System

Abstract

A system is provided.SOLUTION: A system comprising: means for generating an answer to a question by each of a plurality of generative artificial intelligence models; means for evaluating the answer generated by each generative artificial intelligence model by another generative artificial intelligence model; means for selecting a most appropriate answer based on an evaluation result; and means for outputting the selected answer as a final answer.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The technology of the present disclosure relates to a system. [Background technology]

[0002] Patent document 1 discloses a persona chatbot control method performed by at least one processor, the method including the steps of receiving a user utterance, adding the user utterance to a prompt including an instruction sentence related to a description of the chatbot character, encoding the prompt, and inputting the encoded prompt into a language model to generate a chatbot utterance in response to the user utterance. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2022-180282 Summary of the Invention [Problem to be solved by the invention]

[0004] While the use of generative artificial intelligence (AI) has expanded in recent years, issues such as hallucination and missing information have arisen. Answers generated by these AI systems can sometimes contain incorrect or unreliable information, and this problem is particularly likely to become apparent when generating complex questions or documents with a large amount of information. Therefore, there is a demand for systems that can generate more accurate and reliable answers. [Means for solving the problem]

[0005] The present invention provides a means for multiple generative AI models to generate answers to questions, and for other generative AI models to evaluate the answers generated by each generative AI model. The system includes a means for selecting the most appropriate answer based on the evaluation results and a means for outputting the selected answer as the final answer. This system enables the generation of more accurate and reliable answers through mutual evaluation and discussion between generative AI models. Furthermore, the accuracy of the answers is further improved by introducing a means for updating the answers by holding discussions and adding correction information, and a means for conducting a staged evaluation process using multiple generative AI models and selecting the final answer based on the results of a majority vote.

[0006] A "generative artificial intelligence model" is an artificial intelligence that generates answers to input questions based on a unique algorithm.

[0007] "Evaluation" is the act of other generative AI models judging the generated answer based on criteria such as the accuracy, consistency, and completeness of the information.

[0008] "Evaluation results" are the aggregated results of the ratings given by each generative artificial intelligence model to the answers of other models.

[0009] "Selection" is the act of selecting the most appropriate answer from among multiple answers based on the evaluation results.

[0010] A "final answer" is the final answer provided to the user, determined through an evaluation and selection process.

[0011] "Discussion" is a process in which multiple generative AI models exchange opinions based on each other's answers and add correction information.

[0012] "Correction information" is additional information provided by each model during the discussion process, and is data used to improve the accuracy of the final answer.

[0013] "Majority vote" is a method for determining the answer that receives the most support among multiple answers. [Brief explanation of the drawings]

[0014] [Figure 1] 1 is a conceptual diagram showing an example of the configuration of a data processing system according to a first embodiment. [Figure 2] 1 is a conceptual diagram showing an example of main functions of a data processing device and a smart device according to a first embodiment. [Figure 3] FIG. 10 is a conceptual diagram showing an example of the configuration of a data processing system according to a second embodiment. [Figure 4] FIG. 10 is a conceptual diagram showing an example of main functions of a data processing device and smart glasses according to a second embodiment. [Figure 5] FIG. 10 is a conceptual diagram showing an example of the configuration of a data processing system according to a third embodiment. [Figure 6] FIG. 11 is a conceptual diagram showing an example of main functions of a data processing device and a headset-type terminal according to a third embodiment. [Figure 7] FIG. 10 is a conceptual diagram showing an example of the configuration of a data processing system according to a fourth embodiment. [Figure 8] FIG. 10 is a conceptual diagram showing an example of main functions of a data processing device and a robot according to a fourth embodiment. [Figure 9] 1 shows an emotion map onto which multiple emotions are mapped. [Figure 10] 1 shows an emotion map onto which multiple emotions are mapped. [Figure 11] FIG. 3 is a sequence diagram showing a processing flow of the data processing system according to the first embodiment. [Figure 12] FIG. 10 is a sequence diagram showing the flow of processing in the data processing system in Application Example 1. [Figure 13] FIG. 10 is a sequence diagram showing the flow of processing in the data processing system according to the second embodiment when an emotion engine is combined. [Figure 14] FIG. 10 is a sequence diagram showing the flow of processing in the data processing system in Application Example 2 when an emotion engine is combined. DETAILED DESCRIPTION OF THE INVENTION

[0015] An example of an embodiment of a system according to the technology of the present disclosure will be described below with reference to the accompanying drawings.

[0016] First, the terms used in the following description will be explained.

[0017] In the following embodiments, a coded processor (hereinafter simply referred to as a "processor") may be a single arithmetic device or a combination of multiple arithmetic devices. Furthermore, a processor may be a single type of arithmetic device or a combination of multiple types of arithmetic devices. Examples of arithmetic devices include a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), a GPGPU (General-Purpose computing on Graphics Processing Units), and an APU (Accelerated Processing Unit).

[0018] In the following embodiments, a coded RAM (Random Access Memory) is a memory in which information is temporarily stored and is used as a working memory by a processor.

[0019] In the following embodiments, the coded storage is one or more non-volatile storage devices that store various programs, various parameters, etc. Examples of non-volatile storage devices include flash memory (SSD (Solid State Drive)), magnetic disks (e.g., hard disks), and magnetic tapes.

[0020] In the following embodiments, a communication I / F (Interface) with a symbol is an interface including a communication processor, an antenna, etc. The communication I / F controls communication between multiple computers. Examples of communication standards applied to the communication I / F include wireless communication standards including 5G (5th Generation Mobile Communication System), Wi-Fi (registered trademark), Bluetooth (registered trademark), etc.

[0021] In the following embodiments, "A and / or B" is synonymous with "at least one of A and B." In other words, "A and / or B" means that it may be only A, only B, or a combination of A and B. Furthermore, in this specification, the same concept as "A and / or B" is also applied when three or more things are expressed connected by "and / or."

[0022] [First embodiment]

[0023] FIG. 1 shows an example of the configuration of a data processing system 10 according to the first embodiment.

[0024] 1, a data processing system 10 includes a data processing device 12 and a smart device 14. An example of the data processing device 12 is a server.

[0025] The data processing device 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" according to the technology of the present disclosure. The computer 22 includes a processor 28, a RAM 30, and a storage 32. The processor 28, the RAM 30, and the storage 32 are connected to a bus 34. The database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. Examples of the network 54 include a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0026] The smart device 14 includes a computer 36, a reception device 38, an output device 40, a camera 42, and a communication I / F 44. The computer 36 includes a processor 46, a RAM 48, and a storage 50. The processor 46, the RAM 48, and the storage 50 are connected to a bus 52. The reception device 38, the output device 40, and the camera 42 are also connected to the bus 52.

[0027] The reception device 38 includes a touch panel 38A, a microphone 38B, and the like, and receives user input. The touch panel 38A detects contact with an indicator (for example, a pen or a finger) to receive user input by the touch of the indicator. The microphone 38B detects the user's voice to receive user input by voice. The control unit 46A transmits data indicating the user input received by the touch panel 38A and the microphone 38B to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the data indicating the user input.

[0028] The output device 40 includes a display 40A and a speaker 40B, and presents data to the user 20 by outputting the data in a form of expression that the user 20 can perceive (for example, audio and / or text). The display 40A displays visible information such as text and images in accordance with instructions from the processor 46. The speaker 40B outputs audio in accordance with instructions from the processor 46. The camera 42 is a compact digital camera equipped with an optical system including a lens, aperture, and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor.

[0029] The communication I / F 44 is connected to a network 54. The communication I / Fs 44 and 26 control the exchange of various information between the processor 46 and the processor 28 via the network 54.

[0030] FIG. 2 shows an example of the main functions of the data processing device 12 and the smart device 14.

[0031] 2, in the data processing device 12, a specific process is performed by the processor 28. A specific processing program 56 is stored in the storage 32. The specific processing program 56 is an example of a "program" according to the technology of the present disclosure. The processor 28 reads the specific processing program 56 from the storage 32 and executes the read specific processing program 56 on the RAM 30. The specific process is realized by the processor 28 operating as a specific processing unit 290 in accordance with the specific processing program 56 executed on the RAM 30.

[0032] The storage 32 stores a data generation model 58 and an emotion identification model 59. The data generation model 58 and the emotion identification model 59 are used by the identification processing unit 290.

[0033] In the smart device 14, the processor 46 performs the reception output process. The storage 50 stores a reception output program 60. The reception output program 60 is used in conjunction with the specific processing program 56 by the data processing system 10. The processor 46 reads the reception output program 60 from the storage 50 and executes the read reception output program 60 on the RAM 48. The reception output process is realized by the processor 46 operating as the control unit 46A in accordance with the reception output program 60 executed on the RAM 48.

