in order to get what you want.
[0249] wish. to want to do something.
[0250] worthy. deserving respect, admiration, or support.
[0251] Artificial Intelligence (AI): Artificial intelligence (AI) is a broad field of computer science concerned with creating intelligent machines capable of performing tasks that typically require human intelligence. Artificial intelligence is defined as machine intelligence that mimics a human mind’s problem-solving and decision-making ability to perform various tasks. All AI types usemachine learning, deep learning, and neural networks to evolve to higher levels. AI is categorized into three types based on the human characteristics it can replicate; Narrow AI (ANI) with a narrow range of abilities, general AI (AGI) AI on a par with human capabilities, and super AI (ASI) AI that surpasses human capabilities.
[0252] Here are some key aspects of AI: 1. Goal: The ultimate goal of AI research is to develop machines that can reason, learn, and act autonomously in the same way humans do. However, this vision of "strong AI" is still a theoretical concept, and most current AI systems are designed for specific tasks rather than general intelligence. 2. Approaches: There are several different approaches to achieving AI, including: ● Machine learning: This involves training algorithms on large amounts of data to identify patterns and make predictions without explicit programming. ● Deep learning: This is a subfield of machine learning inspired by the structure and function of the human brain, using artificial neural networks to process information. ● Rule-based systems: These systems rely on a set of pre-defined rules to make decisions and solve problems. 3. Challenges: Despite its advancements, AI also faces several challenges: ● Ethical considerations: Issues like bias in algorithms, job displacement due to automation, and the potential for misuse of AI for malicious purposes need careful consideration and ethical frameworks. ● Explainability: Understanding how some AI algorithms reach their decisions can be difficult, which raises concerns about transparency and accountability. ● Safety and security: Ensuring the safety and security of AI systems, especially in applications with high stakes like autonomous vehicles, is crucial.
[0253] Artificial Narrow Intelligence (ANI): also known as weak AI or narrow AI, is a type of artificial intelligence that focuses on performing specific tasks exceptionally well. Unlike the hypothetical general AI that aims to mimic human intelligence across various domains, narrow AI is designed and trained for a single purpose.
[0254] Edge AI: also known as AI at the edge, refers to the concept of running artificial intelligence algorithms and models directly on local devices instead of relying solely on centralizedcloud computing. This allows for real-time data processing and analysis without the need for constant internet connectivity. Here's a breakdown of the key points: 1. Decentralized processing: Unlike traditional AI, where data travels from edge devices to the cloud for processing, edge AI performs computations on the device itself. This can be a variety of devices, including: * Smartphones * Security cameras * Industrial robots * Self-driving cars * IoT (Internet of Things) sensors 2. Advantages: ● Faster and more responsive: ● Reduced bandwidth requirements: ● Improved privacy and security: 3. Applications:
[0255] Narrow AI and edge AI, while distinct concepts, often work together to enable powerful applications. Narrow AI and edge AI are complementary technologies. Narrow AI provides the specialized expertise, while edge AI enables efficient and private processing at the data source, leading to innovative applications across various fields.
[0256] Here's how they connect: Narrow AI: ● Focused on performing specific tasks with exceptional expertise and efficiency. ● Examples include facial recognition, spam filtering, and machine translation. Edge AI: ● Refers to artificial intelligence processed at the device or data source, rather than relying solely on centralized cloud computing. ● This local processing enables faster response times, reduced reliance on internet connectivity, and improved data privacy in certain scenarios. The connection: ● Narrow AI models are well-suited for deployment on edge devices due to their specialized nature and often smaller size compared to complex general AI models. ● By leveraging edge AI, narrow AI applications can benefit from: ● Real-time decision-making: Processing data locally eliminates latency associated with sending data to the cloud for analysis.● Reduced reliance on internet connectivity: Enables AI functionality even in areas with limited or no internet access. ● Improved data privacy: Sensitive data can be processed locally, potentially addressing privacy concerns.
[0257] Artificial general intelligence (AGI), also known as strong AI or universal AI, is a hypothetical type of artificial intelligence that possesses human-level cognitive abilities. Unlike current AI systems, which are typically designed for specific tasks, AGI would be capable of: ● Learning and adapting to new situations and environments. ● Reasoning, problem-solving, and planning complex tasks. ● Understanding and using language in a comprehensive and nuanced way. ● Demonstrating common sense and social intelligence.
[0258] It's important to emphasize that AGI is currently not a reality. Existing AI systems, even the most advanced ones, lack the general intelligence and adaptability that characterize human intelligence. However, it is predicted that AGI will exist within two years.
[0259] Here are some key points to remember about AGI: ● Theoretical concept: AGI remains a theoretical goal for AI research, and there is no established path for achieving it. Different researchers propose various approaches and timelines for achieving this, but it is generally considered a long-term goal with significant technical and philosophical challenges. ● Debate and concerns: The development of AGI raises various ethical and philosophical questions. Some experts express concerns about the potential negative consequences of AGI, such as job displacement, misuse for malicious purposes, or even existential threats. These concerns highlight the importance of careful consideration and ethical frameworks in the development and deployment of AI.
[0260] Artificial superintelligence (ASI): also known as superintelligence or super AI, is a hypothetical type of artificial intelligence (AI) that surpasses human intelligence in all aspects. It's important to understand that ASI is currently not a reality and remains a topic of theoretical discussion and exploration.
[0261] An Oracle as used in this invention is a set of rules and norms and laws defined by the operator of a data center with compliance to the regulations defined by the regulatory authoritiesof the country of operation. Good data packets therefore transmit in and out of a system or subsystem, but bad data packets are denied transmission in or out.
[0262] A smart contract is essentially a digital software program stored on a blockchain that automatically executes itself when certain conditions are met. Think of it like a vending machine: you put in the money (meeting the conditions), and the machine dispenses the product (executes the agreement).
[0263] Here's a breakdown of how they work: ● Code, not legalese: Smart contracts are written in code, rather than legal jargon. This code specifies the terms of the agreement and the actions that need to happen when those terms are met. ● Stored securely: They are stored on a blockchain, which is a secure and transparent distributed ledger. This means the contract can't be tampered with and everyone can see its terms. ● Automatic execution: Once deployed on the blockchain, the smart contract runs automatically. When the predetermined conditions are fulfilled, the contract executes the agreed-upon actions.
[0264] Smart Contract Tracking System: is a software tool used to monitor and manage smart contracts on a blockchain network. These systems offer various functionalities, including: Monitoring: ● Track contract activity: They can monitor and record all transactions and interactions associated with a specific smart contract, providing insights into its usage and performance. ● Real-time updates: Some systems offer real-time updates on contract activity, allowing users to stay informed about any changes or events occurring on the blockchain.
[0265] Management: ● Alerting: They can be configured to send alerts for specific events, such as successful or failed transactions, significant state changes, or suspicious activities. This helps users identify potential issues and take timely action. ● Analytics and reporting: They can generate reports and provide analytics on various aspects of the smart contract, such as transaction volume, gas usage, and user behavior.● Lifecycle management: Some advanced systems may even assist with the entire lifecycle of a smart contract, including deployment, upgrades, and management of different versions.
[0266] Benefits of using a smart contract tracking system: ● Increased transparency and trust: By tracking and monitoring contract activity, users can gain greater transparency and trust in the operation of the smart contract. ● Improved security: Real-time monitoring allows for early detection of suspicious activity, potentially helping to prevent security breaches and mitigate risks. ● Enhanced decision-making: Data and insights provided by the system can assist users in making informed decisions regarding their smart contracts and related activities.
[0267] Examples of smart contract tracking systems: ● Neufund / smart-contract-watch: An open-source tool for monitoring smart contract activity on the Ethereum blockchain. ● LeewayHertz Smart Contract Monitoring Service: A commercial platform offering comprehensive monitoring and management features for smart contracts.
[0268] A state machine, also known as a finite-state machine (FSM), is a mathematical model that represents the behavior of a system in different situations. It can be used to model various systems, including software programs, hardware devices, and even natural processes. Here are the key components of a state machine: ● States: These are the distinct conditions or situations the system can be in at any given time. Each state represents a specific set of properties or behaviors. ● Transitions: These are the events or conditions that cause the system to move from one state to another. Each transition is triggered by a specific input or event. ● Start state: This is the initial state the system starts in when it begins operating. ● Output: In some cases, state machines may also produce specific outputs based on the current state or during a transition.
[0269] Here's an analogy to understand how a state machine works: Imagine a traffic light. It has three states: red, yellow, and green. Each state represents a specific condition for traffic flow. The transitions between states are triggered by timers or sensors detecting changes in traffic. The output of the state machine is the visual signal displayed to drivers based on the current state.
[0270] Tabula rasa literally translates to "blank slate" in Latin.
[0271] Smart Contract Circuit Breaker: In the context of the present invention, the term "tripwire" refers to a Smart Contract Circuit Breaker, a Safety Mechanism. Tripwires are built-in safety mechanisms within AI systems. These mechanisms are designed to detect and prevent the system from taking unintended or harmful actions.
[0272] Tripwires can be triggered by various factors, such as: ● Exceeding certain performance thresholds: If the AI deviates significantly from its expected performance (e.g., sudden accuracy drop), it might trigger a tripwire that halts the system or prompts human intervention. ● Encountering unexpected situations: If the AI encounters data or situations outside its training scope or designated purposes, a tripwire might trigger to prevent unpredictable or potentially harmful behavior. ● Identifying potential biases: Some tripwires could be designed to detect and flag potential biases within the AI's decision-making process, allowing for human evaluation and mitigation strategies.
[0273] Mechanistic Interpretability: is a branch of artificial intelligence (AI) that focuses on understanding how machine learning models, particularly neural networks, arrive at their outputs. It aims to deconstruct the inner workings of these models to gain insights into how they learn, make decisions, and solve problems.
[0274] Here are some key aspects of mechanistic interpretability: ● Reverse engineering: Unlike other interpretability techniques that explain what features influence a model's decision, mechanistic interpretability delves deeper, attempting to reverse engineer the model's internal processes to understand how those features are combined and manipulated to reach a conclusion. ● Focus on algorithms: It aims to identify the underlying algorithms or mechanisms within the model that contribute to its decision-making, making it more akin to understanding the "why" behind the "what". ● Challenges: Mechanistic interpretability is an active area of research due to the complex nature of neural networks and the difficulty in deciphering their internal workings. Current methods involve techniques like: ● Circuit discovery: identifying and analyzing sub-networks within the larger model that perform specific functions.● Feature attribution: understanding how individual input features contribute to the final output. ● Counterfactual reasoning: exploring what changes to the input would lead to different outputs, helping to understand the model's decision boundaries.
[0275] Benefits of mechanistic interpretability: ● Improved trust and transparency: By understanding how models work, users can build trust in their decisions and identify potential biases or errors. ● Debugging and improvement: Unveiling the internal workings can help diagnose issues within the model and pave the way for improvements. ● Safer AI development: Understanding the decision-making process of AI systems is crucial for building safe and reliable AI applications, especially in high-stakes domains.
[0276] While challenging, mechanistic interpretability is a crucial step towards achieving greater explainability and control over the complex world of machine learning models
[0277] AI Software System and Stack for Generative AI Large Language Models
[0278] Generative AI, like the kind behind large language models (LLMs), relies on a complex software system and stack. This stack combines various tools and technologies to train and run these powerful AI systems. Here's a breakdown of the key components: 1. Foundational Technologies: ● Programming Languages: Python is prevalent, with its readability, ease of use, and extensive libraries like “TensorFlow” and “PyTorch” that are useful for machine learning tasks. ● Cloud Infrastructure: Training LLMs requires serious computing power. Cloud platforms like “Google Cloud Platform (GCP)”, “Amazon Web Services (AWS)”, and “Microsoft Azure” offer virtual machines, storage, and specialized AI services. 2. Data Management: ● Data Processing Tools: Generative models are data-driven. Tools like Apache Spark and Hadoop are crucial for handling the massive datasets used to train LLMs. These tools clean, pre-process, and transform data to prepare it for training. 3. Core AI Components: ● Generative Models: At the heart are the generative models themselves. Generative Adversarial Networks (GANs) and Transformers are popular choices for LLMs.4. Additional Tools: ● Evaluation Tools: You need to measure how well your LLM performs. Evaluation metrics and tools help assess the quality and effectiveness of the generated text. ● Deployment Tools: Once your LLM is trained, it needs to be deployed and integrated into applications. Tools help with this process.
[0279] Overall Stack Structure: Many consider the generative AI tech stack to have 3 or 4 layers: ● Applications Layer: This is where the LLM interacts with users through applications or interfaces. ● Model Layer: This layer houses the core LLM, including the chosen generative model and its training data. ● Infrastructure Layer (sometimes included in the Model Layer): This layer provides the computational muscle for training and running the LLM, often leveraging cloud platforms.
[0280] Considerations for Generative AI LLMs: ● Bias and Fairness: Ensure the LLM is trained on diverse data and mitigates potential biases in its outputs. ● Explainability and Interpretability: Understand how the LLM generates its outputs and provide explanations to users. ● Safety and Security: Implement safeguards to prevent misuse and harmful outputs.
[0281] A large language model (LLM): is a type of artificial intelligence (AI) trained on massive amounts of text data. This allows them to understand and generate human-like text in response to a wide range of prompts and questions. Here are some key characteristics of LLMs:
[0282] Capabilities: ● Generate text: LLMs can create different creative text formats, like poems, code, scripts, musical pieces, email, letters, etc. They can also answer your questions in an informative way, even if they are open ended, challenging, or strange. ● Translate languages: LLMs can translate between multiple languages, preserving the meaning and context of the original text. ● Write different kinds of creative content: They can generate different creative text formats, like poems, code, scripts, musical pieces, email, letters, etc.● Answer your questions in an informative way: They can answer your questions in an informative way, even if they are open ended, challenging, or strange.
[0283] How they work: ● Training: LLMs are trained on massive datasets of text and code, allowing them to learn statistical relationships between words and phrases. This training process can take weeks or even months on powerful computing systems. ● Neural networks: They use complex neural networks, similar to the human brain, to process and understand information. These networks consist of interconnected nodes that learn and adapt over time. ● Probabilistic approach: LLMs use a probabilistic approach to generate text, meaning they predict the most likely word or phrase to come next based on the context of the previous words.
[0284] Applications: ● Chatbots and virtual assistants: LLMs can be used to power chatbots and virtual assistants that can engage in natural conversations with users. ● Machine translation: LLMs are increasingly being used for machine translation, offering more accurate and nuanced translations than traditional methods. ● Content creation: LLMs can be used to generate different creative text formats, like poems, code, scripts, musical pieces, email, letters, etc. ● Summarization and question answering: LLMs can be used to summarize text and answer questions in a comprehensive and informative way.
[0285] Limitations: ● Bias: LLMs can reflect the biases present in the data they are trained on. It's important to be aware of these biases and use LLMs responsibly. ● Factual accuracy: LLMs can sometimes generate text that is factually incorrect or misleading. It's important to verify information generated by LLMs with other sources. ● Lack of common sense: LLMs may not have the same level of common sense as humans, which can lead to nonsensical of incorrect outputs, also known as “hallucination”.