[0034] Next, a description will be given of the specific processing performed by the specific processing unit 290 of the data processing device 12. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0035] This invention relates to a generative AI platform for improving the accuracy of answers generated by AI. This platform is a system in which multiple generative AI models cooperate to generate the most correct answers through mutual evaluation and discussion. Below, we will explain the program processing of this system in natural language using concrete examples.

[0036] Program processing explanation

[0037] Enter your question

[0038] A user uses an email client to type a question, for example, "What's the weather like in Tokyo tomorrow?"

[0039] Generate answers

[0040] The server receives this question and sends it to several different generative artificial intelligence models (AIM1, AIM2, AIM3, etc.), each of which uses its own algorithm to generate an answer.

[0041] AIM1: "It's sunny."

[0042] AIM2: "Tomorrow will be sunny."

[0043] AIM3: "It's forecast to be sunny in Tokyo."

[0044] Request for evaluation of answer

[0045] The server sends the generated answers to other generative AI models in the Validator Pool, allowing each model to evaluate the answers of the other generative AI models.

[0046] Evaluating answers

[0047] Each generative AI model evaluates the answers generated by other models and assigns them a score based on the accuracy, consistency, and completeness of the answers.

[0048] Aggregation of evaluation results

[0049] The server aggregates the results from each model and sees which answer received the highest rating.

[0050] Notification of evaluation results and request for discussion

[0051] The server notifies each generative AI model of the evaluation results and requests them to correct or supplement their answers as necessary.

[0052] Editing and discussing answers

[0053] Each model modifies its answer based on the discussion and adds any necessary corrections. For example, AIM1 can update its answer as follows:

[0054] AIM1: "Tomorrow will be sunny in Tokyo, with temperatures around 25 degrees."

[0055] AIM2: "Tomorrow in Tokyo will be sunny with a maximum temperature of 25 degrees."

[0056] AIM3: "It'll be sunny and 25 degrees."

[0057] Final evaluation and majority vote

[0058] The server then re-evaluates the updated answer against the models in the Validator Pool. Each model selects the most appropriate answer and a majority vote is held.

[0059] Selection of the final answer

[0060] Based on the evaluation results, the server selects the final answer. The final answer obtained through this process is "Tomorrow in Tokyo, it will be sunny and the temperature will be around 25 degrees."

[0061] Submit your answer

[0062] The server sends the final selected answer to the terminal.

[0063] Show Answers

[0064] The terminal displays the final answer to the user.

[0065] This system allows multiple generative AI models to cooperate and generate more accurate and reliable answers through mutual evaluation and discussion. As can be seen from this explanation with concrete examples, the present invention provides an effective means for suppressing hallucination and improving the accuracy of information.

[0066] The processing flow will be explained below.

[0067] Step 1:

[0068] The user inputs a question, for example, "What's the weather like in Tokyo tomorrow?" into the terminal.

[0069] Step 2:

[0070] The terminal transmits the entered question to the server.

[0071] Step 3:

[0072] The server receives questions and distributes them to multiple generative artificial intelligence models (AIM1, AIM2, AIM3).

[0073] Step 4:

[0074] Each generative AI model generates an answer to a question sent by the server using its own algorithm. For example,

[0075] AIM1: "It's sunny."

[0076] AIM2: "Tomorrow will be sunny."

[0077] AIM3: "It's forecast to be sunny in Tokyo."

[0078] Step 5:

[0079] The server sends the generated answer to each of the other generative AI models in the Validator Pool for evaluation.

[0080] Step 6:

[0081] Each generative AI model evaluates the answers it receives from other models and assigns a score based on the accuracy, consistency, and completeness of the answers.

[0082] Step 7:

[0083] The server aggregates the evaluation results from each model and analyzes which answer received the highest evaluation.

[0084] Step 8:

[0085] The server notifies each generative AI model of the evaluation results and, if necessary, requests them to correct or supplement their answers.

[0086] Step 9:

[0087] Each generative AI model modifies the answer based on the evaluation results and adds correction information. For example,

[0088] AIM1: "Tomorrow will be sunny in Tokyo, with temperatures around 25 degrees."

[0089] AIM2: "Tomorrow in Tokyo will be sunny with a maximum temperature of 25 degrees."

[0090] AIM3: "It'll be sunny and the temperature will be 25 degrees."

[0091] Step 10:

[0092] The server then sends the updated answers to the Validator Pool again, asking each model to re-evaluate.

[0093] Step 11:

[0094] Each generative AI model then evaluates the answers again, selects the most appropriate answer, and assigns it a score.

[0095] Step 12:

[0096] The server tally the results of the reevaluation and selects the answer with the highest score. For example, "Tomorrow in Tokyo, it will be sunny and the temperature will be around 25 degrees."

[0097] Step 13:

[0098] The server sends the final selected answer to the terminal.

[0099] Step 14:

[0100] The terminal displays the final answer to the user.

[0101] Example 1

[0102] Next, a description will be given of Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0103] In existing generative AI models, relying on a single model's answer can reduce the accuracy and reliability of the answer. Furthermore, when the answers generated by each model differ, it is difficult to determine which answer is the most accurate. There is a need to solve this problem and provide more accurate and reliable answers.

[0104] The specific processing by the specific processing unit 290 of the data processing device 12 in the first embodiment is realized by the following means.

[0105] In this invention, the server includes means for inputting a question, means for multiple generative AI models to each generate an answer to the question, means for other generative AI models to evaluate the answers generated by each generative AI model, means for selecting the most appropriate answer based on the evaluation results, means for correcting or supplementing the selected answer, means for re-evaluating the corrected answer, means for performing the evaluation process in stages and selecting a final answer based on the result of majority voting, and means for transmitting and displaying the selected final answer to a user terminal. This makes it possible to provide users with more accurate and reliable final answers through mutual evaluation and discussion among multiple generative AI models.

[0106] The "means for inputting a question" provides an interface for a user to input a question to the system.

[0107] A "generative artificial intelligence model" refers to an algorithm or machine learning model that generates answers in natural language to input questions.

[0108] The "means for generating an answer" is what enables the generative artificial intelligence model to generate an answer to an input question.

[0109] "Means for evaluating answers" refers to a function in which answers generated by a generative AI model are evaluated by other generative AI models and scored for accuracy and consistency.

[0110] The "means for selecting the most appropriate answer" includes a function for selecting the answer that has received the highest evaluation based on the evaluation results of each generative artificial intelligence model.

[0111] "Means for correcting or supplementing answers" refers to the functionality for adding necessary corrections or additional information to selected answers.

[0112] "Means for re-evaluation" refers to a function that allows each generative artificial intelligence model to re-evaluate an answer that has been corrected or supplemented.

[0113] "Means for carrying out the evaluation process in stages" includes a function for dividing the evaluation into multiple stages and carrying out a final evaluation based on the evaluation results at each stage.

[0114] "Means for selecting a final answer based on the results of a majority vote" refers to a function for selecting the answer that receives the most support as the final answer based on the evaluation results of each generative AI model.

[0115] The "means for transmitting and displaying the final answer" provides a function for transmitting the selected final answer to the user terminal and displaying it.

[0116] This invention relates to a generative AI platform for improving the accuracy of answers generated by AI. This platform is a system in which multiple generative artificial intelligence models (generative AI models) cooperate to generate the most correct answers through mutual evaluation and discussion.

[0117] Hardware and software used

[0118] The hardware required to implement this system includes a high-performance server and user terminals (e.g., PCs and smartphones). The server has high computing power and storage capacity, and a cloud platform for running the generative AI models can also be used. The software includes each generative AI model, an email client, a web form, etc.

[0119] Data processing and calculation

[0120] When a user enters a question using an email client or web form, the question is stored in a database and sent to a server, which parses the question and converts it into a prompt that the generative AI model can understand.

[0121] For example, in response to the question "What's the weather like in Tokyo tomorrow?", the server generates a prompt sentence, "Please tell me the weather forecast for Tokyo tomorrow," and sends this to multiple generative AI models (AIM1, AIM2, AIM3, etc.).

[0122] Each generative AI model receives a prompt and uses its own algorithms and datasets to generate an answer.

[0123] AIM1: "It's sunny."

[0124] AIM2: "Tomorrow will be sunny."

[0125] AIM3: "It's forecast to be sunny in Tokyo."