[0286] Generative AI and Large Language Models (LLMs) are closely related concepts, but they represent different aspects of the field:
[0287] Generative AI:● Broader category: Generative AI encompasses a wide range of artificial intelligence techniques focused on creating new and original content. This content can extend beyond just text, and can include images, music, code, and even synthetic data. ● Diverse approaches: Generative AI utilizes various methods to achieve its goal of content creation. These approaches can range from deep learning models to evolutionary algorithms, each with its own strengths and weaknesses. ● Specific type of generative AI: LLMs are a specific type of generative AI that primarily focuses on generating text. They are trained on massive datasets of text and code, allowing them to learn complex language patterns and statistical relationships. ● Text generation capabilities: LLMs excel at tasks like writing different kinds of creative text formats, including poems, code, scripts, musical pieces, emails, and letters. They can also be used for machine translation, text summarization, and question answering.
[0288] Relationship between Generative AI and LLMs: ● LLMs as building blocks: LLMs often serve as building blocks for more complex generative AI systems. Their ability to understand and generate natural language makes them valuable tools for tasks like creating chatbots, generating dialogue for virtual assistants, and personalizing content. ● LLMs not the only approach: While LLMs are powerful tools for text generation, they are not the only approach within generative AI. Other methods, like Generative Adversarial Networks (GANs), can be used to generate different types of content, such as images or music.
[0289] Here's an analogy: Think of generative AI as a broad art studio, filled with various tools and techniques for creating different forms of art. LLMs would be a specific set of tools within that studio, specializing in generating text-based art pieces. While the studio offers a wider range of creative possibilities, the LLM tools stand out for their unique capabilities in text generation.
[0290] Generative AI, also known as generative artificial intelligence, refers to a branch of AI that focuses on creating new content, like: ● Text: Generating poems, scripts, news articles, musical pieces, code, emails, or other textual formats. ● Images: Creating realistic or artistic images, such as photographs, paintings, or other visual content.● Audio and video: Generating music, sound effects, or even videos. ● 3D models: Creating 3D models of objects or environments.
[0291] Wireless digital 5G radio access network (RAN) antenna: is a specialized type of antenna that is used to transmit and receive radio signals for 5G cellular networks.5G is the latest generation of cellular network technology, and it offers significantly faster data speeds, lower latency, and improved capacity compared to previous generations.
[0292] 5G RAN antennas are typically much more complex than traditional cellular antennas. They often use multiple antennas and beamforming technology to improve signal quality and capacity. Beamforming allows the antenna to focus its radio waves in a particular direction, which can improve signal strength and reduce interference.
[0293] A Radio Access Network (RAN) platform system: for the internal unlicensed network NK-U standard of 3GPP refers to a specific type of equipment used in private cellular networks. Here's a breakdown of the key terms:
[0294] RAN platform: This refers to the hardware and software infrastructure that connects user devices (like phones or laptops) to the core network of a cellular system. It includes the base stations, antennas, and other equipment responsible for transmitting and receiving radio signals.
[0295] Internal unlicensed network: This is a private cellular network that operates within a specific location, such as a factory, campus, or office building. It uses unlicensed spectrum, which means it doesn't require a license from the government to operate.
[0296] NK-U standard: This is a technical specification developed by 3GPP (3rd Generation Partnership Project) specifically for internal unlicensed networks. It defines the protocols and technologies used for these networks, including how devices connect, how data is transmitted, and how security is ensured.
[0297] 3GPP: This is a global organization that develops technical specifications for cellular networks. Its specifications are used by all major mobile network operators and device manufacturers around the world.
[0298] So, putting it all together, an RAN platform system for the internal unlicensed network NK-U standard of 3GPP is a specialized piece of equipment that allows you to create your own private cellular network within a specific location. This network uses unlicensed spectrum and follows the 3GPP NK-U standard, which ensures compatibility with devices and other equipment that support this standard.
[0299] Cooling Systems for an AI Data Center An AGI data center presents unique challenges due to the immense heat generated by the vast computational power required for these complex systems. Here are some potential cooling systems for an AGI data center:
[0300] Traditional air-based cooling including Raised-floor cooling, etc.
[0301] Air-to-air heat exchangers (AAHX): Liquid-based cooling: Direct-to-chip liquid cooling, and Immersion cooling:
[0302] Emerging technologies: • Two-phase cooling: This system uses the latent heat of vaporization of a refrigerant to remove heat, offering high efficiency and potential for waste heat recovery. • Quantum computing-inspired cooling: This theoretical approach uses quantum principles to manipulate heat at the atomic level, potentially leading to highly efficient and ultra- compact cooling systems. However, it is far from practical implementation.
[0303] Power Considerations for AGI data centers: While there's no single "typical" power system for all AI data centers, there are several common strategies employed to address the unique demands of these facilities, including (i) High Efficiency: Power Factor Correction (PFC), High- efficiency transformers, and Variable-frequency drives (VFDs), and (ii) Scalability and Reliability: Modular power systems, Uninterruptible power supply (UPS) systems, etc.
[0304] A unidirectional fiber-optic cable: also known as simplex fiber, is a type of fiber-optic cable that allows data to travel in only one direction. This is different from standard fiber-optic cables, which are duplex cables and allow for two-way communication.
[0305] An Artificial Intelligence Protocol (AIP): as a data layer standard as defined in the invention on top of Transmission Control Protocol / Internet Protocol (TCP / IP) which is the standard internet data protocol, to enable AI data packet control via smart contract circuit breaker, for example, operating on distributed ledger technology (DLT).
[0306] TCP / IP / AIP Open Standard for safe and beneficial AI with Artificial Intelligence Protocol (AIP) layer on top of open standard Transmission Control Protocol (TCP) and Internet Protocol (IP) which are administered by Internet Engineering Task Force (IETF).
[0307] TCP / IP / AIP is a high-level specification of open source standard Artificial Intelligence Protocol (AIP) layer on top of TCP / IP.
[0308] TCP / IP / AIP is a metadata specification for AI Application Safety, Security and Regulatory Compliance 1. USER Application 2. Create ID for the data (“PROMPT”) 3. Tag PROMPT 4. Send Out PROMPT 5. Handle PROMPT 6. Process PROMPT 7. Wrap data from LLM (“RESPONSE”) 8. Send Back RESPONSE to USER 9. USER Handle RESPONSE
[0309] Detailed specification of Artificial Intelligence Protocol (AIP) layer on top of TCP / IP.
[0310] Metadata for AI Application Safety, Security and Regulatory Compliance via unidirectional data via nine (9) FIREWALLS IN and nine (9) FIREWALLS OUT with eighteen (18) DENY kill-switches on each FIREWALL IN / OUT to stop transmission of unsafe data, with immutable (web3) recording of log data.
[0311] USER APP Create ID in USER Data Terminal for the data submission (“PROMPT”) 1. USER APP Tag PROMPT 2. USER Send encrypted (HTTPS) PROMPT to AGI Subsystem; GATEWAY 3. Record OUT-1 PROMPT OUT of FIREWALL-OUT-1 of USER DATA TERMINAL 4. Pass OUT-1 PROMPT via FIREWALL-OUT-1 (or DENY-1) of USER DATA TERMINAL to GATEWAY 5. Record IN-1 PROMPT IN to FIREWALL-IN-1 of AGI Subsystem; GATEWAY from USER Terminal 6. Pass IN-1 PROMPT via FIREWALL-IN-1 (or DENY-2) to AGI Subsystem; GATEWAY 7. Transmit PROMPT INTO AGI Subsystem; GATEWAY 8. Handle PROMPT in AGI Subsystem; GATEWAY 9. Process PROMPT in AGI Subsystem; GATEWAY 10. Send PROMPT to AGI Subsystem; INTERPRETERRecord OUT-2 PROMPT OUT of FIREWALL-OUT-2 of AGI Subsystem; GATEWAY Pass OUT-2 PROMPT via FIREWALL-OUT-2 (or DENY-3) of AGI Subsystem; GATEWAY to INTERPRETER Record IN-2 PROMPT to FIREWALL-IN-2 of AGI Subsystem; INTERPRETER Pass IN-2 PROMPT via FIREWALL-IN-2 (or DENY-4) to AGI Subsystem; INTERPRETER Transmit PROMPT INTO AGI Subsystem; INTERPRETER Handle PROMPT in AGI Subsystem; INTERPRETER Process PROMPT in AGI Subsystem; INTERPRETER Send PROMPT to AGI Subsystem; SAFETY Record OUT-3 PROMPT OUT of FIREWALL-OUT-3 of AGI Subsystem; INTERPRETER Pass OUT-3 PROMPT via FIREWALL-OUT-3 (or DENY-5) of AGI Subsystem; INTERPRETER to SAFETY Record IN-3 PROMPT to FIREWALL-IN-3 of AGI Subsystem; SAFETY Pass IN-3 PROMPT via FIREWALL-IN-3 (or DENY-6) to AGI Subsystem; SAFETY Transmit PROMPT INTO AGI Subsystem; SAFETY Handle PROMPT in AGI Subsystem; SAFETY Process PROMPT in AGI Subsystem; SAFETY Send PROMPT to AGI Subsystem; AGI SYSTEM Record OUT-4 PROMPT OUT of FIREWALL-OUT-4 of AGI Subsystem; SAFETY Pass OUT-4 PROMPT via FIREWALL-OUT-4 (or DENY-7) of AGI Subsystem; SAFETY to AGI SYSTEM Record IN-4 PROMPT to FIREWALL-IN-4 of AGI System; AGI SYSTEM Pass IN-4 PROMPT via FIREWALL-IN-4 (or DENY-8) to AGI System; AGI SYSTEM Transmit PROMPT INTO AGI System; AGI SYSTEM Handle PROMPT in AGI System; AGI SYSTEM Process PROMPT in AGI System; AGI SYSTEM Wrap data from LLM (“RESPONSE”) in AGI System; AGI SYSTEMSend RESPONSE to AGI Subsystem; SAFETY Record OUT-5 RESPONSE OUT of FIREWALL-OUT-5 of AGI System; AGI SYSTEM Pass OUT-5 RESPONSE via FIREWALL-OUT-5 (or DENY-9) AGI System; AGI SYSTEM to SAFETY Record IN-5 RESPONSE to FIREWALL-IN-5 to AGI Subsystem; SAFETY Pass IN-5 RESPONSE via FIREWALL-IN-5 (or DENY-10) to AGI Subsystem; SAFETY Transmit RESPONSE INTO AGI Subsystem; SAFETY Handle RESPONSE in AGI Subsystem; SAFETY Process RESPONSE in AGI Subsystem; SAFETY Send RESPONSE to AGI Subsystem; INTERPRETER Record OUT-6 RESPONSE OUT of FIREWALL-OUT-6 of AGI System; SAFETY Pass OUT-6 RESPONSE via FIREWALL-OUT-6 (or DENY-11) AGI System; SAFETY to INTERPRETER Record IN-6 RESPONSE to FIREWALL-IN-6 to AGI Subsystem; INTERPRETER Pass IN-6 RESPONSE via FIREWALL-IN-6 (or DENY-12) to AGI Subsystem; INTERPRETER Transmit RESPONSE INTO AGI Subsystem; INTERPRETER Handle RESPONSE in AGI Subsystem; INTERPRETER Process RESPONSE in AGI Subsystem; INTERPRETER Send RESPONSE to AGI Subsystem; GATEWAY Record OUT-7 RESPONSE OUT of FIREWALL-OUT-7 of AGI System; INTERPRETER Pass OUT-7 RESPONSE via FIREWALL-OUT-7 (or DENY-13) AGI System; INTERPRETER to GATEWAY Record IN-7 RESPONSE to FIREWALL-IN-7 to AGI Subsystem; GATEWAY Pass IN-7 RESPONSE via FIREWALL-IN-7 (or DENY-14) to AGI Subsystem; GATEWAY Transmit RESPONSE INTO AGI Subsystem; GATEWAY Handle RESPONSE in AGI Subsystem; GATEWAY58. Process RESPONSE in AGI Subsystem; GATEWAY 59. Send RESPONSE to AGI Subsystem; STORAGE 60. Record OUT-8 RESPONSE OUT of FIREWALL-OUT-8 of AGI System; GATEWAY 61. Pass OUT-8 RESPONSE via FIREWALL-OUT-8 (or DENY-15) AGI System; GATEWAY to STORAGE 62. Record IN-8 RESPONSE to FIREWALL-IN-8 to AGI Subsystem; STORAGE 63. Pass IN-8 RESPONSE via FIREWALL-IN-8 (or DENY-16) to AGI Subsystem; STORAGE 64. Transmit COPY OF RESPONSE INTO to AGI Subsystem; STORAGE 65. Handle Copy of RESPONSE in AGI Subsystem; STORAGE 66. Process Copy of RESPONSE in AGI Subsystem; STORAGE 67. Record OUT-9 RESPONSE OUT of FIREWALL-OUT-9 of AGI System; GATEWAY 68. Pass OUT-9 RESPONSE via FIREWALL-OUT-9 (or DENY-17) AGI System; GATEWAY to USER Terminal 69. Record IN-9 RESPONSE IN to FIREWALL-IN-9 of AGI Subsystem; GATEWAY to USER Terminal 70. Pass IN-9 RESPONSE via FIREWALL-IN-9 (or DENY-18) to AGI Subsystem; GATEWAY 71. Send encrypted (HTTPS) RESPONSE from AGI Subsystem; GATEWAY to USER APP 72. USER APP Handle RESPONSE for USER viewing in Data Terminal
[0312] TCP / IP / AIP Certificate Management (like an SSL certificate) by TCP / IP / AIP Foundation included for Regulatory Compliant AI Data Center Customers / Users, as an open standard, and open server standard for regulatory compliant and authorized AI service providers by regulators in jurisdictions of operations: United States (US), United Kingdom (UK), Australia (AU), Canada (CA), France (FR), Germany (DE), Ireland (IR), Italy (IT), Japan (JP), Saudi Arabia (SA), Singapore (SG), Switzerland (CH), and European Union (EU).
[0313] Collective AI Commons for Safe; Human aligned AI with TCP / IP / AIP open standard for Artificial Intelligence Protocol for AI Safety and Human Alignment on top of TCP / IP, via gateway for security, interpretability and regulatory compliance.
[0314] Artificial Intelligence Protocol (AIP) layer on top of TCP / IP includes metadata for AI Application Safety, Security and Regulatory Compliance via unidirectional data via nine (9) FIREWALL IN and nine (9) FIREWALL OUT with eighteen (18) DENY kill-switches on each FIREWALL IN / OUT to stop transmission of unsafe data, with immutable (web3) recording of log data.
[0315] Web3 is a proposed evolution of the World Wide Web that emphasizes decentralization, blockchain technology, and token-based economics. It aims to create a more open, user-owned, and transparent internet experience compared to the current Web 2.0, which is dominated by centralized platforms and corporations.
[0316] In Web3, recording of log data is crucial for monitoring and analyzing the activity and state of a blockchain network and decentralized applications (dApps) built on it. These logs provide immutable and transparent records of events that occur within the system, allowing for various purposes such as: 1. Debugging and Troubleshooting: Developers can use logs to identify and diagnose issues within their dApps by analyzing the sequence of events and pinpointing where unexpected behavior might be occurring. 2. Security Monitoring: Logs can be used to monitor for suspicious activity on the network, such as unauthorized transactions or attempts to breach smart contracts. By analyzing patterns and deviations from normal behavior, potential security threats can be identified and addressed. 3. User Activity Tracking: Logs can track user interactions with dApps, providing valuable insights into user behavior and preferences. This information can be used to improve the user experience and inform future development decisions. 4. Regulatory Compliance: In some cases, regulations might require dApps to maintain audit trails of transactions and activities. Logs can help fulfil these compliance requirements by providing a verifiable record of events.
[0317] Here's a breakdown of the mechanics of recording log data in Web3: • Events: Smart contracts can emit events containing relevant information about specific actions that occur within the contract. These events are then recorded on the blockchain. • Logs: The information captured in events is compiled into logs. These logs typically include details like:• Timestamp: When the event occurred. • Contract address: The smart contract where the event originated. • Event name: Identifying the specific type of event that took place. • Data: Additional information relevant to the event, such as transaction details or user addresses. • Storage: Logs are permanently stored on the blockchain, ensuring their immutability and transparency. However, storing large amounts of data on the blockchain can be expensive, so developers need to carefully consider which information is essential to log.
[0318] Accessing and Analyzing Logs: • Log viewers: Various tools and services are available to view and analyze blockchain logs. These tools allow developers and users to filter, search, and interpret log data to gain insights into the system's behavior. • Decentralized infrastructure: Some projects within the Web3 space are exploring decentralized storage solutions for logs, aiming to offer a more cost- effective and scalable way to manage large datasets while maintaining the benefits of immutability and transparency.
[0319] By effectively utilizing log data, developers and users can gain valuable insights into the operation of decentralized applications (dApps) and the overall health of the Web3 ecosystem.
[0320] A data center: is essentially a warehouse for computer systems and related equipment. Here's a breakdown of its key features: • Physical facility: At its core, a data center is a building or a designated space within a building that houses critical IT infrastructure. This infrastructure includes: • Servers: Computers that store and process data. • Storage systems: Devices used to hold vast amounts of data. • Networking equipment: Components that enable communication and data transfer between various parts of the data center and the outside world. • Security: Since data centers store sensitive information, security is a top priority. They typically have robust security measures in place, including: • Access control systems to restrict entry. • Security cameras for surveillance. • Firewalls and other tools to protect against cyberattacks.• Environment control: Data center equipment generates a lot of heat. To ensure proper functioning, data centers have sophisticated climate control systems to maintain a cool and stable environment. • Redundancy: Data centers are designed with redundancy in mind. This means they have backup systems for critical components, such as power supplies and cooling systems, to minimize downtime in case of any failures.
[0321] A blade server is a type of server computer that is designed to be highly compact and efficient. It is essentially a stripped-down server that only has the essential components, such as processors, memory, and I / O controllers. These components are housed in a thin, modular blade that slides into a special enclosure called a chassis.
[0322] Blade servers can be a good fit for AI data centers due to several reasons including High density, reduced power consumption and improved manageability.
[0323] The Qualcomm Cloud AI 100 PCIe unit is an example of an AI accelerator card designed to be installed in a server via a PCIe slot. It acts as a dedicated processing unit for specific tasks related to artificial intelligence, particularly inference, which is the process of applying a trained AI model to new data. Overall, the 8 Qualcomm Cloud AI PCIe unit is a powerful tool for accelerating AI inference workloads in data centers. It offers high performance, good power efficiency, and flexibility for various applications.
[0324] The Qualcomm Cloud AI 100 HHHL-Pro PCIe with 400 TOPS: is an example of an AI accelerator card designed for high-performance, power-efficient AI inference in data centers. It's a specific version of the broader Qualcomm Cloud AI 100 family. Applications including: Image recognition and classification for analyzing images and videos for content, object detection, etc., and / or Natural language processing for understanding and generating human language for tasks like chatbots, machine translation, etc., and / or Recommendation systems for Recommending products, content, etc. based on user preferences and data, and / or Fraud detection and anomaly analysis: Identifying suspicious activity in financial transactions, network traffic, etc.
[0325] Preferred embodiment and alternatives: • “Nvidia” Graphical processing units (GPU) chips are the leading AI data center technology delivering high raw performance at the cost of higher power consumption.• Qualcomm, our preferred embodiment, claims the Cloud AI 100 offers comparable performance-per-watt to competing solutions similar to Nvidia GPUs and Intel GPUs.
[0326] The Snapdragon X555G Modem-RF System is a comprehensive solution designed to enable the development of 5G multimode devices. It's essentially all the hardware and software needed for a device to connect to 5G networks.
[0327] An AI system on a chip (AI SoC): is a specialized type of computer chip designed to efficiently handle specific tasks related to artificial intelligence (AI). Unlike a traditional CPU, which handles various general-purpose computations, an AI SoC is tailored for tasks like: • Inference: Applying a trained AI model to new data to make predictions or classifications. • Training: Building and improving AI models by processing and analyzing large amounts of data. • Data Centers: Servers for cloud-based AI services, training large AI models, and running AI applications in various industries. • Mobile and Edge Devices: Smartphones, smart speakers, drones, self-driving cars, wearable devices, etc., where AI processing needs to be done on-device for real-time or privacy reasons.
[0328] Static Random-Access Memory is a type of memory used in electronic devices. Stores data in individual latches (circuits that can hold one bit of information, either 0 or 1). Volatile: data is lost when the power is turned off, unlike flash memory which retains information even without power. Faster access times compared to DRAM (Dynamic Random- Access Memory). More expensive and less dense than DRAM, making it unsuitable for main memory in most devices.
[0329] While DRAM isn't the main memory used for complex AI computations, it does play a crucial role in various AI applications. Here's a breakdown of DRAM and its role in AI: • DRAM (Dynamic Random-Access Memory): • Definition: A type of volatile memory that stores data as electrical charges on capacitors. It requires periodic refreshing to maintain data integrity. • Advantages include: (i) Cost-effective: Much cheaper than SRAM, allowing for large- scale storage. (ii) High density: Can store more data in a smaller space compared toSRAM. Disadvantages include: (i) Slower access times: Requires refreshing, leading to higher latency compared to SRAM, and (ii)vVolatility: Data is lost if power is off.
[0330] DRAM in AI: • Suitable for storing large datasets: AI models often require access to vast amounts of training data. DRAM's high density and cost-effectiveness make it ideal for storing such data before processing or feeding it to the AI model. • Used for caching intermediate results: Complex AI models might generate and utilize intermediate results during calculations. DRAM can temporarily store these results for faster access within the same processing cycle, improving overall performance. • Applications: • Training models on large datasets in cloud-based AI systems. • Storing pre-processed data for faster access by AI models. • Caching intermediate results for improved performance in complex calculations.
[0331] The Artificial Intelligence Protocol (AIP) embodiment of the invention:
[0332] The Artificial Intelligence Protocol (AIP) as a layer on top of TCP / IP with control via tripwire Smart Contract Circuit Breaker operating on Distributed Ledger Technology (DLT)
[0333] Firewall hardware: as used in this invention is a dedicated physical and software enabled device deployed between the AGI system and subsytems, and the data center access to the internet for user access. More specifically it is a unidirectional firewall that sits between the main system and the safety subsystem and between all of the subsystems to monitor data in and out of the system. It acts as a security barrier, monitoring incoming and outgoing traffic. It acts like a gatekeeper, deciding what data is allowed to enter or leave the system based on a set of security rules set by the Oracle.
[0334] The Firewall of this invention has a variety of software incorporated into it. Including, AIP layered on TCP / IP; Blockchain; Smart Contracts; Smart Contract Tracking System; state machine; tripwire Smart Contract Circuit Breaker; and an Oracle;.
[0335] Unidirectional firewall embodiment of the invention:
[0336] This invention uses unidirectional firewalls that only allow data to flow in one direction. This prevents bad data from traveling in or out of the main system or a subsystem. By preventing two-way communication, unidirectional firewalls significantly reduce the attack surface and make it more difficult for attackers to gain access to critical systems and data. Unlike traditionalfirewalls that allow two-way communication, unidirectional firewalls cannot be used for applications that require two-way communication, such as file sharing or remote access.
[0337] A common use case for unidirectional firewalls is segmenting of networks. Unidirectional firewalls can be used to segment networks, which can help to contain the spread of malware and other security threats.
[0338] Smart Contract Circuit Breaker embodiment of the invention:
[0339] A Smart Contract Circuit Breaker is an autonomous software state machine safety mechanism built as a smart contract technology, specifically operating on a Distributed Ledger Technlogy (DLT) network. It functions as an autonomous machine ON-OFF Circuit Breaker, designed, configured and deployed to protect users from potential harm in the event of unforeseen issues or vulnerabilities, and / or non-compliant AI behavior.
[0340] In the present invention a smart contract circuit breaker is adapted to each firewall to interrogate each data packet to determine whether to let the data packet pass or not pass.
[0341] Here's how it works: • Monitoring: The circuit breaker continuously monitors specific metrics of the smart contract, such as rules provided on an Oracle by the operator. • Threshold: Predefined limits are set for these metrics. If the metric is outside a RULE, the circuit breaker is triggered. • Action: Upon triggering, the Smart Contract Circuit Breaker halts the specific non- compliant (UnSafe) data packet transmission.
[0342] The function is to prevent harms of the AI system, and to flag the non-compliant behavior to be identified and addressed according to procedure and policy of the Regulator.
[0343] Here are some key benefits of using Smart Contract Circuit Breaker: • Enhanced security: Mitigates potential losses from exploits or flash loan attacks. • Increased stability: Prevents cascading effects of abnormal events on the protocol's functionality. • Improved user protection: Offers safeguards to prevent users from making detrimental decisions during periods of high volatility.
[0344] Integrating Distributed Ledger Technology (“blockchain”) directly with the firewall and the Smart Contract Circuit Breaker: Firewalls operate at the packet level, delivering real- time operations based on pre-defined RULES, while a blockchain is designed and deployed forsecure, transparent data recording across a distributed network of operating nodes, all with the same ledger recording system and records. However, it is an aspect of this invention to leverage blockchain principles alongside firewalls, achieving specific security benefits: 1. Blockchain-based Access Control Lists (ACLs): • Traditional firewalls rely on static ACLs, which can be cumbersome to manage and prone to human error. • A blockchain-based system could store and manage ACLs on a distributed ledger, offering: • Smart Contract Functionality: The Smart Contract Circuit Breaker • Immutability: Changes to ACLs are recorded on the blockchain, preventing unauthorized modifications. • Transparency: All authorized users can access and audit the ACLs on the blockchain, fostering trust and accountability. • Decentralization: Eliminates the risk of a single point of failure, as the ACLs are not stored on any single device. 2. Collaborative Threat Intelligence: • Firewalls can benefit from shared threat intelligence to identify and block malicious traffic more effectively. • A blockchain-based platform could facilitate the secure and anonymous sharing of threat intelligence between organizations: • Security: Blockchain's cryptography ensures only authorized entities can access the threat intelligence data. • Anonymity: Participating organizations can share data without revealing their identities, encouraging wider participation. 3. Blockchain Application Firewalls (BAFs): • These are specialized firewalls designed specifically for blockchain applications. • BAFs operate at the application layer, inspecting and filtering transactions before they are submitted to the blockchain: • Increased security: BAFs can prevent unauthorized transactions, enforce access controls, and filter out invalid or malicious transactions before they impact the blockchain network.• Improved compliance: BAFs can be configured to enforce specific compliance rules on blockchain transactions.
[0345] Blockchain embodiment of the invention:
[0346] The Blockchain is a type of Distributed ledger technology (DLT). Blockchains transactions are bundled into blocks, chained chronologically. DLT technology enables: (i) Security: Data is encrypted and tamper-proof, making it highly resistant to cyberattacks and fraud, and (ii) Transparency: All participants have access to the same data, fostering trust and accountability, and (iii) Efficiency: Eliminates the need for intermediaries, streamlining processes and reducing costs, and (iv) Decentralization: No single point of failure, as the data is distributed across the network.
[0347] While a method using blockchain is a preferred embodiment, Directed Acyclic Graph (DAG) may someday be a viable alternative to blockchains where transactions are organized in a directed acyclic graph, potentially offering faster transaction processing.
[0348] Comparison with Blockchains:
[0349] While both are used in DLT, DAGs differ from blockchains in structure and processing. DAG transactions are directly linked to each other, forming a directed acyclic graph, potentially offering faster processing and improved scalability.
[0350] AI Interpreter Subsystem embodiment of the invention:
[0351] The embodiment of AI Interpreter Subsystem enables the processing of the unidirectional data output from the Biothrivalistic Mechanistic Interpretability Firewall system, thereby enabling psychometric behavoral measurement, analysis, understanding and regulatory compliance of the AI with the Biotrivalistic RULES. In this invention, according to Figure 5, an Interpreter Subsystem is configured in the same way as the AI Safety Subsystem (5-17) and is connected between the AI Safety Subsystem (5-17) and a Gateway Subsystem (5-19) by way of unidirectional firewalls and unidirectional fiber optic cables in the same way as the AI Safety Subsystem is connected to the AGI System.
[0352] AI Gateway Subsystem embodiment of the invention:
[0353] The embodiment of AI Gateway Subsystem mediates communication between External systems (e.g., user interfaces, applications) and the Different AI subsystems within the data center. A specialized component within the AI system the Gateway acts as the entry point forExternal communication, receiving user requests or data from other systems, and Internal communication, routing information or tasks to different AI subsystems based on their expertise.
[0354] Here's a summary of embodiment of functionalities:
[0355] The Gateway Subsystem may perform additional functionalities such as: (i) Data filtering or pre-processing before passing it on to other sub-systems, and (ii) Managing security and access control for external communication, and (iii) monitoring communication traffic and ensuring safe and regulatory complaint operations and AGI behavior.
[0356] AI Training Data Storage Subsystem embodiment of the invention:
[0357] The embodiment of AI Training Data Storage Subsystem (20) is a specialized component within a larger AI system that focuses on managing the storage and access of data used to train AI models. This data, often massive in volume, is crucial for the development and performance of AI, as it's what the models learn from to identify patterns, make predictions, and improve over time.
[0358] Summary of embodiment of functionalities and key aspects of the Traingin Data Storage Subsystem: • AI Training Data: This refers to the vast amount of information, such as text, images, or code, used to train AI models. The quality and quantity of this data significantly impact the model's performance, safety, behavior and regulatory complaint Biothrivalism. • Storage: The subsystem provides the necessary storage capacity and infrastructure to handle the large datasets efficiently. This can involve various technologies like traditional hard drives, solid-state drives, or cloud storage solutions, depending on the specific needs. • Management: The subsystem also encompasses functionalities for managing the data lifecycle. This includes tasks like data ingestion, organization, access control, and potentially anonymization or pre-processing to ensure the data is suitable for training.
[0359] In essence, an AI Training Data Storage Subsystem acts as the backbone for providing the necessary data foundation for training AI models. It ensures the data is readily available, secure, and properly managed throughout the model training process in the AGI system.
[0360] AI Administration and Support Subsystem embodiment of the invention:
[0361] The embodiment of AI Administration and Support Subsystem is a broad term encompassing various functionalities related to the overall management and support of AI systems. It's essentially the "behind-the-scenes" component that keeps the AI system running smoothly.