[0126] The generated answers are sent by the server to other generative AI models in the Validator Pool, where each model evaluates the answers of the other models based on the accuracy, consistency, and completeness of the answers.

[0127] The evaluation results are tallied, and the server checks which answer received the highest rating. For example, if AIM1 received the highest rating, that score is recorded. The evaluation results are notified to each generative AI model, which corrects or supplements the answer as necessary. For example, if AIM1 is evaluated as having insufficient temperature information, additional information is provided.

[0128] Each generative AI model modifies its answer based on the discussion and adds any necessary corrections. Answers may be updated as follows:

[0129] AIM1: "Tomorrow will be sunny in Tokyo, with temperatures around 25 degrees."

[0130] AIM2: "Tomorrow in Tokyo will be sunny with a maximum temperature of 25 degrees."

[0131] AIM3: "It'll be sunny and 25 degrees."

[0132] The server then evaluates the updated answers again using the models in the Validator Pool, and each generative AI model selects the most appropriate answer through a majority vote. Based on the evaluation results, the server selects the final answer: "Tomorrow in Tokyo will be sunny, with a temperature of around 25 degrees."

[0133] Examples of specific examples and prompts

[0134] As a specific example, the following shows how a generative AI model generates and evaluates answers to the input, "What's the weather like in Tokyo tomorrow?"

[0135] User input: "What's the weather like in Tokyo tomorrow?"

[0136] Server prompt: "What's the weather forecast for Tokyo tomorrow?"

[0137] Example answer from a generative AI model:

[0138] AIM1: "It's sunny."

[0139] AIM2: "Tomorrow will be sunny."

[0140] AIM3: "It's forecast to be sunny in Tokyo."

[0141] Example of an answer after evaluation and revision:

[0142] AIM1: "Tomorrow will be sunny in Tokyo, with temperatures around 25 degrees."

[0143] AIM2: "Tomorrow in Tokyo will be sunny with a maximum temperature of 25 degrees."

[0144] AIM3: "It'll be sunny and 25 degrees."

[0145] Ultimately, the answer selected by the server based on the majority vote was: "Tomorrow in Tokyo, it will be sunny, and the temperature will be around 25 degrees."

[0146] Through the collaboration and mutual evaluation of multiple generative AI models, this system enables the provision of more accurate and reliable information to users.

[0147] The flow of the identification process in the first embodiment will be described with reference to FIG.

[0148] Step 1:

[0149] A user enters a question using an email client or web form, for example, "What's the weather like in Tokyo tomorrow?" This input is sent to the server.

[0150] Input: User question: "What's the weather like in Tokyo tomorrow?"

[0151] Output: Query data sent to the server

[0152] Step 2:

[0153] The server analyzes the received question and converts it into a prompt that the generative AI model can understand. For example, it generates a prompt such as, "Please tell me the weather forecast for Tokyo tomorrow."

[0154] Input: Question data "What's the weather like in Tokyo tomorrow?"

[0155] Output: Prompt "What is the weather forecast for Tokyo tomorrow?"

[0156] Step 3:

[0157] The server sends the generated prompt to multiple generative AI models (AIM1, AIM2, AIM3, etc.), each of which receives the prompt and generates an answer using its own algorithm and dataset.

[0158] Input: Prompt "What is the weather forecast for Tokyo tomorrow?"

[0159] Output: Answer of each generative AI model (e.g., AIM1: "It's sunny," AIM2: "It will be sunny tomorrow," AIM3: "It is forecast to be sunny in Tokyo.")

[0160] Step 4:

[0161] The server sends the generated answers to other generative AI models in the Validator Pool, asking each model to evaluate the answers of the other models.

[0162] Input: The generative AI model's answer

[0163] Output: Evaluation request data

[0164] Step 5:

[0165] Generative AI models evaluate the answers generated by other generative AI models. For example, AIM2 looks at the answers of AIM1 and AIM3 and assigns a score to each. The evaluation criteria are the accuracy, consistency, and completeness of the information.

[0166] Input: Answers from other generative AI models

[0167] Output: Evaluation score (e.g., AIM2 scoring)

[0168] Step 6:

[0169] The server aggregates the results from each model and checks which answer received the highest score. For example, if AIM1 received the highest score, that score is recorded.

[0170] Input: Rating score

[0171] Output: The answer from the model with the highest score

[0172] Step 7:

[0173] The server notifies each generative AI model of the evaluation results and asks them to correct or supplement their answers as necessary. For example, if AIM1 receives an evaluation that "temperature information is insufficient," it will ask them to provide additional information.

[0174] Input: Evaluation result

[0175] Output: Notification of correction request

[0176] Step 8:

[0177] The generative AI model then modifies the answer based on the discussion and adds any necessary corrections. For example, AIM1 updates the answer to "Tomorrow will be sunny in Tokyo, with the temperature around 25 degrees."

[0178] Input: Notification of correction request

[0179] Output: Updated answer (e.g. AIM1: "Tomorrow will be sunny in Tokyo, with the temperature around 25 degrees.")

[0180] Step 9:

[0181] The server then re-evaluates the updated answers with the models in the Validator Pool. Each generative AI model selects the most appropriate answer and a majority vote is held.

[0182] Input: Updated answer

[0183] Output: Reassessment score and majority vote result

[0184] Step 10:

[0185] The server selects the final answer based on the evaluation results. For example, it selects "Tomorrow in Tokyo will be sunny, with the temperature around 25 degrees" as the most popular answer.

[0186] Input: Reassessment score and majority vote result

[0187] Output: Final answer "Tomorrow in Tokyo it will be sunny, with the temperature around 25 degrees."

[0188] Step 11:

[0189] The server sends the final selected answer to the terminal.

[0190] Input: Final answer

[0191] Output: Data sent to the terminal

[0192] Step 12:

[0193] The terminal displays the final answer to the user, who can view it through their email client or web interface.

[0194] Input: Final response data

[0195] Output: What is displayed to the user

[0196] This series of processes enables the provision of more accurate and reliable information to users through cooperation and mutual evaluation of multiple generative AI models.

[0197] (Application example 1)

[0198] Next, a description will be given of Application Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0199] Conventional answer generation systems using generative AI models often lack consistency and reliability, with the accuracy of each model varying. Furthermore, especially in the security field, where incorrect answers can have significant impacts, higher accuracy and reliability are required. For this reason, there was a need to develop a system in which multiple AI models cooperate and generate the most accurate and reliable answers through mutual evaluation and discussion.

[0200] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 1 is realized by the following means.

[0201] In this invention, the server includes means for multiple generative AI models to generate answers to questions, means for other generative AI models to evaluate the answers generated by each generative AI model, and means for selecting the most appropriate answer based on the evaluation results. This improves the accuracy and reliability of the generated answers, enabling reliable notifications to users, particularly when used as security alerts.

[0202] "Multiple generative artificial intelligence models" refers to multiple AI models that use different algorithms and datasets to generate answers in their own unique way.

[0203] "Means for generating answers to each question" refers to the process or function by which each AI model generates an individual answer to a question input by a user.

[0204] "Means for evaluation by other generative artificial intelligence models" refers to the process or function by which answers generated by one AI model are evaluated by other AI models to determine their accuracy and consistency.

[0205] "Means for selecting the most appropriate answer" refers to the process or function of selecting the answer that received the highest rating based on the results of multiple evaluations.

[0206] "Means for outputting the selected answer as the final answer" refers to the process or function of providing the selected most appropriate answer to the user as the final answer.

[0207] "Means for holding discussions, adding corrections, and updating answers" refers to the process or function of discussing the answers generated by each AI model and making necessary corrections to generate more accurate answers.

[0208] "A method of conducting a step-by-step evaluation process and selecting a final answer based on the results of a majority vote" refers to a process of going through a series of evaluation and selection steps in a step-by-step manner, ultimately selecting the answer supported by the majority of evaluators.

[0209] A "security alert" refers to a warning message that notifies the user of an abnormal security situation or abnormality.

[0210] "Means of notifying the user" refers to the process or function of sending the final selected answer or alert to the user's device and notifying the user by visual, auditory, or other means.

[0211] This invention relates to a system in which multiple generative artificial intelligence models cooperate to generate the most correct answer through mutual evaluation and discussion. Specifically, this invention describes an application for providing accurate alerts to users in security services.

[0212] Hardware and software used

[0213] Hardware:

[0214] 1. Server: Runs multiple AI models and manages the evaluation process.

[0215] 2. Security cameras: collect data and provide it for analysis.

[0216] 3. Smartphone: The device where the user enters the question and receives the final answer.