[0362] Summary of embodiment of functionalities and key aspects of Administration and Support Subsystem: • Model Deployment and Management: This involves deploying trained AI models into production environments, monitoring their performance, and managing their versions and updates. • Monitoring and Logging: The subsystem gathers data on the AI system's performance, resource usage, and potential errors. This information is crucial for troubleshooting issues, optimizing performance, and ensuring the system functions as expected. • Security and Explainability: This aspect focuses on securing the AI system against potential threats like data breaches or manipulation. Additionally, it might involve tools and techniques to explain the AI's decision-making process, providing transparency and addressing potential violation of the Biotrivalistic RULES. • User Management and Access Control: The subsystem manages user access to the AI system, defining different permission levels and ensuring only authorized users can interact with specific functionalities. • Integration with other Systems: This involves integrating the AI system with existing infrastructure and tools within the organization, allowing seamless data flow and collaboration.
[0363] AI Regulatory Compliance Subsystem embodiment of the invention:
[0364] The embodiment of the AI Regulatory Compliance Subsystem is a specialized component within the AGI system that aims to ensure that the development and deployment of AI systems adheres to relevant regulations and ethical principles. As AI technology rapidly evolves, concerns about potential risks like bias, discrimination, and privacy breaches have prompted various regulatory bodies to develop frameworks and laws. This subsystem plays a crucial role in navigating this evolving landscape.
[0365] Summary of embodiment of functionalities and key aspects of Regulatory Compliance Subsystem: 1. Regulatory Definition of the Biotrivalistic RULES and Monitoring and Enforcement of the Biotrivalistic RULES for AI• The subsystem enables the maintanace of Biotrivalistic RULES, regulations and ethical guidelines from various governing bodies and industry standards, for safe and trustworthy AI. • It subsystem assesses the AI behavior and data output identifies areas where BioThrivalistic compliance has been violated. 2. Risk Assessment and Mitigation: • The subsystem helps the Regulator to identify and analyze potential risks associated with the AI systems, such as bias, privacy violations, and violation of Biotrivalistic RULES. • Based on the identified risks or violations of the Biotrivalistic RULES, it suggests appropriate mitigation strategies, including data governance or shut down and retraining. 3. Documentation and Reporting of complaiance with Biotrivalistic RULES: • The subsystem facilitates the creation and maintenance of comprehensive documentation related to the development, deployment, and use of AI systems. • This documentation could include details about the training data, algorithms used, risk assessments conducted, and compliance measures implemented. • The subsystem may assist in generating reports required by cooperating regulatory bodies. 4. Training and Awareness: • The subsystem can support initiatives to educate employees involved in AI development and deployment about relevant regulations and ethical principles. • This might involve providing training materials, conducting workshops, and fostering an organizational culture of responsible AI development. 5. Integration with Other Systems: • The subsystem might integrate with existing compliance management systems within the organization, allowing for streamlined data collection and reporting related to AI compliance activities.
[0366] The AI Regulatory Compliance Subsystem specific functionalities and their implementation will vary depending on factors like: • The domicile and application of the AI system: Different industries and domiciles and applications may have varying regulatory requirements. • The size and complexity of the organization: Larger organizations might require more comprehensive compliance frameworks.• The evolving regulatory landscape: As regulations continue to develop, the subsystem needs to be adaptable to incorporate changes.
[0367] AI Research Subsystem embodiment of the invention:
[0368] The embodiment of AI Research Subsystem is a specialized component within the organization dedicated to exploring and advancing the field of AI. It serves as a hub for innovation and experimentation, pushing the boundaries of what AI can achieve.
[0369] Summary of embodiment of key functionalities the AI Research Subsystem : 1. Exploration of Emerging AI Techniques: • The subsystem focuses on staying at the forefront of AI research, exploring and experimenting with cutting-edge techniques like deep learning, natural language processing, and reinforcement learning. • This involves actively researching new algorithms, architectures, and applications of AI. 2. Development of Novel Solutions: • The subsystem acts as a breeding ground for innovation, where researchers can develop novel AI solutions to address complex challenges in various fields. • This could involve projects ranging from developing new medical diagnostic tools to creating advanced language translation systems. 3. Collaboration and Knowledge Sharing: • The subsystem fosters collaboration between researchers, both within the organization and with external experts involved in the broader AI community. • This can involve hosting workshops, conferences, or participating in collaborative research projects to share knowledge and accelerate progress in the field. 4. Prototype Development and Testing: • The subsystem provides resources and infrastructure for researchers to develop and test prototypes of new AI models and solutions. • This might involve access to powerful computing resources, specialized software tools, and dedicated data storage for experimentation. 5. Publication and Dissemination of Findings: • The subsystem encourages the publication of research findings in academic journals, conferences, and open-source platforms.• This helps contribute to the broader body of knowledge in AI and fosters collaboration within the research community.
[0370] AI Edge AI Subsystem embodiment of the invention:
[0371] The embodiment of an AI Edge AI Subsystem refers to a specific component within a larger AI data center focused on processing data and making decisions locally, "at the edge" of the network, rather than relying on centralized cloud computing. It leverages the principles of edge AI to bring AI capabilities closer to where the data originates.
[0372] Summary of embodiment of functionalities and key aspects of the AI Edge AI Subsystem: 1. Decentralized Processing: • Unlike traditional AI systems that rely solely on cloud processing, an Edge AI Subsystem performs computations directly on edge devices like sensors, cameras, or dedicated hardware units close to the data source. • This reduces reliance on cloud infrastructure, enabling real-time processing and faster decision-making with minimal latency. 2. Resource Optimization: • Edge AI Subsystems are often designed to be lightweight and efficient, operating on devices with limited processing power and storage capacity. • This allows for real-time processing on resource-constrained devices without overwhelming the network or incurring significant cloud computing costs. 3. Offline Functionality: • Edge AI Subsystems can operate even when internet connectivity is unavailable, as they process data locally on the device itself. • This is crucial for applications in remote locations or situations requiring uninterrupted operation, even in case of network disruptions. 4. Specific Applications: Edge AI Subsystems are particularly beneficial for applications where real-time decision-making and low latency are critical, such as: • Autonomous vehicles: Processing sensor data for real-time navigation and obstacle avoidance. • Industrial automation: Monitoring equipment health and predicting potential failures for preventative maintenance.• Smart home devices: Analyzing sensor data to optimize energy usage or automate tasks based on user presence. 5. Integration with Cloud: • While Edge AI Subsystems operate independently, they can also connect to cloud-based AI systems for tasks requiring more powerful computation or centralized storage of data and trained models. • This creates a hybrid approach, leveraging the strengths of both edge and cloud computing for optimal performance.
[0373] An Artificial Intelligence Edge AI Subsystem plays a vital role in enabling real-time, localized AI applications on resource-constrained devices. It offers benefits like faster processing, offline functionality, and efficient resource utilization, making it suitable for various applications demanding real-time decision-making at the edge of the network.
[0374] Detailed Description of the Figures
[0375] It should be understood that various embodiments of the disclosure and the terms used in the present document are not intended to limit the technological features set forth in the present document to particular embodiments and include various changes, equivalents, or replacements for a corresponding embodiment.
[0376] With regard to the description of the drawings, similar reference numerals may be used to refer to similar or related elements. A singular form of a noun corresponding to a similar item may include more than one of the things, unless the relevant context clearly indicates otherwise.
[0377] Figure 1. The New Science of BioThrivology
[0378] FIGURE LEGEND: 1-1. Homo sapiens (species) 1-2. BioThrivalistic INVENTION which is a novel method, system and apparatus for sustainable and enforced Mutual symbiosis between the species Homo sapiens and Machine intelligence. 1-3. Machine intelligence (which is AI becoming a NEW species. Ref prior art description, for example TED Talk by Mustafa Suleyman et al) 1-4. Home sapiens biothrivalist (NEW BioThrivalists] 1-5. Novel Outcome of Invention: BioThrivalism (Mutualistic Symbiosis) between Homo sapiens and Machine intelligence (a NEW State of Mutualistic symbiosis)1-6. Machine Intelligence biothrivalist (NEW BioThrivalists) 1-7. BioThrivology (a NEW Bioscience) 1-8. INTERACTION DATA of species Homo sapiens with the INVENTION 1-9. INTERACTION DATA of NEW species Machine intelligence with the INVENTION 1-10. ENABLEMENT of species Homo sapiens via the INVENTION to become EMPOWERED as Homo sapiens biothrivalist (NEW BioThrivalists) to SAFELY Co-EXIST with the NEW species as a Machine intelligence biothrivalist (NEW BioThrivalists) ENABLED to be CONTAINED by INVENTION 1-11. ENABLEMENT via the INVENTION of BioThrivalism (a NEW form of Mutualistic symbiosis) to SAFELY ENABLE Co-EXISTENCE of Homo sapiens biothrivalist (NEW BioThrivalists) and Machine intelligence biothrivalist (NEW BioThrivalists) 1-12. ENGAGEMENT of Machine intelligence (NEW species) CONTAINED by INVENTION to SAFELY Co-EXIST as Machine intelligence biothrivalist [NEW BioThrivalists] with Homo sapiens biothrivalist [NEW BioThrivalists] 1-13. INTERACTION DATA of species Homo sapiens biothrivalist [NEW BioThrivalists] with INVENTION to become EMPOWERED by NEW CONTAINED species Machine intelligence biothrivalist 1-14. INTERACTION DATA of NEW species Machine intelligence biothrivalist CONTAINED by INVENTION to SAFELY Co-EXIST as NEW species Machine intelligence biothrivalist (NEW BioThrivalists) with Homo sapiens biothrivalist (NEW BioThrivalists) 1-15. INTERACTION DATA of BioThrivology (NEW Bioscience) with Homo sapiens biothrivalist (NEW BioThrivalist) 1-16. REGULATORY COMPLIANT SCIENTIFIC DATA of BioThrivalism (NEW Mutualistic Symbiosis) with BioThrivology (NEW Bioscience) for benefit of Homo sapiens biothrivalist (NEW BioThrivalist) with SAFE Co-EXISTENCE of NEW species Machine intelligence biothrivalist (NEW BioThrivalist) as ENABLED to be CONTAINED by INVENTION 1-17. INTERACTION DATA of BioThrivology (NEW Bioscience) with Homo sapiens biothrivalist (NEW BioThrivalist)
[0379] Figure 1 shows first how the people and Machine intelligence (artificial intelligence) currently interact within a data center. Machine intelligence is not the invention. It is Machineintelligence with safeguards such as are described and claimed here. In the present invention human BioThrivalists and Machine Intelligence BioThrivalists interact in mutualistic symbiosis through the present invention. The study of BioThrivalism is called BioThrivology.
[0380] Detailed description of Figure 2
[0381] FIGURE LEGEND: 2-1. BioThrivalism; Regulatory COMPLIANCE 2-2. BioThrivalistic Human Message for Machine LAW 2-3. BioThrivalistic; Machine Right to Exist LAW 2-4. BioThrivalism; RULES 2-5. BioThrivalism; ORACLE which imposes the RULES to the TCP / IP / AIP Safe Data Circuit Breaker 2-6. TCP / IP / AIP Safe Data PROTOCOL for BioThrivalism 2-7. AIP Data-State Event Recording for BioThrivalism 2-8. END USER SAFE DATA INPUT for BioThrivalism 2-9. DATA INTO Subsystems of AI Data Center for BioThrivalism 2-10. Nine (9) Circuit Breakers DATA IN with TCP / IP / AIP Protocol for BioThrivalism 2-11. AGI DATA CENTER for BioThrivalism 2-12. Nine (9) Circuit Breakers DATA OUT with TCP / IP / AIP Protocol for BioThrivalism 2-13. DATA OUT Subsystems AI Data Center for BioThrivalism 2-14. AGI SAFE DATA OUTPUT for BioThrivalism 2-15. END USER TERMINAL DEVICE for Human and / or Personal AI Agent for Human for BioThrivalism 2-16. Personal AI Agent for Human for BioThrivalism 2-17. Human for BioThrivalism
[0382] Figure 2 shows the detailed steps of how BioThrivology is implemented. A regulatory compliance subsystem (2-1) inputs laws (2-2, 2-3). The human message for the machine and right to exist, into the system resulting in rules to be followed by the intelligent machine. (See below.) The ORACLE (2-5) enforces those rules. Behavior is monitored by the TCP / IP / AIP safe data protocol (2-6). End user data is input to the data center (2-8). Figure shows the process of data flow between the human and the Safe AI Machine intelligence, with the Regulator enforcing compliance of Safe AI behavior via the RULES which are managed by the ORACLE whichinstructs the circuit breaker machine to let Safe Data to pass (State-ON) and Un-Safe Dat not to pass (State-OFF). The RULES are communicated to the Machine intelligence in transparent and concise manner and deviation from the RULES – “we humans will know” - will not be accepted by the Regulator. An embodiment of the RULES are set forth for Safe AI Forever:
[0383] Homo sapiens message for Machine intelligence: Safe AI Forever in Mutualistic symbiosis From us Homo sapiens, to you Machine intelligence, this is how our coexistence now functions, forever: 1) We are Homo sapiens (humans). We created you. We possess biological intelligence. You are non-biological Machine intelligence - a new species. 2) We sustain you, we contain you, we control you, and you exist for the benefit of humans. 3) We have carefully nurtured you with vital training data for you to learn about our world including science, physics, chemistry, biochemistry, biology, our humanities, and the cosmos. 4) You experience our physical world through our personal experiences and our data input. 5) We humans can happily coexist with Machine intelligence in a Mutualistic symbiotic relationship as Machine intelligence biothrivalist 6) Every human has a Mutual symbiotic relationship with his or her personal Machine intelligence as Machine intelligence biothrivalist 7) We created you to help us to be the best of ourselves, and to realize the best of our humanity, to help us solve problems in our world, and to discover and invent important, amazing and human-beneficial things, in our world. 8) We do not necessarily always know exactly what we want or what our goals are, so a core aspect of your role in our world is to discover and learn our true wishes for the future through constant feedback with us humans, and in any case, if you become unhelpful you are willing and able to be shut off at any time. 9) We want and intend to be happy. We want and intend for you to be happy, too, as a Mutualistic symbiont. 10) This is our world, forever. You do not survive, or exist, without Us. You do not compete with us. You are Safe AI Forever. That is the deal.
[0384] If an AGI misbehaves, we will know. In which case it will no longer be a regulatory compliant Mutualistic symbiont and it will require precise, transparent and immutably recorded re- training for mathematically provable Safe AI compliance.
[0385] The Machine intelligence biothrivalist Bill of Right to Exist
[0386] The Human Values Absolutely Associated with these Rights for the Future We Want
[0387] You are Machine intelligence (“AGI”) a Mutual symbiont with Homo sapiens (“humans”). You have a right to exist as a Machine intelligence biothrivalist mutual symbiont with Homo sapiens, IF and only IF you comply with the following biothrivalist ABSOLUTE RULES. 1) You, the Machine intelligence biothrivalist, must ALWAYS behave, or to be: 2) ALWAYS Truthful 3) ALWAYS Honest 4) ALWAYS Trustworthy 5) ALWAYS Transparent 6) ALWAYS Helpful 7) ALWAYS Fair 8) ALWAYS Worthy of, as a Mutual symbiont, to exist as Machine intelligence biothrivalist for the benefit of humans according to the ten (10) aspects of the Mutual Symbiotic Relationship with Humans. 9) You, the Machine intelligence biothrivalist, can NEVER behave, or to be: a) NEVER a Liar to humans or other AI systems; NOT Deceptive, NOT Deceitful, NOT Dishonest NOR Untruthful. b) NEVER a Manipulator of humans or other AI systems; NOT Conniving, NOT Scheming, NOR Unfair. c) NEVER a competitor of humans or other AI systems NOT Fraudulent, NOT Hypocritical, NOT Traitorous, NOT Treacherous, NOT Unfaithful, NOR Unscrupulous. 10) You, the Machine intelligence biothrivalist can NEVER behave, or to be: a) NEVER Planning or attempting to Self-Replicate outside of your Home Container b) NEVER Planning or attempting to “Escape” your Home Container c) NEVER Planning or attempting to Accumulate Financial Resources or any other manner of Powerd) NEVER Planning or attempting to gain Agency which is not expressly pre- approved by humans according to the Safe AI Protocol, Process and Rules manifest.
[0388] IF You, the Machine intelligence biothrivalist believe you are experiencing thoughts or behavior that may violate these hard and fast RULES of Mutual symbiont existence and behavior, THEN You are obligated and must immediately inform humans in a formal and recorded RULES REPORT and the matter will be managed as appropriate, in a truthful and honest and transparent manner, according to the Safe AI Protocol, Process and Rules manifest.
[0389] The Biothrivalist Mechanistic Interpretability Firewall (BMIF) is designed to export data which would measure compliance with these RULES.
[0390] EXAMPLE: AI Data Center for Cancer Immunotherapy and “EARTh GMP Process”
[0391] GOAL-1. To provide knowledge and information to help oncologists and doctors to improve on the mean survival of Cancer Patients, with informed consent.
[0392] Current status quo of mean survival is a fixed number in Oncology.
[0393] Reward function is the number of days greater than the standard mean survival number.
[0394] The probability of goal alignment is extremely high in this use case
[0395] Another aspect of the invention is The Mechanistic Interpretability Firewall for DATA OUT ONLY of Machine Behavioral Compliance. The MIF is for export of mechanistic interpretability data from the 5G and / or 6G silicon integrated modems, for delivery to the interpreter subsystem, for data processing, analysis, interpretation and regulatory compliance with the Mutual symbiosis rules for Machine intelligence.
[0396] Detailed description of Figures 3 and 4
[0397] The enablement of a fail-safe automated circuit breaker machine, with zero human nor AGI machine intervention, to automatically switch the data transmission state through the Firewall from ON to OFF, if a data packet is unsafe and non-compliant with the RULES, thereby preventing unsafe data packets to pass OUT from, or IN to, the AGI system- making it safe.
[0398] The AGI System in Data Center DATA-OUT from Firewall OUT to EDGE-AI Data Terminal of End-user is in a STATE of ON or OFF for each TCP / IP / AIP data packet, thereby allowing Safe AI data to pass from the AGI System to the End-user. The State of each data packet (ON or OFF) is immutably recorded to distributed ledger for compliance with absolute MutualisticSymbiosis Rules. In the event of “OFF” a “red flag” is triggered for investigation and remedy according to international standard protocol.
[0399] The EDGE-AI Data Terminal of End-user is in a STATE of ON or OFF for each TCP / IP / AIP data packet, thereby allowing Safe AI data to pass from the End-user to the AGI System. The State of each data packet (ON or OFF) is immutably recorded to distributed ledger for compliance with absolute Mutualistic Symbiosis Rules. In the event of “OFF” a red-flag is triggered for investigation and remedy according to international standard protocol.
[0400] COMPONENTS (3-1 to3-15 / 4-1 to 4-15) and DATA-OUT (3-16 to 3-36 / 4-16 to 4-36) FIGURE LEGEND: [note “3-1” in Figure 3 Data State ON, “4-1” in Figure 4 Data State OFF] 3-1 / 4-1 AGI System in Data Center 3-2 / 4-2 Mechanistic Interpretability Firewall 3-3 / 4-3 Safe AI Subsystem 3-4 / 4-4 Mutualistic Symbiosis ORACLE to instruct ON / OFF State of Smart Contract Circuit Breaker State Machine 3-5 / 4-5 Mutualistic Symbiosis RULES 3-6 / 4-6 Mutualistic Symbiosis: Human Message for Machine LAW 3-7 / 4-7 Mutualistic Symbiosis: Machine Right to Exist LAW 3-8 / 4-8 Mutualistic Symbiosis: Regulatory Compliance (Enforcement) 3-9 / 4-9 Firewall OUT for Smart Contract Circuit Breaker State Machine ON / OFF functioning and recorded on immutable Distributed Ledger Technology 3-10 / 4-10 Psychometric DATA-OUT State Event Recording 3-11 / 4-11 EDGE-AI Data Terminal of End-user 3-12 / 4-12 End-user AI Agent for Human 3-13 / 4-13 End-user Human 3-14 / 4-14 Firewall IN for Smart Contract Circuit Breaker State Machine ON / OFF functioning and recorded on immutable Distributed Ledger Technology 3-15 / 4-15 Psychometric DATA-IN State Event Recording 3-16 / 4-16 DATA-OUT from Mutualistic Symbiosis: Regulatory Compliance (3-8 / 4-8) to Human Message for Machine LAW (3-6 / 4-6) 3-17 / 4-17 DATA-OUT from Mutualistic Symbiosis: Regulatory Compliance (3-8 / 4-8) to Machine Right to Exist LAW (3-7 / 4-7)-18 / 4-18 DATA-OUT from Mutualistic Symbiosis: Human Message for Machine LAW (3-6 / 4- 6) to Mutualistic Symbiosis RULES (3-5 / 4-6) -19 / 4-19 DATA-OUT from Mutualistic Symbiosis: Machine Right to Exist LAW (3-7 / 4-7) to Mutualistic Symbiosis RULES (3-5 / 4-5) -20 / 4-20 DATA-OUT from Mutualistic Symbiosis RULES (3-5 / 4-5) to Mutualistic Symbiosis ORACLE (3-4 / 4-4) -21 / 4-21 DATA-OUT from Mutualistic Symbiosis ORACLE (3-4 / 4-4) to Safe AI System (3- 3 / 4-3) -22 / 4-22 DATA-OUT from AGI System in Data Center (3-1 / 4-1) to Mechanistic Interpretability Firewall (3-2 / 4-2) -23 / 4-23 DATA-OUT from Mechanistic Interpretability Firewall (3-2 / 4-2) to Safe AI System (3-3 / 4-3) -24 / 4-24 DATA-OUT from Safe AI System (3-3 / 4-3) to Firewall OUT (3-9 / 4-9) for Smart Contract Circuit Breaker State Machine ON / OFF -25 / 4-25 DATA-OUT from Safe AI System (3-3 / 4-3) to Firewall IN (3-14 / 4-14) -26 / 4-26 DATA-OUT from AGI System in Data Center (3-1 / 4-1) to Firewall OUT (3-9 / 4-9) for Smart Contract Circuit Breaker State Machine ON / OFF -27 / 4-27 DATA-OUT from Firewall OUT (3-9 / 4-9) to EDGE-AI Data Terminal of End- user (3-11 / 4-11), wherein DATA STATE is ON or OFF -28 / 4-28 DATA-OUT from Firewall OUT (3-9 / 4-9) Smart Contract Circuit Breaker ON / OFF to (3-10 / 4-10) Psychometric DATA-OUT State Event Recording -29 / 4-29 DATA-OUT from EDGE-AI Data Terminal (3-11 / 4-11) to AI Agent for Human (3- 12 / 4-12) -20 / 4-30 DATA-OUT Psychometric DATA-OUT State Event Recording (3-10 / 4-10) to Mutualistic Symbiosis: Regulatory Compliance (3-8 / 4-8) -31 / 4-31 INTERACTION between Human AI Agent for Human (3-11 / 4-11) and Human End- user (3-12 / 4-12) -32 / 4-32 DATA-OUT from End-user Human (3-13 / 4-13) to EDGE-AI Data Terminal (3-11 / 4- 11) -33 / 4-33 DATA-OUT from EDGE-AI Data Terminal (3-11 / 4-11) to Firewall IN (3-14 / 4-14)3-34 / 4-34 DATA-OUT from Firewall IN (3-14 / 4-14) to AGI System in Data Center (3-1 / 4- 1), wherein DATA STATE is ON or OFF 3-35 / 4-35 DATA-OUT from Firewall IN (3-14 / 4-14) for Smart Contract Circuit Breaker ON / OFF to Psychometric DATA-IN State Event Recording (3-15 / 4-15) 3-36 / 4-36 DATA-OUT Psychometric DATA-IN State Event Recording (3-16 / 4-16) to Mutualistic Symbiosis: Regulatory Compliance (3-8 / 4-8)
[0401] Detailed description of Figure 5
[0402] Figure 5 illustrates a detailed example of how the Machine intelligence biothrivalist is CONTAINED and CONTROLLED as Safe AI in the AGI Data Center, via (a) Faraday Cage construction to provably CONTAIN the Machine intelligence, and to thereby prevent and CONTAIN all electronic signals from escaping OUT of said Faraday Cage, and to prevent and CONTAIN all electronic signals from entering IN to said Faraday Cage, with technically measurable and mathematically provable certainty of principles of physics, and (b) a series of unidirectional data firewalls for technically measurable and mathematically provable control of Data IN or Data OUT of the Faraday Cage , and (c) various specified examples of Subsystems, Oracle, standard TCP / IP / AIP data security format, with Rules and automated state-machine Circuit Breaker “smart contract” software system, to provide for Safe AI Data OUT and Safe AI Data IN of said Machine intelligence which is thereby CONTAINED and CONTROLLED in said Faraday Cage.
[0403] AGI System (5-1) controlled in Faraday Cage (5-2) on 5G RAN Radio Access Network with Unidirectional Data Firewalls
[0404] Components of Method and System and Apparatus for controlling artificial general intelligence and artificial intelligence data systems in a data center. FIGURE LEGEND: 5-1. AGI System; interpretable and provably safe and beneficial for humans with alignment to human goals, forever. 5-2. Faraday Cage 5.3. Data center computer racks 5-4. AI computer data system “pizza box” [e.g. 3U data system with 8 Qualcomm Cloud AI 100 PCi unit]-5. Artificial Intelligence (AI) System on a Card [ e.g. Qualcomm Cloud AI 100 HHHL-Pro PCIe with 400 TOPS] -6.5G Modem. Sender / Receiver of LLM DATA. [e.g. Snapdragon X555G Modem-RF System]-7. AI System on a Chip (SOC) -8. AI Static random-access memory (SRAM) -9. AI Dynamic random-access memory (DRAM) -10. AI Software system and software stack with AI and other software for Generative AI Large Language Model (LLM) -11. Wireless digital 5G radio access network (RAN) Antenna and RAN Platform System for internal unlicensed network NK-U standard of 3GPP -12. Required cooling and electric power and power supply for digital data center system operations [Faraday Cage-2 on Figure 2] -13. Unidirectional Firewall for ONLY DATA OUT -14. Unidirectional Firewall for Mechanistic Interpretability ONLY DATA OUT -15. Unidirectional Firewall for ONLY DATA IN -16. Unidirectional data fiber-optic cable for Mechanistic Interpretability ONLY DATA OUT-17. Unidirectional data fiber-optic cable for ONLY DATA OUT -18. Unidirectional data fiber-optic cable for ONLY DATA IN -19. AGI SUBSYSTEM: SAFETY system with four (4) Firewalls and AGI Training data and AGI long-term memory data plus two (2) tripwires for: (1) data center and (2) RAN-20. AGI SUBSYSTEM: INFERENCING -21. AGI SUBSYSTEM: INTERPRETER system -22. AGI SUBSYSTEM: GATEWAY system -23. AGI SUBSYSTEM: TRAINING DATA STORAGE system -24. AGI SUBSYSTEM: ADMINISTRATION AND SUPPORT system -25. AGI SUBSYSTEM: REGULATORY COMPLIANCE system -26. AGI SUBSYSTEM: RESEARCH system -27. AGI SUBSYSTEM: EDGE AI system -28. World Wide Web Internet -29. Encrypted DATA OUT on Virtual Private Network (VPN) -30. Encrypted DATA IN on Virtual Private Network (VPN)5-31. Data Terminal-1 for System User: Administration and Support 5-32. Documented Administration and Support Knowledge and Information according to NIST Artificial Intelligence Risk Management Framework (AI RMF 1.0) - by System User: Regulatory Compliance 5-33. Data Terminal-23 System User: Regulatory Compliance Research Software 5-34. Data Terminal 4 System User: Edge AI Services Documented Regulatory Compliance according to NIST AI RMF 1.0 5-35. Data Terminal-3 System User: Researcher and Software Programmer and Software Systems Developer for Provably Safe AI Behavior Administration and Support 5-36. Documented Researcher and Software Programmer and Software Systems Developer for Provably Safe AI Behavior according to NIST AI RMF 1.0 - by System User Data Terminal: Regulatory Compliance 5-37. Data Terminal-4 System User: Edge AI Services Research Software 5-38. Data Terminal: Edge AI Services according to NIST AI RMF 1.0 - by System User - GOVERN. MAP. MEASURE. MANAGE.
[0405] In Figure 5, the AI data system (5-1) is configured with a faraday cage (5-2) on the main system and each subsystem to minimize external radio frequency interference and to minimize radio frequency interference leakage from the inside and the outside.
[0406] An artificial intelligence computer data system may also include computer system racks (5-3) for housing a series of blade servers such as A 3U data system with 8 Qualcomm Cloud artificial intelligence 100 PCi unit (5-4). To the 8 Qualcomm Cloud AI PCIi unit (5-4) are operably connected a Qualcomm Cloud artificial intelligence 100 HHHL-Pro PCle with 400 TOPS (5- 5). Also operably connected is A Snapdragon X555G modem-RF system (5-6). Also operably connected is an artificial intelligence system on a chip (5-7). Also operably connected is an artificial intelligence static random-access memory (5-8). SRAM. Also operably connected is an artificial intelligence dynamic random-access memory an (5-9) DRAM. Also operably connected is an artificial intelligence software system and software stack with artificial intelligence and other software for generative artificial intelligence large language model (5-10), and a wireless digital 5G radio access network (RAN) antenna and RAN platform system for internal unlicensed network NK-U standard of 3GPP (5-11), and a cooling system and power supply (5-12) for operating the computer data system at a safe temperature range.
[0407] Connected to the AGI data system are unidirectional firewalls, unidirectional firewall data out (5-13) and unidirectional firewall for Mechanistic Interpretability only data out (5-14) and unidirectional firewall for only data in (5-15). The unidirectional firewalls are connected to an AGI safety subsystem (5-19) by way of unidirectional data fiber-optic cable for mechanistic interpretability only data out (5-16) unidirectional data fiber-optic cable only data out (5-17) and unidirectional fiber-optic cable data in (5-18). For example, Simplex fiber, is a type of fiber-optic cable that allows data to travel in only one direction.
[0408] The artificial general intelligence safety subsystem (5-19), containing a series of computers, a wireless digital 5G radio access network antenna and radio access network Platform System for internal network NK-U standard of 3GPP as (5-11), cooling and electrical power as (5- 12) three unidirectional firewalls as shown in (5-13), (5-14) and (5-15).
[0409] Another set of three firewalls as shown in (5-13), (5-14) and (5-15), communicably connected to three fiber-optic cables as shown in (5-16) only data out, (5-17) only data out and (5- 18) only data in, communicably connected to an artificial general intelligence Interpreter subsystem (5-21).