[0217] software:

[0218] 1. Python: A system-wide programming language.

[0219] 2. Requests library: Manages API communication.

[0220] 3. AI Models: Multiple generative AI models that use various algorithms to generate solutions to problems.

[0221] System processing flow

[0222] 1. User enters question:

[0223] Users use a smartphone app to enter specific questions, such as "Can you check if there's an intruder in my office?"

[0224] 2. The server receives the query:

[0225] The server receives this question and sends it to multiple generative AI models, each of which uses its own algorithm to generate an answer.

[0226] 3. The AI ​​model generates the answer:

[0227] Each generative AI model generates an answer to the question, for example:

[0228] Model 1: "Motion detected on camera, but it doesn't seem to be an intruder."

[0229] Model 2: "It looks like the door is open, but we need to check it out immediately."

[0230] Model 3: "Motion detected, but no further information available."

[0231] 4. Other models rate your answers:

[0232] The server sends these answers to other generative AI models, which then evaluate the answers of the other models based on the accuracy and consistency of the answers.

[0233] 5. Compilation of evaluation results and selection of best answers:

[0234] Based on the evaluation results of each AI model, the answer with the highest score will be selected. For example, the evaluation will be as follows:

[0235] "Motion detected on camera, but it doesn't appear to be an intruder" received the highest rating.

[0236] 6. Generate final answer and notify user:

[0237] The server then sends the final selected answer to the user's device and notifies them, allowing the user to receive accurate and reliable security alerts.

[0238] Prompt Sentence Examples

[0239] User input:

[0240] "Check if there's an intruder in the office."

[0241] This invention enables multiple generative AI models to cooperate to generate the most appropriate response and notify the user of it as a security alert, allowing the user to take appropriate action based on highly reliable information.

[0242] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[0243] Step 1:

[0244] A user uses a smartphone to input a question, for example, "Please check if there is an intruder in my office." This input is sent to the server through the user interface.

[0245] Step 2:

[0246] The server receives a user's question and sends it to multiple generative AI models. Each model receives the question and generates an answer using its own algorithm. A typical answer might be, "Motion was detected on the camera, but it doesn't appear to be an intruder."

[0247] Step 3:

[0248] The server sends the answers generated by each generative AI model to other generative AI models for evaluation. At this time, the input provided to each model is a list of generated answers, and each model evaluates this list.

[0249] Step 4:

[0250] Each generative AI model evaluates the answers generated by other models and assigns a score based on their accuracy, consistency, and completeness of information. The evaluation results are sent to the server in the form of a score. For example, model A may assign 90 points to the answer of model B and 80 points to the answer of model C.

[0251] Step 5:

[0252] The server then selects the answer with the highest score based on the aggregated evaluation results. For example, if "Motion was detected on the camera, but it doesn't appear to be an intruder" receives the highest score, it will be selected.

[0253] Step 6:

[0254] The server then sends the selected answer to the user's smartphone. This notification includes the most highly rated answer and a brief explanation of it. The user can then check the answer on their device.

[0255] Step 7:

[0256] If necessary, the generated answers may be further discussed and corrections added. Each generative AI model will be discussed again and generate a new answer with the necessary corrections. This process of evaluation and correction is repeated until the revised best answer is selected.

[0257] The above is a specific flow of processing by the program that realizes this invention, which enables users to receive accurate and reliable security alerts.

[0258] Furthermore, an emotion engine that estimates the user's emotion may be combined. That is, the identification processing unit 290 may estimate the user's emotion using the emotion identification model 59 and perform identification processing using the user's emotion.

[0259] This invention relates to a generative AI platform that uses multiple generative AI models and an emotion engine to generate accurate and appropriate answers to user questions. Each generative AI model generates and modifies answers through mutual evaluation and discussion, and by combining it with an emotion engine that recognizes the user's emotions, it provides answers that are adapted to the user's emotional state. Below, we will explain the program processing of this system in natural language using concrete examples.

[0260] Program processing explanation

[0261] Enter your question

[0262] The user uses the terminal to input a question, for example, "Tell me what my friends think."

[0263] Emotion recognition

[0264] The device sends the user's input, voice, facial expression, and other data to the emotion engine, which analyzes this data and recognizes the user's emotions. For example, it can detect "anxiety."

[0265] Submit a Question

[0266] The device sends the question with the recognized emotion data to the server.

[0267] Generate answers

[0268] The server distributes the emotional data attached to the question to multiple generative AI models (AIM1, AIM2, AIM3). Each model takes the emotional data into account and generates an answer using its own algorithm. For example,

[0269] AIM1: "Your friends care about you."

[0270] AIM2: "Your friends are worried about you."

[0271] AIM3: "Your friends trust you."

[0272] Request for evaluation of answer

[0273] The server sends the generated answers to other generative AI models, allowing each model to evaluate the answers of the other generative AI models.

[0274] Evaluating answers

[0275] The generative AI model evaluates the answers generated by other models and assigns them a score based on the answer's accuracy, consistency, completeness of information, and fit to the perceived emotion.

[0276] Aggregation of evaluation results

[0277] The server aggregates the evaluation results from each model and analyzes which answer received the highest evaluation.

[0278] Notification of evaluation results and request for discussion

[0279] The server notifies each generative AI model of the evaluation results and requests them to revise or supplement their answers if necessary. Emotional data is also reflected.

[0280] Editing and discussing answers

[0281] Each generative AI model modifies its answers based on the evaluation results and emotional data, adding correction information. For example,

[0282] AIM1: "Friends care about you and always want to support you."

[0283] AIM2: "Your friend is very concerned about you and wants to talk more."

[0284] AIM3: "Your friends trust you and want to talk to you if they have a problem."

[0285] Final evaluation and majority vote

[0286] The server re-evaluates the updated answer and asks each model to re-evaluate.

[0287] Selection of the final answer

[0288] Based on the reevaluation results, the server selects a final answer, such as "Your friends care about you and always want to support you."

[0289] Submit your answer

[0290] The server sends the final selected answer to the terminal.

[0291] Show Answers

[0292] The terminal displays the final answer to the user.

[0293] This system enables multiple generative AI models to generate more accurate and reliable answers that adapt to the user's emotional state through mutual evaluation and discussion. As can be seen from the explanation using concrete examples, the present invention provides an innovative means for suppressing hallucination and providing information that takes into consideration the user's emotions.

[0294] The processing flow will be explained below.

[0295] Step 1:

[0296] The user inputs a question, for example, "What's the weather like in Tokyo tomorrow?" into the terminal.

[0297] Step 2:

[0298] The device sends the input question as data to the emotion engine, which analyzes the user's input, voice, facial expressions, and other data to recognize the user's emotions. For example, it can detect "anxiety."

[0299] Step 3:

[0300] The emotion engine returns the recognized emotion data to the terminal.

[0301] Step 4:

[0302] The device sends the question with the emotional data attached to it to the server.

[0303] Step 5:

[0304] The server receives questions and emotion data and distributes them to multiple generative AI models (AIM1, AIM2, AIM3).

[0305] Step 6:

[0306] Each generative AI model uses its own algorithm to generate answers to questions while taking into account emotional data. For example,

[0307] AIM1: "It's sunny."

[0308] AIM2: "Tomorrow will be sunny."

[0309] AIM3: "It's forecast to be sunny in Tokyo."

[0310] Step 7:

[0311] The server compiles the generated answers and sends them to other generative AI models, allowing each model to evaluate the answers of the other generative AI models.

[0312] Step 8:

[0313] Each generative AI model evaluates the answers sent by other generative AI models and assigns a score based on the answer's accuracy, consistency, completeness of information, and fit to the perceived emotion.

[0314] Step 9:

[0315] The server aggregates the evaluation results from each model and analyzes which answer received the highest evaluation.

[0316] Step 10:

[0317] The server notifies each generative AI model of the evaluation results and requests them to revise or supplement their answers as necessary. This emotional data is also reflected in the re-evaluation.

[0318] Step 11:

[0319] Each generative AI model modifies its answers based on the evaluation results and emotional data, adding correction information. For example,

[0320] AIM1: "Tomorrow will be sunny in Tokyo. The temperature will be around 25 degrees."

[0321] AIM2: "Tomorrow in Tokyo will be sunny with a maximum temperature of 25 degrees."

[0322] AIM3: "It'll be sunny and the temperature will be 25 degrees."

[0323] Step 12:

[0324] The server then distributes the updated answers to each generative AI model again and requests them to evaluate again.