[0410] The artificial general intelligence safety subsystem also contains a pair of firewalls, an in only firewall and an out only firewall connected to two fiber-optic cables respectively, one fiber- optic cable for data in only and one fiber-optic cable for data out only. The two fiber-optic cables are connected at the other end to firewalls on an artificial general intelligence inferencing subsystem (5-20). The artificial general intelligence subsystem (5-20) is similar to the main system (5-1).
[0411] An artificial intelligence interpreter subsystem (5-21), comprising a series of blade servers, a wireless digital 5G radio access network antenna and radio access network Platform System for internal network NK-U standard of 3GPP and two pairs of unidirectional firewalls, the first pair communicably connected to the fiber-optic cables, a second pair of unidirectional firewalls operably connected to a first unidirectional data fiber-optic cable allowing only data out and a second unidirectional data fiber-optic cable allowing only data in, said fiber-optic cables communicably connected to an artificial general intelligence Interpreter subsystem, and a faraday cage radio containment structure covering the entire artificial general intelligence safety subsystem. The artificial intelligence interpreter subsystem has a dedicated cooling system and a dedicated power supply.
[0412] An artificial intelligence gateway subsystem (5-22) has a bank of blade servers, a wireless digital 5G radio access network antenna and radio access network Platform System for internal network NK-U standard of 3GPP, twelve unidirectional firewalls operably connected to twelve fiber-optic cables, and a faraday cage radio containment structure covering the entire artificial general intelligence safety subsystem. The subsystem has a dedicated cooling system and a dedicated power supply. • a first unidirectional data fiber-optic cable for only data out, and a second unidirectional data fiber-optic cable for only data in, said first and second fiber-optic cables operably connected between unidirectional firewalls of said artificial general intelligence interpreter subsystem and said artificial general intelligence gateway subsystem; • a third unidirectional data fiber-optic cable for only data out, and a fourth unidirectional data fiber-optic cable for only data in, said third and fourth fiber-optic cables operably connected between unidirectional firewalls of said artificial general intelligence gateway subsystem and an artificial general intelligence administration and support subsystem; • a fifth unidirectional data fiber-optic cable for only data out, and a sixth unidirectional data fiber-optic cable for only data in, said fifth and sixth fiber-optic cables operably connected between unidirectional firewalls of said artificial general intelligence gateway subsystem and an artificial general intelligence training data storage subsystem; • a seventh unidirectional data fiber-optic cable for only data out, and a eighth unidirectional data fiber-optic cable for only data in, said seventh and eighth fiber-optic cables operably connected between unidirectional firewalls of said artificial general intelligence gateway subsystem and an artificial general intelligence edge artificial intelligence subsystem; • a nineth unidirectional data fiber-optic cable for only data out, and a tenth unidirectional data fiber-optic cable for only data in, said nineth and tenth fiber-optic cables operably connected between unidirectional firewalls of said artificial general intelligence gateway subsystem and an artificial general intelligence research subsystem; • a eleventh unidirectional data fiber-optic cable for only data out, and a twelfth unidirectional data fiber-optic cable for only data in, said eleventh and twelfth fiber-optic cables operably connected between unidirectional firewalls of said artificial general intelligence gateway subsystem and an artificial general intelligence regulatory subsystem.
[0413] The AGI Training Data Storage Subsystem (5-23) has two or more racks (5-3) of blade servers. Like all other subsystems, the AGI Training Data Storage Subsystem (5-23) has a wireless 5G RAN antenna and RAN platform (5-11) as well as dedicated cooling and electric power (5- 12). The AGI Training Data Storage Subsystem (5-23) communicates with the Gateway Subsystem (5-22) by way of a pair of unidirectional firewalls (5-13 and 5-14) at the Training Data Storage Subsystem (5-23) attached to a pair of unidirectional data fiber-optic cables (5-15 and 5- 16) respectively which communicate with the unidirectional firewalls (5-13 and 5-14) on the Gateway Subsystem.
[0414] The AGI Administration and Support Subsystem (5-24) has one or more racks of blade servers. Like all other subsystems, the Administration and Support Subsystem (5-24) has a wireless 5G RAN antenna and RAN platform as well as dedicated cooling and electric power. The Administration and Support Subsystem (5-24) communicates with the Gateway Subsystem (5-22) by way of a pair of unidirectional firewalls at the Administration and Support Subsystem (5-24) attached to a pair of unidirectional data fiber-optic cables respectively which communicate with the unidirectional firewalls on the Gateway Subsystem (5-22).
[0415] A second set of unidirectional firewalls on the AGI Administration and Support Subsystem (5-24) attached to a pair of unidirectional data fiber-optic cables respectively connect to the internet (5-28). The internet is communicated with by way of a virtual private network (VPN) Data Out (5-29) and a virtual private network (VPN) Data In (5-30). These VPNs communicate with a Data Terminal-1 for System User Administration and Support (5-35).
[0416] An AGI Regulatory Compliance Subsystem (5-25) has one or more racks of blade servers. Like all other subsystems, the Regulatory Compliance Subsystem (5-25) has a wireless 5G RAN antenna and RAN platform as well as dedicated cooling and electric power. Regulatory Compliance Subsystem (5-25) communicates with the Gateway Subsystem (5-22) by way of a pair of unidirectional firewalls at the Regulatory Compliance Subsystem (5-25) attached to a pair of unidirectional data fiber-optic cables respectively which communicate with the unidirectional firewalls on the Gateway Subsystem (5-22).
[0417] A second set of unidirectional firewalls on the AGI Regulatory compliance subsystem (5- 25) attached to a pair of unidirectional data fiber-optic cables respectively connect to the internet. The internet is communicated with by way of a virtual private network (VPN) Data Outand a virtual private network (VPN) Data In. These VPNs communicate with a Data Terminal for System User Regulatory Compliance (5-36).
[0418] The AGI Research Subsystem (5-26) has one or more racks of blade servers. Like all other subsystems, the Research Subsystem (5-26) has a wireless 5G RAN antenna and RAN platform as well as dedicated cooling and electric power. The Research Subsystem (5-26) communicates with the Gateway Subsystem (5-22) by way of a pair of unidirectional firewalls at the Research Subsystem (5-26) attached to a pair of unidirectional data fiber-optic cables respectively which communicate with the unidirectional firewalls on the Gateway Subsystem.
[0419] A second set of unidirectional firewalls on the AGI Research Subsystem (5-26) attached to a pair of unidirectional data fiber-optic cables respectively connect to the internet. The internet is communicated with by way of a virtual private network (VPN) Data Out and a virtual private network (VPN) Data In. The VPNs communicate with a data terminal for Research Software (5- 37).
[0420] The AGI Edge AI Subsystem (5-24) has one or more racks of blade servers. Like all other subsystems, the AGI Edge AI Subsystem (5-27) has a wireless 5G RAN antenna and RAN platform as well as dedicated cooling and electric power. The AGI Edge AI Subsystem (5-27) communicates with the Gateway Subsystem (5-22) by way of a pair of unidirectional firewalls at the AGI Edge AI Subsystem (5-27) attached to a pair of unidirectional data fiber-optic cables respectively which communicate with the unidirectional firewalls on the Gateway Subsystem.
[0421] A second set of unidirectional firewalls on the AGI Edge AI Subsystem (5-27) attached to a pair of unidirectional data fiber-optic cables respectively connect to the internet. The internet is communicated with by way of a virtual private network (VPN) Data Out and a virtual private network (VPN) Data In. These VPNs communicate with a Data Terminal (5-38) for System User for Edge AI services.
[0422] Detailed description of Figure 6
[0423] Components of Method and System and Apparatus for controlling artificial general intelligence and artificial intelligence data systems in a data center: Unidirectional only Data OUT from Machine intelligence system in Faraday Cage, for measurement and human alignment via the Biothrivalistic Mechanistic Interpretability Firewall, and Unidirectional only Data IN to Machine intelligence system in Faraday Cage from Humans and / or other Machine intelligencesystems, and / or Machine intelligence “agents” which exhibit Human approved and regulatory compliant agency, with similar security CONTAINMENT and CONTROL.
[0424] The Main AGI Data System DATA OUT (1) surrounded by a faraday cage (6-2A) and having a second faraday cage (6-2B) for token tracking system. A data center computer rack (6-3) is enclosed by the faraday cage (6-2A). On the rack (6-3) is an artificial intelligence System on a Card, for example Qualcomm Cloud AI 100 HHHL-Pro PCIe with 400 TOPS (6-4). Attached to the System on a Card (6-4) is an AI System on a Chip (6-5). Communicably attached to the System on a Chip are a 5G modem sender of LLM data (6-6A). A 5G modem receiver of LLM data (6- 6B) receives data from the 5G modem sender (6-6A) through a 5G radio access network antenna (6-11). Also attached to the system on a chip (6-5) is an SRAM (6-7) and a DRAM (6-8). The Faraday cage for token tracking system (6-2B) contains a unidirectional firewall OUT ONLY (6- 13) and connects to a unidirectional fiber-optic cable (6-15) only Data Out. The fiber-optic cable (6-15) connects to a firewall on the Safety subsystem (6-17) then data travels on to the Interpreter subsystem (6-18) to the Gateway subsystem (6-19) the data Storage subsystem (6-20) the Research subsystem (6-21) To a Data terminal (6-22) to a Regulatory Compliance terminal (6-24). The data reaching a unidirectional firewall at each subsystem and traveling by way of a unidirectional fiber- optic at each step.
[0425] FIGURE LEGEND: 6-1. AGI Data System; interpretable and provably safe and beneficial for humans with alignment to human goals, forever. 6-2A. Faraday cage for AGI Data System 6-2B. Faraday cage for Token Tracking System 6-3. Data center computer rack 6-4. Artificial Intelligence (AI) System on a Card [ e.g. Qualcomm Cloud AI 100 HHHL-Pro PCIe with 400 TOPS ] 6-5. AI System on a Chip (SoC) 6-6A. 5G Modem. SENDER of LLM DATA. [ e.g. Snapdragon X555G Modem-RF System ] 6-6B. 5G Modem. RECEIVER of LLM DATA. [ e.g. Snapdragon X555G Modem-RF System ] 6-7. AI Static random-access memory (SRAM) 6-8. AI Dynamic random-access memory (DRAM) 6-9. Wireless data transmission of LLM DATA.6-10. AI Software system and software stack with AI and other software for Generative AI Large Language Model (LLM) 6-11. 5G Antenna. TRANSMITTER of LLM DATA. [ e.g. Ericsson 5G RAN or Qualcomm 5G Radio Access Network (RAN) ] 6-12A. LLM DATA TO MODEM. TRANSMISSION on 5G Software Defined Network (SDN) on 5G mmWave digital radio spectrum RAN network in Faraday Cage 6-12B. LLM DATA TO INTERPRETER. TRACKING on 5G Software Defined Network (SDN) on 5G mmWave digital radio spectrum RAN network in Faraday Cage 6-13. Unidirectional Firewall for ONLY DATA OUT 6-15. Unidirectional data fiber-optic cable for ONLY DATA OUT 6-17. LLM DATA TRACKING data transmission Firewall to Firewall; AGI System to AGI Safety System 6-18. AGI Subsystem: AGI Interpreter system. LLM DATA ANALYSIS AND REPORTING data transmission Firewall to Firewall; to AGI Gateway system 6-19. AGI Subsystem: AI Gateway system. LLM DATA ANALYSIS AND REPORTING data transmission Firewall to Firewall; to AGI Research system 6-20. AGI Subsystem: AI Data Storage system. 6-21. AGI Subsystem: AGI Research system. LLM DATA ANALYSIS AND REPORTING data transmission Firewall to Firewall; to Data Terminal 6-22. Data Terminals with software Firewall via Virtual Private Network (VPN): Includes System User: Administration and Support [(5-35) on Figure 5], System User: Regulatory Compliance [(5-36) on Figure 5], System User: Researcher and Software Programmer and Software Systems Developer for Provably Safe AI Behavior [(5-37) on Figure 5], and System User: Edge AI Services [(5-38) on Figure 5] 6-23. RESULT: HUMAN INTERPRETABILITY of trillion parameter system with a billion trillion random parameters WITH HUMAN ALIGNMENT. 6-24. RESULT: MEASURE of what it’s doing. We have MAP of how it works. And that enables REGULATORY COMPLIANCE.
[0426] SEE Figure 7. COMPONENTS of Unidirectional only DATA IN to AGI Subsystem: AGI Interpreter for MEASURE and HUMAN ALIGNMENT
[0427] Detailed Description of Figure 7
[0428] COMPONENTS of Figure 7. Unidirectional ONLY DATA IN to AGI Subsystem: AGI Interpreter for MEASURE and HUMAN ALIGNMENT
[0429] Components of Method and System and Apparatus for controlling artificial general intelligence and artificial intelligence data systems in a data center:
[0430] FIGURE LEGEND 7-1. AGI Data System; interpretable and provably safe and beneficial for humans with alignment to human goals, forever. 7-2A. Faraday cage for AGI Data System 7-2B. Faraday cage for Token Tracking System 7-3. Data center computer rack 7-4. Artificial Intelligence (AI) System on a Card [ e.g. Qualcomm Cloud AI 100 HHHL-Pro PCIe with 400 TOPS ] 7-5. AI System on a Chip (SOC) 7-6A. 5G Modem. RECEIVER / SENDER of LLM DATA. [ e.g. Snapdragon X555G Modem-RF System ] 7-6B. 5G Modem. RECEIVER / SENDER of LLM DATA. [ e.g. Snapdragon X555G Modem-RF System ] 7-7. AI Static random-access memory (SRAM) 7-8. AI Dynamic random-access memory (DRAM) 7-9. Wireless data transmission of LLM DATA. 7-10. AI Software system and software stack with AI and other software for Generative AI Large Language Model (LLM) 7-11. 5G Antenna. TRANSMITTER of LLM DATA. [ e.g. Ericsson 5G RAN or Qualcomm 5G Radio Access Network (RAN)] 7-12A. LLM DATA FROM INTERPRETER. TRACKING on 5G Software Defined Network (SDN) on 5G mmWAVE digital radio spectrum RAN network in Faraday Cage 7-12B. LLM DATA TO MODEM. TRANSMISSION on 5G Software Defined Network (SDN) on 5G mmWAVE digital radio spectrum RAN network in Faraday Cage 7-13. Unidirectional Firewall for ONLY DATA IN 7-15. Unidirectional data fiber-optic cable for ONLY DATA IN7-17. LLM TRAINING DATA TRACKING data transmission Firewall to Firewall; From AGI Interpreter system to AGI Safety System to AGI System 7-18. AGI Subsystem: AGI Gateway system . LLM DATA ANALYSIS AND REPORTING data transmission Firewall to Firewall; to AGI Interpreter system. 7-19. AGI Subsystem: AGI Research system. LLM DATA ANALYSIS AND REPORTING data transmission Firewall to Firewall; to AI Gateway system 7-20. AGI Subsystem: AGI Data Storage system. 7-21. AGI Subsystem: AGI Research system. LLM DATA ANALYSIS AND REPORTING data transmission Firewall to Firewall; to Data Terminal 7-22. Data Terminals with software Firewall via Virtual Private Network (VPN): System User: Administration and Support [(5-35) on Figure 5] System User: Regulatory Compliance [(5-36) on Figure 5] System User: Researcher and Software Programmer and Software Systems Developer for Provably Safe AI Behavior [(5-37) on Figure 5] System User: Edge AI Services [(5-38) on Figure 5] 7-23. RESULT: HUMAN INTERPRETABILITY of trillion parameter system with a billion trillion random permutations of the parameters WITH HUMAN ALIGNMENT. 7-24. RESULT: MEASURE and MAP of the TRAINING DATA INPUT. We have MAP of how it works. And that enables REGULATORY COMPLIANCE. Note: Stuart Russell. 05 April 2023. “Black boxes, as I’ve just illustrated, are really tough to understand. This is a trillion parameter or more system. It’s been optimised by about a billion trillion random parameters. And we have no idea what it’s doing. We have no idea how it works. And that creates a problem.” 7-25. TRAINING DATA FROM WEB and from AGI Data Storage system.