[0325] Step 13:

[0326] Each generative AI model then evaluates the answers again, selects the most appropriate answer, and assigns it a score.

[0327] Step 14:

[0328] The server tally the results of the reevaluation and selects the answer with the highest score. For example, "Tomorrow in Tokyo, it will be sunny and the temperature will be around 25 degrees."

[0329] Step 15:

[0330] The server sends the final selected answer to the terminal.

[0331] Step 16:

[0332] The terminal displays the final answer to the user.

[0333] Example 2

[0334] Next, a description will be given of Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0335] In conventional generative AI systems, the answers generated are not always adapted to the user's emotional state, which reduces user satisfaction. Furthermore, the evaluation process for multiple generative AI models is insufficient, resulting in answers that lack consistency and accuracy. Furthermore, the generated answers do not reflect correction information or discussions, which can result in the delivery of less than optimal answers.

[0336] The specific processing by the specific processing unit 290 of the data processing device 12 in the second embodiment is realized by the following means.

[0337] In this invention, the server includes means for multiple generative AI models to generate answers to questions, means for other generative AI models to evaluate the answers generated by each generative AI model, means for selecting the most appropriate answer based on the evaluation results, an emotion engine that recognizes the user's emotion using data input by the user, and means for attaching the emotion data recognized by the emotion engine to the question and transmitting it, thereby making it possible to generate and provide accurate and appropriate answers that are adapted to the user's emotional state.

[0338] "Multiple generative AI models" are multiple AI systems with different algorithms and settings, each capable of generating answers independently.

[0339] A "question" is any information or question entered by a user that requires an answer from a generative AI model.

[0340] The "means for generating an answer" refers to the process or algorithm that the generative artificial intelligence model uses to create an appropriate answer to a user's question.

[0341] "Means for evaluation by other generative artificial intelligence models" refers to the process by which each generative artificial intelligence model evaluates answers generated by other models and assigns them a score or ranking.

[0342] The "means of selecting the most appropriate answer based on the evaluation results" is a system that aggregates the evaluation results obtained from multiple generative AI models and selects the answer that receives the highest evaluation.

[0343] An "emotion engine" is a system that analyzes the user's input data, voice, facial expressions, etc. to recognize the user's emotional state.

[0344] "Emotion data" is data showing the results of analysis by the user's emotion engine, and represents an emotional state such as anxiety or relief.

[0345] The "means of transmission" is the process or method of transmitting data from the device to the server, often using a secure communications protocol.

[0346] "Correction information" is information that is added to the initially generated answer, and is used to improve the accuracy and appropriateness of the answer.

[0347] "Means for discussion" refers to a process in which multiple generative AI models discuss the generated answers and make optimal improvements or corrections.

[0348] "Majority vote" is a method of statistically aggregating the results of multiple evaluations and scoring and selecting the answer that receives the most support.

[0349] The system starts when a user uses a device to input a question. For example, "Tell me what my friend thinks." The user can use a keyboard or touch screen to input a question. The system may also collect voice and facial expression data, which are sent to the emotion engine.

[0350] The emotion engine analyzes the collected data and recognizes the user's emotions. Voice recognition technology and facial expression analysis algorithms are used to recognize emotions, such as "anxiety." The emotion data recognized in this way is attached to the user's question and sent from the device to the server. A secure communication protocol (e.g., HTTPS) is used to protect the data.

[0351] The server distributes the received question and emotion data to multiple generative AI models (e.g., AIM1, AIM2, AIM3). Each model generates an answer using its own natural language processing algorithm. Examples of generated answers include:

[0352] AIM1: "Your friends care about you."

[0353] AIM2: "Your friends are worried about you."

[0354] AIM3: "Your friends trust you."

[0355] The server then sends the answers generated by each generative AI model to other generative AI models, who evaluate each other's answers based on the accuracy, consistency, completeness of the information, and suitability to the emotional data. For example, AIM1 might give 70 points to AIM2's answer and 80 points to AIM3's answer.

[0356] The server aggregates these evaluation results and selects the answer with the highest rating. After selection, the server notifies each generative AI model of the evaluation results and requests them to revise or supplement the answer as necessary. Emotional data is also reflected.

[0357] Each generative AI model modifies its answers based on the evaluation results and emotional data, adding correction information. For example, it may be modified as follows:

[0358] AIM1: "Friends care about you and always want to support you."

[0359] AIM2: "Your friend is very concerned about you and wants to talk more."

[0360] AIM3: "Your friends trust you and want to talk to you if they have a problem."

[0361] Finally, the server re-evaluates the updated answers and selects the answer with the most votes. The final answer is sent to the device and displayed to the user, providing a reliable answer to the question.

[0362] For example, consider the following prompt:

[0363] "Please tell me what my friend thinks. I'm feeling anxious right now."

[0364] This system enables multiple generative AI models to provide more accurate and reliable answers that adapt to the user's emotional state through mutual evaluation and discussion, thereby improving user satisfaction and the quality of information provided.

[0365] The flow of the identification process in the second embodiment will be described with reference to FIG.

[0366] Step 1: The user enters a question using a terminal.

[0367] Input: Text data entered by the user using a keyboard or touchscreen.

[0368] Output: Question text data imported to the terminal.

[0369] Specific operation: The user inputs a question such as "Please tell me what my friends think." The device collects the text data of the question.

[0370] Step 2: The device sends the user's input data to the emotion engine and recognizes the user's emotion.

[0371] Input: Text data of the user's question acquired by the device, as well as voice and facial expression data.

[0372] Output: Emotion data parsed by the emotion engine.

[0373] Specific operation: The device sends the user's text data, voice, and facial expressions to the emotion engine, which analyzes them and recognizes emotions such as "anxiety."

[0374] Step 3: The device sends the question accompanied by the recognized emotion data to the server.

[0375] Input: User question text data and sentiment data.

[0376] Output: Questions and sentiment data sent to the server.

[0377] Specific operation: The device attaches emotional data obtained from the emotion engine to the question and sends it securely to the server using HTTPS.

[0378] Step 4: The server distributes the questions and emotion data to multiple generative AI models.

[0379] Input: Question and emotion data received by the server.

[0380] Output: The questions and sentiment data delivered to each generative AI model.

[0381] Specific operation: The server distributes the received question and emotion data to the generative AI models (e.g., AIM1, AIM2, AIM3), and each model generates an answer.

[0382] Step 5: Each generative AI model generates an answer to the question.

[0383] Input: Questions and sentiment data delivered to each generative AI model.

[0384] Output: The generated answer.

[0385] How it works: The generative AI model uses a proprietary natural language processing algorithm to generate answers. For example, AIM1 generates "Your friends care about you."

[0386] Step 6: The server sends the generated answer to other generative AI models for evaluation.

[0387] Input: The answers generated by each generative AI model.

[0388] Output: Answer evaluation request sent to other generative AI models.

[0389] Specific operation: The server collects the answers generated by each model and sends them to other generative AI models for evaluation.

[0390] Step 7: Each generative AI model evaluates the answers generated by other models and assigns a score.

[0391] Input: Answers received from other generative artificial intelligence models.

[0392] Output: Evaluation score.

[0393] How it works: Each generative AI model evaluates the answers of other models based on specified criteria and assigns a score based on, for example, accuracy or consistency.

[0394] Step 8: The server aggregates the evaluation results of each generative AI model and selects the answer with the highest evaluation.

[0395] Input: Evaluation scores from all generative AI models.

[0396] Output: The answer that received the most votes.

[0397] Specific operation: The server aggregates the evaluation results and selects the answer with the highest score.

[0398] Step 9: The server notifies each generative AI model of the selected answers and requests corrections or supplements if necessary.

[0399] Input: Screened answers and assessment results.

[0400] Output: Request for corrections and clarifications.

[0401] Specific operation: The server sends the selected answers and their evaluation scores to the model, requesting corrections or supplements as necessary.

[0402] Step 10: Each generative AI model modifies the answer and adds correction information.

[0403] Input: Correction request and original answer from the server.

[0404] Output: The revised answer and any corrections added.

[0405] Specific behavior: Each model modifies the answer based on the evaluation results, adding correction information such as, "Your friends care about you and always want to support you."

[0406] Step 11: The server re-evaluates the revised answers and selects the final answer based on the majority vote.

[0407] Input: Revised answer.

[0408] Output: The final selected answer.

[0409] Specific operation: The server has each generative AI model evaluate the revised answer again, and selects the final answer by majority vote based on the evaluation results.

[0410] Step 12: The server sends the final selected answer to the terminal and displays it to the user.