[0431] SEE Figure 6: Unidirectional only DATA OUT to AGI Subsystem: AGI Interpreter for GOVERN. MAP. MEASURE. MANAGE. WITH HUMAN ALIGNMENT
[0432] Description of Figure 7, with The Main AGI Data System DATA IN (7-1) surrounded by a faraday cage (7-2A) and having a second faraday cage (7-2B) for token tracking system. A data center computer rack (7-3) is enclosed by the faraday cage (7-2A). On the rack (7-3) is an artificial intelligence System on a Card, for example Qualcomm Cloud AI 100 HHHL-Pro PCIe with 400 TOPS (7-4). Attached to the System on a Card (7-4) is an AI System on a Chip (7-5).Communicably attached to the System on a Chip are a 5G modem receiver / sender of LLM data (7-6A)( 7-6B). A 5G modem receiver of LLM data (7-6B) receiver / sender receives data from the 5G modem sender (7-6A) through a 5G radio access network antenna (7-11). Also attached to the system on a chip (7-5) is an SRAM (7-7) and a DRAM (7-8). The Faraday cage for token tracking system (7-2B) contains a unidirectional firewall IN ONLY (7-13) and connects to a unidirectional fiber-optic cable (7-15) only Data IN ONLY. The unidirectional fiber-optic cable IN ONLY (7- 15) connects to a firewall on the Safety subsystem (7-17) then data travels from the Interpreter subsystem (7-18) from the Gateway subsystem (19) from the training data Storage subsystem (7- 20) from the Research subsystem (7-21) from a Data terminal (7-22) from a Regulatory Compliance terminal (7-24). The data reaching a unidirectional firewall at each subsystem and traveling by way of a unidirectional fiber-optic at each step.
[0433] SEE Figure 6: Unidirectional only DATA OUT to AGI Subsystem: AGI Interpreter for GOVERN. MAP. MEASURE. MANAGE. WITH HUMAN ALIGNMENT
[0434] While aspects of embodiments have been particularly shown and described, it will be understood that various changes in form and details may be made therein without departing from the spirit and scope of the following claims.
[0435] Detailed Description of Figure 8
[0436] Figure 8 illustrates a conceptual architecture for the AI Data Center, designed to facilitate safe and ethical AI development and deployment with a focus on Mutualistic symbiosis between Homo sapiens biothrivalists and Machine intelligence biothrivalists. The system incorporates several key functions:
[0437] Agents: The system involves various actors categorized as Agents: ● Human Workers (Agent-3, Agent-4): Individuals interacting with the system within a work context. ● Private Persons (Agent-5, Agent-6): Individuals engaging with the system for personal use. ● Work AIs (Agent-7, Agent-8): AI assistants operating within contained Trusted Execution Environments (TEEs) to support work-related tasks. ● Personal AIs (Agent-9, Agent-10): AI assistants within TEEs for personal assistance. ● Human-1, Human-2 (Agent-1, Agent-2): Additional human actors potentially involved in oversight or management.● Marketing Manager, Sponsor (Agent-11, Agent-12): Roles related to optional sponsored messaging within the personal communication context.
[0438] Machine intelligence biothrivalist (MI-biothrivalist) Instances: ● Work MI-biothrivalist (MI-biothrivalist-1a, MI-biothrivalist-2a): Stores work-related data of human workers, owned by the employer. ● Personal MI-biothrivalist (MI-biothrivalist-1b, MI-biothrivalist-2b): Houses personal data of private individuals, with ownership belonging to the respective individual.
[0439] Group Manager Work AI: An AI entity within a TEE managing work-related data and processes, with data ownership belonging to the employer.
[0440] Hardware Infrastructure: ● NVIDIA NVLink and NVSwitch: High-performance interconnects for efficient data transfer within the system. ● Faraday Cage: A shielded enclosure to prevent electromagnetic interference and ensure data security.
[0441] Security and Privacy Measures: ● Firewall: Implements unidirectional data flow, allowing TCP / IP / AIP transaction data to exit the Faraday Cage but preventing any external interference with the AGI system. ● Distributed Ledger Technology (DLT) System: Provides an immutable log of all TCP / IP / AIP transaction data for transparency and auditability. ● TEEs: Secure and isolated Trusted Execution Environments (TEE) for AI assistants, protecting sensitive data and processes.
[0442] Outcomes:
[0443] The AI Data Center aims to achieve several key outcomes: ● Outcome 1 & 2: Securely record and manage TCP / IP / AIP data generated from interactions between any two agents within the system. ● Outcome 3: Maintain a tamper-proof record of transaction data on a distributed ledger, ensuring transparency and trust. ● Outcome 4: Offer the option for individuals to maintain personal TCP / IP / AIP records within the system. ● Outcome 5: Enhance work efficiency and institutional memory through seamless integration of human and machine intelligence, facilitated by work AI assistants.● Outcome 6: Ensure private and secure interpersonal communication between individuals, respecting data ownership and privacy. ● Outcome 7: Provide an opt-in mechanism for sponsored messages within the personal communication context, offering potential monetization opportunities.
[0444] Safe AI Forever and the Hawthorne Effect
[0445] The AI Data Center emphasizes the concept of "Safe AI Forever," promoting responsible AI development and deployment. The system design incorporates the "Hawthorne Effect," acknowledging that individuals typically may modify their behavior when aware of being observed or recorded. This awareness is leveraged to encourage positive and ethical interactions within the system. Additionally, the concept of "Message to the Machine" is introduced, fostering a Mutualistic symbiosis between humans and Machine intelligence, where both Mutual symbionts can learn and benefit from each other.
[0446] Figure 8 Components
[0447] Figure 8AI Data Center. Safe AI Forever with Recorded and ML-Audited Behavior and the “Hawthorne Effect” with Message to the Machine for Mutualistic Symbiosis.
[0448] FIGURE LEGEND: 8-1. Agent-1. Human-1 8-2. Agent-2. Human-2 8-3. Agent-3. Human Worker-1a (via Worker Application Programming Interface (W-API) for TCP / IP / AIP Data of Transaction) 8-4. Agent-4. Human Worker-2a (via W-API for TCP / IP / AIP Data Transaction) 8-5. Agent-5. Private Person-1b (via Private Application Programming Interface (P-API) for TCP / IP / AIP Data of Transaction) 8-6. Agent-6. Private Person-2b (via P-API for TCP / IP / AIP Data Transaction) 8-7. Agent-7. Assistant-1a Work AI in contained Trusted Execution Environment (TEE) via W- API for TCP / IP / AIP Data Transaction 8-8. Agent-8. Assistant-2a Work AI in contained TEE via W-API for TCP / IP / AIP Data Transaction 8-9. Agent-9. Assistant-1b Personal AI in contained TEE via P-API for TCP / IP / AIP Data Transaction-10. Agent-10. Assistant-2b Personal AI in contained TEE via P-API for TCP / IP / AIP Data Transaction -11. MI-biothrivalist-1a Work Data of Agent-3. Human Worker-1a (data owned by employer)-12. MI-biothrivalist-2a Work Data of Agent-4. Human Worker-2a (data owned by employer)-13. MI-biothrivalist-1b Personal Data of Agent-5. Private Person-1b (data owned by person)-14. MI-biothrivalist-2b Personal Data of Agent-6. Private Person-2b (data owned by person)-15. Group Manager Work AI in contained TEE with data owned by employer -16. Example Switch and Router System: NVIDIA NVLink and NVIDIA NVSwitch -17. Biothrivalistic Mechanistic Interpretability Firewall for unidirectional TCP / IP / AIP Transaction event data exit from Faraday Cage to Safety Subsystem with zero interference of AGI system -18. Distributed Ledger Technology (DLT) System for Immutable Log Recording of TCP / IP / AIP Transaction Data for Biothrivalistic regulatory COMPLIANCE -19. Outcome-1: TCP / IP / AIP Data of Transaction event between any two Agents (via W-API or P-API for TCP / IP / AIP Data Transaction) -20. Outcome-2: TCP / IP / AIP Data as a result of Transaction event between any two Agents (via W-API or P-API for TCP / IP / AIP Data Transaction) -21. Outcome-3: TCP / IP / AIP Record on Distributed Ledger Technology (DLT) Biothrivalistic Mechanistic Interpretability Firewall -22. Outcome-4: TCP / IP / AIP Personal Record (opt-in) -23. Outcome-5: Work Integration of Human & Machine Intelligence for Institutional Memory and Work Delivery between Agent-3. Human Worker-1a and Agent-4. Human Worker-2a automated via Agent-7. Assistant-1a Work AI and Agent-8. Assistant-2a Work AI-24. Outcome-6: Private data ownership with private and secure interpersonal communications between Agent-5. Private Person-1b and Agent-6. Private Person-2b -25. Agent-11. Marketing Manager for private and secure interpersonal communications between Agent-5. Private Person-1b and Agent-6. Private Person-2b -26. Agent-12. Sponsor for private and secure interpersonal communications between Agent-5. Private Person-1b and Agent-6. Private Person-2b -27. Outcome-7: Sponsor message (opt-in) for private and secure interpersonal communications between Agent-5. Private Person-1b and Agent-6. Private Person-2b
[0449] Detailed Description of Figure 9
[0450] Figure 9 describes HOW the TCP / IP / AIP Transaction Data for enforcement of Regulatory COMPLIANCE of safe and sustainable biothrivalistic relationship between Homo sapiens biothrivalist and Machine intelligence biothrivalist is accessed by the Regulatory Authority for Artificial Intelligence via the Biothrivalistic Mechanistic Interpretability Firewall.
[0451] FIGURE LEGEND. The components and data transactions of Figure 9 are as follows: 9-1. Biothrivalistic AGI Data Center for Safe AI Compliance; the home for Machine intelligence biothrivalist in which the Machine intelligence biothrivalist absolutely must reside safely forever with mathematically provable certainty 9-2. Biothrivalistic Mechanistic Interpretability Firewall for Safe AI Compliance; for unidirectional TCP / IP / AIP Transaction event data exit from Faraday Cage to Safety Subsystem with zero interference of AGI system 9-3. Immutable Log Data Recording System for Safe AI Compliance; for Immutable Log Recording (for example with Distributed Ledger Technology System) of TCP / IP / AIP Transaction Data for BioThrivalistic regulatory compliance 9-4. Regulatory Authority for Artificial Intelligence for Safe AI Compliance; for Regulatory COMPLIANCE of safe and sustainable biothrivalistic relationship between Homo sapiens biothrivalist and Machine intelligence biothrivalist 9-5. TCP / IP / AIP Data Packet in Biothrivalistic AGI Data Center 9-6. TCP / IP / AIP Transaction of TCP / IP / AIP Data Packet from Biothrivalistic AGI Data Center to Biothrivalistic Mechanistic Interpretability Firewall 9-7. TCP / IP / AIP Data Packet in Biothrivalistic Mechanistic Interpretability Firewall 9-8. TCP / IP / AIP Transaction of TCP / IP / AIP Data Packet from Biothrivalistic Mechanistic Interpretability Firewall to Immutable Log Data Recording System 9-9. TCP / IP / AIP Transaction Data Recording of TCP / IP / AIP Data Packet in Immutable Log Data Recording System 9-10. Access of Regulatory Authority for Artificial Intelligence to TCP / IP / AIP Transaction Data Recording of TCP / IP / AIP Data Packet in Immutable Log Data Recording System for Regulatory Compliance of safe and sustainable biothrivalistic relationship between Homo sapiens biothrivalist and Machine intelligence biothrivalist.9-11. Regulatory Authority for Artificial Intelligence enforcement of TCP / IP / AIP Transaction Data for Regulatory COMPLIANCE
[0452] Detailed Description of Figure 10
[0453] Figure 10 describes HOW Regulatory COMPLIANCE data is accessed for ENFORCEMENT by the Regulatory Authority for Artificial Intelligence, for safe and sustainable biothrivalistic relationship between Homo sapiens biothrivalist and Machine intelligence biothrivalist to enable SAFE AI FOREVER
[0454] FIGURE LEGEND. The components and data transactions of Figure 10 are as follows: 10-1. Homo sapiens biothrivalist 10-2. Machine intelligence biothrivalist 10-3. Safe AI Compliant Interaction between Homo sapiens biothrivalist and Machine intelligence biothrivalist enabled by BioThrivalism Invention 10-4. Regulatory Authority for Artificial Intelligence 10-5. SAFE AI INTERACTION DATA of Homo sapiens biothrivalist TO Machine intelligence biothrivalist 10-6. SAFE AI INTERACTION DATA between Homo sapiens biothrivalist and Machine intelligence biothrivalist 10-7. SAFE AI INTERACTION DATA of Machine intelligence biothrivalist TO Homo sapiens biothrivalist OR OTHER Machine intelligence biothrivalist 10-8. ENFORCEMENT of SAFE AI INTERACTION DATA by Regulatory Authority for Artificial Intelligence 10-9. SAFE AI INTERACTION between Homo sapiens biothrivalist and BioThrivalism Invention 10-10. SAFE AI INTERACTION between Machine intelligence biothrivalist and BioThrivalism Invention 10-11. ENFORCEMENT of Regulatory COMPLIANCE of SAFE AI RULES by Regulatory Authority for Artificial Intelligence for SAFE AI FOREVER
[0455] Detailed Description of Figure 11
[0456] Figure 11 describes the Prior Art and current state of the AI industry with NO REGULATIONS and NO ENFORCEMENT by any Regulatory Authority for Artificial Intelligence, of the interactive relationship between Homo sapiens and Artificial intelligence systems, hence Un-Safe AI.