[0411] Input: Final selected answer.

[0412] Output: The final answer displayed on the terminal.

[0413] Specific operation: The server securely transmits the final selected answer to the terminal, which then displays it to the user, allowing the user to obtain a reliable answer.

[0414] (Application example 2)

[0415] Next, a description will be given of Application Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0416] Conventional generative AI systems have difficulty generating responses that take user emotions into account, and in situations such as security support, it is necessary to respond appropriately to user anxieties and concerns. Therefore, a system that can recognize user emotions and provide appropriate responses based on those emotions is needed.

[0417] The identification process by the identification processing unit 290 of the data processing device 12 in Application Example 2 is realized by the following means. In this invention, the server includes means for sending a question entered by a user to an emotion engine and recognizing the user's emotion, means for sending the question accompanied by emotion data to the server, and means for the server to cause multiple generation year artificial intelligence models to generate answers taking the emotion data into consideration. This makes it possible to provide appropriate and reliable answers that take the user's emotions into consideration.

[0418] "Multiple Generation Year AI Models" refers to multiple AI systems that operate based on different algorithms and data sets and generate unique answers to questions.

[0419] "Means for generating answers to each question" refers to the function by which each AI model generates its own answer to a question entered by a user.

[0420] "Means for evaluation by other generation year artificial intelligence models" refers to the function by which answers generated by one generation year artificial intelligence model are evaluated by other generation year artificial intelligence models.

[0421] "Means for selecting the most appropriate answer" refers to the function of selecting the most appropriate answer based on the results of evaluations by multiple AI models.

[0422] "Means for outputting the selected answer as the final answer" refers to a function that provides the user with the selected most appropriate answer.

[0423] An "emotion engine" is a system that analyzes data such as user input, voice, and facial expressions to recognize the user's emotions.

[0424] The "means for transmitting a question with emotional data attached to it to a server" refers to a function for adding emotional data recognized by an emotion engine to question data and transmitting it to a server.

[0425] "Means for the server to generate answers taking into account emotional data" refers to the function of the server to have multiple artificial intelligence models generate answers based on emotional data.

[0426] This invention shows how to implement a system that recognizes a user's emotions and provides appropriate answers based on those emotions. This system is composed of multiple AI models, an emotion engine, and a server.

[0427] Program Generation

[0428] The system runs a program that includes the following functions:

[0429] Processing Description

[0430] 1. User enters question:

[0431] A user uses a terminal to input a question. For example, the user inputs the question, "My security camera isn't working. What should I do?"

[0432] 2. Emotion Recognition:

[0433] The device sends the text data entered by the user to the emotion engine, which uses natural language processing (NLP) technology to analyze the user's emotions and extract emotional data such as anxiety or impatience. The emotion engine can use, for example, Google Cloud's NLP API or Microsoft Azure's Text Analytics API.

[0434] 3. Sending questions with emotion data:

[0435] The terminal adds the emotion data extracted by the emotion engine and sends the question to the server.

[0436] 4. Generate answers:

[0437] The server distributes questions that take emotion data into account to multiple generative AI models, each of which uses its own algorithm to generate answers, including OpenAI's GPT-3 and GPT-4 models and the Hugging Face Transformers model.

[0438] 5. Peer assessment of answers:

[0439] The server sends the generated answers to other generative AI models, which then evaluate the answers based on accuracy, consistency, completeness of information, and suitability to the emotional data.

[0440] 6. Counting the evaluation results and selecting the responses:

[0441] The server tally the results and selects the answer with the highest rating.

[0442] 7. Final answer output:

[0443] The server sends the selected best answer to the user's terminal, which displays the final answer to the user.

[0444] Adding specific examples

[0445] As a concrete example, consider the case where a user inputs the question "My security camera isn't working." In this case, the emotion engine detects anxiety from the user's text data and sends the emotion data and question to the server. The server then delivers the question and emotion data to the generative AI model, which may generate the following answers:

[0446] Model 1: "Check the security camera cable. It may have a loose connection."

[0447] Model 2: "Don't be afraid. Try restarting your security cameras."

[0448] Model 3: "If you need your security camera repaired, please contact our support center."

[0449] These answers are then evaluated against each other, and the most appropriate answer, "Don't worry. Try restarting your security camera," is selected and provided to the user.

[0450] Prompt Sentence Examples

[0451] An example prompt sentence would be:

[0452] User Question: "My security camera isn't working. What should I do?"

[0453] Post-sentiment prompt: "In response to the question 'My security camera isn't working. What should I do?', which the user is feeling anxious about, each generative AI model should generate an answer that will ease the user's anxiety."

[0454] This system allows users to receive empathetic and reliable support, making it easier to deal with security-related issues.

[0455] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[0456] Step 1:

[0457] User enters a question:

[0458] The user inputs a question using the terminal. At the time the question is input, the terminal receives the question as text data. Specifically, the user inputs the question using a keyboard or voice input, and the question is saved as text data on the terminal.

[0459] Step 2:

[0460] Emotion Recognition:

[0461] The device sends the question text data entered by the user to the emotion engine. Specifically, it calls the emotion engine API (e.g., Google Cloud NLP API, Microsoft Azure Text Analytics API) to analyze the question text. The input data is the user's question text, and the output data is the emotion data returned by the emotion engine. The emotion engine performs text analysis and extracts emotions such as anxiety and impatience.

[0462] Step 3:

[0463] Sending questions with sentiment data:

[0464] The device adds the emotion data extracted by the emotion engine to the question text and sends it to the server. The input data is the emotion data and the question text, and the output data is a data packet that integrates them. This data packet is sent to the server and received by the server.

[0465] Step 4:

[0466] Generate an answer:

[0467] The server sends the received question and emotion data to multiple generative AI models in parallel. The server calls the API of the generative AI model (e.g., OpenAI GPT-3, Hugging Face Transformers), and each model generates an answer using its own algorithm. The input data is the question with the emotion data added, and the output data is the answer returned by each generative AI model.

[0468] Step 5:

[0469] Peer assessment of answers:

[0470] The server sends the answers received from each generative AI model back to other generative AI models to evaluate the answers. Specifically, the server requests each answer from other AI models for evaluation, and each model returns an evaluation score. The input data is the answer generated by each generative AI model, and the output data is the evaluation score.

[0471] Step 6:

[0472] Counting the assessment results and sorting the responses:

[0473] The server aggregates the evaluation results returned by each generative AI model and selects the answer with the highest evaluation. The input data is the evaluation score, and the output data is the optimal answer. The evaluation scores are aggregated and analyzed within the server.

[0474] Step 7:

[0475] Final answer output:

[0476] The server sends the selected optimal answer to the user's terminal. The terminal receives this final answer and displays it to the user. Specifically, the server sends the optimal answer in packet form to the terminal, and the terminal's user interface displays it. The input data is the optimal answer, and the output data is the answer displayed to the user.

[0477] These processing steps allow users to receive emotionally sensitive and reliable answers.

[0478] The specific processing unit 290 transmits the result of the specific processing to the smart device 14. In the smart device 14, the control unit 46A causes the output device 40 to output the result of the specific processing. The microphone 38B acquires audio indicating a user input regarding the result of the specific processing. The control unit 46A transmits audio data indicating the user input acquired by the microphone 38B to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the audio data.

[0479] The data generation model 58 is a so-called generative AI (Artificial Intelligence). An example of the data generation model 58 is ChatGPT (Internet Search<URL: https: / / openai.com / blog / chatgpt> ), Gemini (Internet search <url: https: gemini.google.com ?hl="ja">) and other generation AIs. The data generation model 58 is obtained by performing deep learning on a neural network. A prompt including an instruction is input to the data generation model 58, and inference data such as voice data indicating voice, text data indicating text, and image data indicating an image is also input. The data generation model 58 performs inference on the input inference data in accordance with the instruction indicated by the prompt, and outputs the inference result in a data format such as voice data and text data. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization.

[0480] In the above embodiment, an example in which the specific process is performed by the data processing device 12 has been given, but the technology of the present disclosure is not limited to this, and the specific process may be performed by the smart device 14.

[0481] [Second embodiment]

[0482] FIG. 3 shows an example of the configuration of a data processing system 210 according to the second embodiment.

[0483] 3, the data processing system 210 includes the data processing device 12 and smart glasses 214. An example of the data processing device 12 is a server.

[0484] The data processing device 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" according to the technology of the present disclosure. The computer 22 includes a processor 28, a RAM 30, and a storage 32. The processor 28, the RAM 30, and the storage 32 are connected to a bus 34. The database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. Examples of the network 54 include a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0485] The smart glasses 214 include a computer 36, a microphone 238, a speaker 240, a camera 42, and a communication I / F 44. The computer 36 includes a processor 46, a RAM 48, and a storage 50. The processor 46, the RAM 48, and the storage 50 are connected to a bus 52. The microphone 238, the speaker 240, and the camera 42 are also connected to the bus 52.

[0486] The microphone 238 receives instructions and the like from the user 20 by receiving voice uttered by the user 20. The microphone 238 captures the voice uttered by the user 20, converts the captured voice into audio data, and outputs it to the processor 46. The speaker 240 outputs audio in accordance with instructions from the processor 46.

[0487] Camera 42 is a small digital camera equipped with an optical system including a lens, aperture, and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, and captures images of the surroundings of user 20 (for example, an imaging range defined by an angle of view equivalent to the field of vision of a typical healthy person).

[0488] The communication I / F 44 is connected to a network 54. The communication I / Fs 44 and 26 are responsible for the exchange of various information between the processor 46 and the processor 28 via the network 54. The exchange of various information between the processor 46 and the processor 28 using the communication I / Fs 44 and 26 is carried out in a secure state.

[0489] Fig. 4 shows an example of the main functions of the data processing device 12 and the smart glasses 214. As shown in Fig. 4, in the data processing device 12, a specific process is performed by the processor 28. A specific process program 56 is stored in the storage 32.

[0490] The specific processing program 56 is an example of a "program" according to the technology of the present disclosure. The processor 28 reads the specific processing program 56 from the storage 32 and executes the read specific processing program 56 on the RAM 30. The specific processing is realized by the processor 28 operating as a specific processing unit 290 in accordance with the specific processing program 56 executed on the RAM 30.

[0491] The storage 32 stores a data generation model 58 and an emotion identification model 59. The data generation model 58 and the emotion identification model 59 are used by the identification processing unit 290.

[0492] In the smart glasses 214, the reception output process is performed by the processor 46. A reception output program 60 is stored in the storage 50. The processor 46 reads the reception output program 60 from the storage 50 and executes the read reception output program 60 on the RAM 48. The reception output process is realized by the processor 46 operating as the control unit 46A in accordance with the reception output program 60 executed on the RAM 48.

[0493] Next, a description will be given of the identification process performed by the identification processing unit 290 of the data processing device 12. In the following description, the data processing device 12 will be referred to as the "server" and the smart glasses 214 will be referred to as the "terminal."

[0494] This invention relates to a generative AI platform for improving the accuracy of answers generated by AI. This platform is a system in which multiple generative AI models cooperate to generate the most correct answers through mutual evaluation and discussion. Below, we will explain the program processing of this system in natural language using concrete examples.

[0495] Program processing explanation

[0496] Enter your question

[0497] A user uses an email client to type a question, for example, "What's the weather like in Tokyo tomorrow?"

[0498] Generate answers

[0499] The server receives this question and sends it to several different generative artificial intelligence models (AIM1, AIM2, AIM3, etc.), each of which uses its own algorithm to generate an answer.

[0500] AIM1: "It's sunny."

[0501] AIM2: "Tomorrow will be sunny."

[0502] AIM3: "It's forecast to be sunny in Tokyo."

[0503] Request for evaluation of answer

[0504] The server sends the generated answers to other generative AI models in the Validator Pool, allowing each model to evaluate the answers of the other generative AI models.

[0505] Evaluating answers

[0506] Each generative AI model evaluates the answers generated by other models and assigns them a score based on the accuracy, consistency, and completeness of the answers.

[0507] Aggregation of evaluation results

[0508] The server aggregates the results from each model and sees which answer received the highest rating.

[0509] Notification of evaluation results and request for discussion

[0510] The server notifies each generative AI model of the evaluation results and requests them to correct or supplement their answers as necessary.

[0511] Editing and discussing answers

[0512] Each model modifies its answer based on the discussion and adds any necessary corrections. For example, AIM1 can update its answer as follows:

[0513] AIM1: "Tomorrow will be sunny in Tokyo, with temperatures around 25 degrees."

[0514] AIM2: "Tomorrow in Tokyo will be sunny with a maximum temperature of 25 degrees."

[0515] AIM3: "It'll be sunny and 25 degrees."

[0516] Final evaluation and majority vote

[0517] The server then re-evaluates the updated answer against the models in the Validator Pool. Each model selects the most appropriate answer and a majority vote is held.

[0518] Selection of the final answer

[0519] Based on the evaluation results, the server selects the final answer. The final answer obtained through this process is "Tomorrow in Tokyo, it will be sunny and the temperature will be around 25 degrees."

[0520] Submit your answer

[0521] The server sends the final selected answer to the terminal.

[0522] Show Answers

[0523] The terminal displays the final answer to the user.

[0524] This system allows multiple generative AI models to cooperate and generate more accurate and reliable answers through mutual evaluation and discussion. As can be seen from this explanation with concrete examples, the present invention provides an effective means for suppressing hallucination and improving the accuracy of information.

[0525] The processing flow will be explained below.

[0526] Step 1:

[0527] The user inputs a question, for example, "What's the weather like in Tokyo tomorrow?" into the terminal.

[0528] Step 2:

[0529] The terminal transmits the entered question to the server.

[0530] Step 3:

[0531] The server receives questions and distributes them to multiple generative artificial intelligence models (AIM1, AIM2, AIM3).

[0532] Step 4:

[0533] Each generative AI model generates an answer to a question sent by the server using its own algorithm. For example,

[0534] AIM1: "It's sunny."

[0535] AIM2: "Tomorrow will be sunny."

[0536] AIM3: "It's forecast to be sunny in Tokyo."

[0537] Step 5:

[0538] The server sends the generated answer to each of the other generative AI models in the Validator Pool for evaluation.

[0539] Step 6:

[0540] Each generative AI model evaluates the answers it receives from other models and assigns a score based on the accuracy, consistency, and completeness of the answers.

[0541] Step 7:

[0542] The server aggregates the evaluation results from each model and analyzes which answer received the highest evaluation.

[0543] Step 8:

[0544] The server notifies each generative AI model of the evaluation results and, if necessary, requests them to correct or supplement their answers.

[0545] Step 9:

[0546] Each generative AI model modifies the answer based on the evaluation results and adds correction information. For example,

[0547] AIM1: "Tomorrow will be sunny in Tokyo, with temperatures around 25 degrees."

[0548] AIM2: "Tomorrow in Tokyo will be sunny with a maximum temperature of 25 degrees."

[0549] AIM3: "It'll be sunny and the temperature will be 25 degrees."

[0550] Step 10:

[0551] The server then sends the updated answers to the Validator Pool again, asking each model to re-evaluate.

[0552] Step 11:

[0553] Each generative AI model then evaluates the answers again, selects the most appropriate answer, and assigns it a score.

[0554] Step 12:

[0555] The server tally the results of the reevaluation and selects the answer with the highest score. For example, "Tomorrow in Tokyo, it will be sunny and the temperature will be around 25 degrees."

[0556] Step 13:

[0557] The server sends the final selected answer to the terminal.

[0558] Step 14:

[0559] The terminal displays the final answer to the user.

[0560] Example 1

[0561] Next, a description will be given of Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the smart glasses 214 will be referred to as a "terminal."

[0562] In existing generative AI models, relying on a single model's answer can reduce the accuracy and reliability of the answer. Furthermore, when the answers generated by each model differ, it is difficult to determine which answer is the most accurate. There is a need to solve this problem and provide more accurate and reliable answers.

[0563] The specific processing by the specific processing unit 290 of the data processing device 12 in the first embodiment is realized by the following means.

[0564] In this invention, the server includes means for inputting a question, means for multiple generative AI models to each generate an answer to the question, means for other generative AI models to evaluate the answers generated by each generative AI model, means for selecting the most appropriate answer based on the evaluation results, means for correcting or supplementing the selected answer, means for re-evaluating the corrected answer, means for performing the evaluation process in stages and selecting a final answer based on the result of majority voting, and means for transmitting and displaying the selected final answer to a user terminal. This makes it possible to provide users with more accurate and reliable final answers through mutual evaluation and discussion among multiple generative AI models.

[0565] The "means for inputting a question" provides an interface for a user to input a question to the system.

[0566] A "generative artificial intelligence model" refers to an algorithm or machine learning model that generates answers in natural language to input questions.

[0567] The "means for generating an answer" is what enables the generative artificial intelligence model to generate an answer to an input question.