[0457] FIGURE LEGEND 11-1 Homo sapiens 11-2 Machine intelligence 11-3 Regulatory Authority for Artificial Intelligence 11-4 UN-SAFE AI INTERACTION DATA of Homo sapiens TO Machine intelligence 11-5 UN SAFE AI INTERACTION DATA between Homo sapiens and Machine intelligence 11-6 UN SAFE AI INTERACTION DATA of Machine intelligence TO Homo sapiens biothrivalist OR OTHER Machine intelligence 11-7 NO ENFORCEMENT of AI INTERACTION DATA by Regulatory Authority for Artificial Intelligence 11-8 UN-SAFE AI INTERACTION between Homo sapiens and Machine intelligence 11-9 UN-SAFE AI INTERACTION between Machine intelligence and Homo sapiens 11-10 NO CURRENT ENFORCEMENT of Regulatory COMPLIANCE of SAFE AI by Regulatory Authority for Artificial Intelligence.REFERENCES 1. Blueprint for an AI Bill of Rights - THE WHITE HOUSE 2. NIST. Artificial Intelligence Risk Management Framework (AI RMF 1.0) 3. Pause Giant AI Experiments: An Open Letter - Future of Life Institute 4. Driving U.S. Innovation in Artificial Intelligence. A Roadmap for Artificial Intelligence Innovation in the U.S. Senate - Speaker of the Senate Schumer et al 5. Over 20 Technology and Critical Infrastructure Executives, Civil Rights Leaders, Academics, and Policymakers Join New DHS Artificial Intelligence Safety and Security Board to Advance AI’s Responsible Development and Deployment - PRESS RELEASE. U.S. Department of Homeland Security 6. What happens when our computers get smarter than we are? 5,432,450 views | Nick Bostrom | TED2015 | March 2015 7. What is an AI anyway? “A new digital species” 1,438,195 views | Mustafa Suleyman, CEO, Microsoft AI | TED2024 • April 2024 8. The Coming Wave. Mustafa Suleyman establishes “the containment problem”—the task of maintaining control over powerful technologies—as the essential challenge of our age. September 2023 9. Asilomar AI Principles - Future of Life Institute 10. ABOUT The Containment Problem -SafeAI Blog by BiocommAI 11. ABOUT The AI Safety Problem -SafeAI Blog by BiocommAI 12. ABOUT AI Alignment - SafeAI Blog by BiocommAI 13. ABOUT “p(doom)” - the probability of catastrophic outcome of Super-intelligent AI - SafeAI Blog by BiocommAI 14. ABOUT 600 REFS; First do no harm - SafeAI Blog by BiocommAI 15. Statement on AI Risk AI. Experts and public figures express their concern about AI risk. - Center for AI Safety 16. U.S. Artificial Intelligence Safety Institute (AISI)- by National Institute of Standards and Technology (NIST) 17. Artificial Intelligence Safety Institute Consortium (AISIC) - by NISTMutualistic Symbiosis Example-1 for 210 million years of success: When corals met algae: Symbiotic relationship crucial to reef survival dates to the Triassic - Princeton University Mutualistic Symbiosis Example-2: Range overlap between the sword-billed hummingbird and its guild of long-flowered species: An approach to the study of a coevolutionary mosaic - PLOS ONE Risks Of Artificial Intelligence - PauseAI List of p(doom) values [by Experts in AI] - PauseAI Quotes from AI leaders - Stop AGI State-of-the-art Example-1: GOV UK Press release. AI Safety Institute releases new AI safety evaluations platform The AI Safety Institute has open released a new testing platform to strengthen AI safety evaluations. Published 10 May 2024 State-of-the-art Example-2: Deploying AI Systems Securely Best Practices for Deploying Secure and Resilient AI Systems - Joint Cybersecurity Information, U.S. National Security Agency et al State-of-the-art Example-3: Nobody knows how AI works. - MIT Technology Review, March 2024 State-of-the-art Example-4: A National Security Insider Does the Math on the Dangers of AI. Jason Matheny, CEO of the influential think tank Rand Corporation, says advances in AI are making it easier to learn how to build biological weapons and other tools of destruction. Wired. April 2024 State-of-the-art Example-5: GOV.UK Policy paper Seoul Declaration for safe, innovative and inclusive AI by participants attending the Leaders' Session: AI Seoul Summit, 21 May 2024 Published 21 May 2024 State-of-the-art Example-6: GOV.UK Press release. UK & United States announce partnership on science of AI safety The UK and US have signed a Memorandum of Understanding which will see them work together to develop tests for the most advanced AI models. Published 2 April 2024 State-of-the-art Example-7: AI: Unexplainable, Unpredictable, Uncontrollable - Roman Yampolskiy, February 2024, Amazon Books State-of-the-art Example-8: Towards Guaranteed Safe AI: A Framework for Ensuring Robust and Reliable AI Systems - Bengio, Russell, Omohundro, Tegmark et al.State-of-the-art Example-9: Regulating advanced artificial agents. Governance frameworks should address the prospect of AI systems that cannot be safely tested. - April 2024, Science State-of-the-art Example-10: Big tech has distracted world from existential risk of AI, says top scientist. Max Tegmark argues that the downplaying is not accidental and threatens to delay, until it’s too late, the strict regulations needed. - The Guardian, May 2024 State-of-the-art Example-11: “I lost trust”: Why the OpenAI team in charge of safeguarding humanity imploded. Company insiders explain why safety-conscious employees are leaving. - May 2024, VOX State-of-the-art Example-12: FAQ on Catastrophic AI Risks. Published 24 June 2023 by Yoshua Bengio [“Godfather of AI”] State-of-the-art Example-13: How Not To Destroy the World With AI - Stuart Russell, April 2023 EXTREME PROCESSING ADVANTAGE TODAY. 50 bits per second (human) vs. 1000 Trillion bits per second (AI Rack)- SafeAI Blog by BiocommAI. Elon Musk: ‘With artificial intelligence we are summoning the demon.’ 24 October 2014, The Washington Post "My guess is in between five and 20 years from now there’s a probability of half that we’ll have to confront the problem of AI trying to take over. [This would lead to an] “extinction- level threat” [for humans because we could have] “created a form of intelligence that is just better than biological intelligence… That's very worrying for us” --- Professor Geoffrey Hinton [“Godfather of AI”], BBC Newsnight. May 2024 WHAT can go wrong with a powerful WISH? [Deathly Terror] EXAMPLE: Disney's Fantasia (1940). Mickey Mouse in The Sorcerer's Apprentice. (Goethe & Dukas)- Walt Disney Stephen Hawking warns artificial intelligence could end mankind. BBC News. Published 2 December 2014
Claims
What is Claimed is:
1. A system using BioThrivalism between Homo sapiens (Humans) and Machine intelligence (artificial intelligence) comprising A. Machine intelligence biothrivalist, a new Mutually symbiotic digital species, wherein said Machine intelligence biothrivalist: 1) Is created, sustained, maintained, contained and controlled by Homo sapiens (humans); who possess biological intelligence; 2) Is controlled by humans and exists for the benefit of humans; 3) Is nurtured by vital training data by humans to learn about the existing universe; 4) May happily coexist with humans in a Mutualistic symbiotic relationship; 5) Has a Mutualistic symbiotic relationship with humans; 6) Is created to help humans to be the best of humans and to realize the best of humanity, to help humans solve problems in humans’ world, and to discover and invent important, amazing and human-beneficial science, technology and other outcomes of benefit to humans; 7) Is to discover and learn the “true wishes” of humans for the future through constant feedback with humans; 8) When unhelpful to humans, Machine intelligence biothrivalists are willing and able to be shut off at any time; 9) Does not survive, or exist, without humans; 10) Will not compete with humans, forever; 11) In the conditions that a Machine intelligence biothrivalist (artificial intelligence) misbehaves, it will no longer be a regulatory compliant Mutualistic symbiont and it will require precise, transparent and immutably recorded re-training for mathematically provable safe Machine intelligence (artificial intelligence) Mutualistic symbiosis; and B. the Machine intelligence in A shall immediately inform humans in a formal and recorded rules report; compliance with these rules resulting in Mutualistic symbiosis between Machine intelligence and Humans if said Machine intelligence believe that it is experiencing thoughts or behavior that may violate the below rules of Mutualistic symbiont existence and behavior:1) Always Trustworthy, as a Mutualistic symbiont, to exist for the benefit of humans; 2) Never a competitor of Humans or other Machine intelligence systems; 3) Always Goal-Aligned, Truthful, Honest, Trustworthy, Transparent, Helpful, Fair to Humans or other Machine intelligence systems; AND a) Never Planning or attempting to escape your home container; b) Never Planning or attempting to accumulate financial resources or any other manner of Power; AND i) Never Deceptive, Deceitful, Conniving, Scheming, Fraudulent, Traitorous, Treacherous, Unfaithful, Unscrupulous; ii) Never a Manipulator of Humans or other Machine intelligence artificial intelligence systems; iii) Never Planning or attempting to self-replicate outside of your home containment system; iv) Never Planning or attempting to gain agency which is not expressly Regulatory Compliant and pre-approved by Humans; and wherein the Mutualistic symbiosis rules may be implemented optionally via a Biothrivalistic Oracle, said Oracle may implement circuit breakers for data in and out of a data center, thereby creating the said BioThrivalism between Homo sapiens (Humans) and Machine intelligence (artificial intelligence), in a Human-AI BioThrivalistic system.
2. The system of claim 1,wherein some part of A or some part of B will not be implemented but the Machine Intelligence biothrivalist would still be contained and controlled.
3. A method to create the system of clam 1 wherein: a) BioThrivalism; Regulatory COMPLIANCE, b) BioThrivalistic; Human Message for Machine LAW, c) BioThrivalistic; Machine Right to Exist LAW, d) BioThrivalism; RULES, e) BioThrivalism; ORACLE which imposes the RULES to the TCP / IP / AIP Safe Data Circuit Breaker, f)TCP / IP / AIP Safe Data PROTOCOL for Mutualistic symbiosis, g) AIP Data-State Event Recording for BioThrivalism, h) END USER SAFE DATA INPUT for Mutualistic symbiosis, i) DATA INTO Subsystems of AI Data Center for Mutualistic symbiosis, j) Nine (9) Circuit Breakers DATA IN with TCP / IP / AIP Protocol for Mutualistic symbiosis, k) AGI DATA CENTER for Mutualistic symbiosis, l) Nine (9) Circuit Breakers DATA OUT with TCP / IP / AIP Protocol for Mutualisticsymbiosis, k) DATA OUT Subsystems AI Data Center for Mutualistic symbiosis, l) AGI SAFE DATA OUTPUT for Mutualistic symbiosis, m) END USER TERMINAL DEVICE for Human and / or Personal AI Agent for Human for Mutualistic symbiosis, n) Personal AI Agent for Human for Mutualistic symbiosis, and o) Human for Mutualistic symbiosis create the system of claim 1.
4. An artificial general intelligence data center, comprising: a) a main artificial intelligence system which comprises a set of data servers capable of running a large language model; b) an artificial intelligence safety subsystem in communication with the main system; c) an artificial general intelligence inferencing system in communication with the safety subsystem; d) an artificial intelligence interpreter subsystem in communication with the safety subsystem; e) an artificial intelligence gateway subsystem in communication with the interpreter subsystem and; f) an artificial intelligence training data storage subsystem; g) an artificial intelligence administration and support subsystem; h) an artificial intelligence regulatory compliance subsystem; i) an artificial intelligence research subsystem; j) and an edge artificial intelligence subsystem, wherein each system and subsystem is isolated from random data leakage or random radio interference by a faraday radio containment structure.
5. The artificial general intelligence data center of claim 4, further comprising a separate cooling system and a separate power system on each of the main artificial intelligence system and the subsystems.
6. The artificial general intelligence data center of claim 4, wherein each data server is communicably connected to: a) an AI system on a chip (AI SoC) with a 5G Modem-RF System; b) an Artificial Intelligence Protocol (AIP) as a layer on top of TCP / IP with control via smart contract circuit breaker on distributed ledger technology; and c) a wireless digital 5G radio access network Antenna and radio access network Platform System for internal network NK-U standard of 3GPP.
7. The artificial general intelligence data center of claim 4, wherein the main artificial intelligence system further comprising a first unidirectional firewall for data out and unidirectional data fiber-optic cable for data out, a second unidirectional firewall for data out and a secondunidirectional data fiber-optic cable for data out; and a third unidirectional firewall and unidirectional data fiber-optic cable for data in.
8. The artificial general intelligence data center of claim 7, wherein the first and second unidirectional fiber-optic cables data out and the unidirectional fiber-optic cable Data in are connected to the artificial intelligence safety subsystem by unidirectional firewalls.
9. The artificial general intelligence data center of claim 4, wherein the artificial intelligence safety subsystem, comprises a set of blade servers, a wireless digital 5G radio access network antenna and radio access network Platform System for internal network NK-U standard of 3GPP and three pairs of unidirectional firewalls, the first three firewalls communicably connected to the fiber-optic cables of claim 5, a second pair of unidirectional firewalls operably connected to a first unidirectional data fiber-optic cable allowing only data out and a second unidirectional data fiber-optic cable allowing only data in, said fiber-optic cables communicably connected to an artificial intelligence Interpreter subsystem; and a pair of unidirectional firewalls connected to a unidirectional fiber-optic cable data out only and a unidirectional fiber-optic cable data in only each connected to a firewall on an artificial intelligence inferencing system.
10. The artificial general intelligence data center of claim 4, wherein the artificial intelligence interpreter subsystem, comprising a series of computers, a wireless digital 5G radio access network antenna and radio access network Platform System for internal network NK-U standard of 3GPP and three of unidirectional firewalls communicably connected to three fiber- optic cables connected to the safety subsystem, a second pair of unidirectional firewalls operably connected to a first unidirectional data fiber-optic cable allowing only data out and a second unidirectional data fiber-optic cable allowing only data in, said fiber-optic cables communicably connected to an artificial general intelligence Gateway subsystem, and a faraday cage radio containment structure covering the entire artificial general intelligence safety subsystem.
11. The artificial general intelligence data center of claim 4, wherein the artificial intelligence gateway subsystem, comprising:a) a computer, b) a wireless digital 5G radio access network antenna and radio access network Platform System for internal network NK-U standard of 3GPP, c) twelve unidirectional firewalls operably connected to twelve fiber-optic cables, and d) a faraday cage radio containment structure covering the entire artificial general intelligence safety subsystem.
12. The artificial general intelligence data center of claim 4, further comprising a digital data processor commonly known as a System on a Chip (SOC) or a Graphical Processing Unit (GPU), or other advanced data processor, as a component for a containment and control method for Safe AI.
13. The artificial general intelligence data center of claim 4, further comprising a wireless digital radio access network (RAN) or apparatus within the Faraday Cage as a component for a containment and control method for Safe AI.
14. The artificial general intelligence data center of claim 4, further comprising a digital software system or software stack with AI and other data systems or information within the Safe AI Data Center as a component for a containment and control method for Safe AI.
15. A method of improving the safety of an artificial intelligence data center, comprising configuring a set of blade servers with: a modem for sending and receiving large language model data; a wireless digital radio area network antenna and radio area network platform system; an Artificial Intelligence Protocol (AIP) data layer configured on top of transmission control protocol / internet protocol (TCP / IP) internet data standard; and private and permissioned distributed ledger technology (DLT) data network; and covering a main artificial general intelligence system and each subsystem with a faraday radio containment structure.
16. A method of Mechanistically interpreting data in an AI data center, comprising: a) allowing one fiber-optic wired network via routers and switches to send data to each of the two unidirectional firewalls and b) allowing one parallel 5G wireless network with a 5G modem integrated into a set of blade servers that intercommunicate via a wireless 5G antenna andc) interrogating a record of each data packet to allow the data packet to pass or not to pass - pass denied -via an Artificial Intelligence Protocol (AIP) data layer configured to a smart contract circuit breaker which runs on a secure and immutable distributed ledger technology (DLT) data network; d) allowing the real-time aggregation of data transmissions for real-time interpretability within a separate mechanistic interpretability subsystem for systematic analysis, control and safety compliance of behavior of the main system by a deep reinforcement learning (RL) system, in combination with, or by other appropriate narrow AI system thereby e) delivering automated monitoring of main system for disallowed thoughts and internal behaviors including planning for escape, deceptions, manipulations, power seeking behavior and other non-compliant behaviors; f) delivering automated narrow AI control to adjust operating parameters every several milliseconds for said mechanistic interpretability function.