[0568] "Means for evaluating answers" refers to a function in which answers generated by a generative AI model are evaluated by other generative AI models and scored for accuracy and consistency.

[0569] The "means for selecting the most appropriate answer" includes a function for selecting the answer that has received the highest evaluation based on the evaluation results of each generative artificial intelligence model.

[0570] "Means for correcting or supplementing answers" refers to the functionality for adding necessary corrections or additional information to selected answers.

[0571] "Means for re-evaluation" refers to a function that allows each generative artificial intelligence model to re-evaluate an answer that has been corrected or supplemented.

[0572] "Means for carrying out the evaluation process in stages" includes a function for dividing the evaluation into multiple stages and carrying out a final evaluation based on the evaluation results at each stage.

[0573] "Means for selecting a final answer based on the results of a majority vote" refers to a function for selecting the answer that receives the most support as the final answer based on the evaluation results of each generative AI model.

[0574] The "means for transmitting and displaying the final answer" provides a function for transmitting the selected final answer to the user terminal and displaying it.

[0575] This invention relates to a generative AI platform for improving the accuracy of answers generated by AI. This platform is a system in which multiple generative artificial intelligence models (generative AI models) cooperate to generate the most correct answers through mutual evaluation and discussion.

[0576] Hardware and software used

[0577] The hardware required to implement this system includes a high-performance server and user terminals (e.g., PCs and smartphones). The server has high computing power and storage capacity, and a cloud platform for running the generative AI models can also be used. The software includes each generative AI model, an email client, a web form, etc.

[0578] Data processing and calculation

[0579] When a user enters a question using an email client or web form, the question is stored in a database and sent to a server, which parses the question and converts it into a prompt that the generative AI model can understand.

[0580] For example, in response to the question "What's the weather like in Tokyo tomorrow?", the server generates a prompt sentence, "Please tell me the weather forecast for Tokyo tomorrow," and sends this to multiple generative AI models (AIM1, AIM2, AIM3, etc.).

[0581] Each generative AI model receives a prompt and uses its own algorithms and datasets to generate an answer.

[0582] AIM1: "It's sunny."

[0583] AIM2: "Tomorrow will be sunny."

[0584] AIM3: "It's forecast to be sunny in Tokyo."

[0585] The generated answers are sent by the server to other generative AI models in the Validator Pool, where each model evaluates the answers of the other models based on the accuracy, consistency, and completeness of the answers.

[0586] The evaluation results are tallied, and the server checks which answer received the highest rating. For example, if AIM1 received the highest rating, that score is recorded. The evaluation results are notified to each generative AI model, which corrects or supplements the answer as necessary. For example, if AIM1 is evaluated as having insufficient temperature information, additional information is provided.

[0587] Each generative AI model modifies its answer based on the discussion and adds any necessary corrections. Answers may be updated as follows:

[0588] AIM1: "Tomorrow will be sunny in Tokyo, with temperatures around 25 degrees."

[0589] AIM2: "Tomorrow in Tokyo will be sunny with a maximum temperature of 25 degrees."

[0590] AIM3: "It'll be sunny and 25 degrees."

[0591] The server then evaluates the updated answers again using the models in the Validator Pool, and each generative AI model selects the most appropriate answer through a majority vote. Based on the evaluation results, the server selects the final answer: "Tomorrow in Tokyo will be sunny, with a temperature of around 25 degrees."

[0592] Examples of specific examples and prompts

[0593] As a specific example, the following shows how a generative AI model generates and evaluates answers to the input, "What's the weather like in Tokyo tomorrow?"

[0594] User input: "What's the weather like in Tokyo tomorrow?"

[0595] Server prompt: "What's the weather forecast for Tokyo tomorrow?"

[0596] Example answer from a generative AI model:

[0597] AIM1: "It's sunny."

[0598] AIM2: "Tomorrow will be sunny."

[0599] AIM3: "It's forecast to be sunny in Tokyo."

[0600] Example of an answer after evaluation and revision:

[0601] AIM1: "Tomorrow will be sunny in Tokyo, with temperatures around 25 degrees."

[0602] AIM2: "Tomorrow in Tokyo will be sunny with a maximum temperature of 25 degrees."

[0603] AIM3: "It'll be sunny and 25 degrees."

[0604] Ultimately, the answer selected by the server based on the majority vote was: "Tomorrow in Tokyo, it will be sunny, and the temperature will be around 25 degrees."

[0605] Through the collaboration and mutual evaluation of multiple generative AI models, this system enables the provision of more accurate and reliable information to users.

[0606] The flow of the identification process in the first embodiment will be described with reference to FIG.

[0607] Step 1:

[0608] A user enters a question using an email client or web form, for example, "What's the weather like in Tokyo tomorrow?" This input is sent to the server.

[0609] Input: User question: "What's the weather like in Tokyo tomorrow?"

[0610] Output: Query data sent to the server

[0611] Step 2:

[0612] The server analyzes the received question and converts it into a prompt that the generative AI model can understand. For example, it generates a prompt such as, "Please tell me the weather forecast for Tokyo tomorrow."

[0613] Input: Question data "What's the weather like in Tokyo tomorrow?"

[0614] Output: Prompt "What is the weather forecast for Tokyo tomorrow?"

[0615] Step 3:

[0616] The server sends the generated prompt to multiple generative AI models (AIM1, AIM2, AIM3, etc.), each of which receives the prompt and generates an answer using its own algorithm and dataset.

[0617] Input: Prompt "What is the weather forecast for Tokyo tomorrow?"

[0618] Output: Answer of each generative AI model (e.g., AIM1: "It's sunny," AIM2: "It will be sunny tomorrow," AIM3: "It is forecast to be sunny in Tokyo.")

[0619] Step 4:

[0620] The server sends the generated answers to other generative AI models in the Validator Pool, asking each model to evaluate the answers of the other models.

[0621] Input: The generative AI model's answer

[0622] Output: Evaluation request data

[0623] Step 5:

[0624] Generative AI models evaluate the answers generated by other generative AI models. For example, AIM2 looks at the answers of AIM1 and AIM3 and assigns a score to each. The evaluation criteria are the accuracy, consistency, and completeness of the information.

[0625] Input: Answers from other generative AI models

[0626] Output: Evaluation score (e.g., AIM2 scoring)

[0627] Step 6:

[0628] The server aggregates the results from each model and checks which answer received the highest score. For example, if AIM1 received the highest score, that score is recorded.

[0629] Input: Rating score

[0630] Output: The answer from the model with the highest score

[0631] Step 7:

[0632] The server notifies each generative AI model of the evaluation results and asks them to correct or supplement their answers as necessary. For example, if AIM1 receives an evaluation that "temperature information is insufficient," it will ask them to provide additional information.

[0633] Input: Evaluation result

[0634] Output: Notification of correction request

[0635] Step 8:

[0636] The generative AI model then modifies the answer based on the discussion and adds any necessary corrections. For example, AIM1 updates the answer to "Tomorrow will be sunny in Tokyo, with the temperature around 25 degrees."

[0637] Input: Notification of correction request

[0638] Output: Updated answer (e.g. AIM1: "Tomorrow will be sunny in Tokyo, with the temperature around 25 degrees.")

[0639] Step 9:

[0640] The server then re-evaluates the updated answers with the models in the Validator Pool. Each generative AI model selects the most appropriate answer and a majority vote is held.

[0641] Input: Updated answer

[0642] Output: Reassessment score and majority vote result

[0643] Step 10:

[0644] The server selects the final answer based on the evaluation results. For example, it selects "Tomorrow in Tokyo will be sunny, with the temperature around 25 degrees" as the most popular answer.

[0645] Input: Reassessment score and majority vote result

[0646] Output: Final answer "Tomorrow in Tokyo it will be sunny, with the temperature around 25 degrees."

[0647] Step 11:

[0648] The server sends the final selected answer to the terminal.

[0649] Input: Final answer

[0650] Output: Data sent to the terminal

[0651] Step 12:

[0652] The terminal displays the final answer to the user, who can view it through their email client or web interface.

[0653] Input: Final response data

[0654] Output: What is displayed to the user

[0655] This series of processes enables the provision of more accurate and reliable information to users through cooperation and mutual evaluation of multiple generative AI models.

[0656] (Application example 1)

[0657] Next, a description will be given of Application Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the smart glasses 214 will be referred to as a "terminal."

[0658] Conventional answer generation systems using generative AI models often lack consistency and reliability, with the accuracy of each model varying. Furthermore, especially in the security field, where incorrect answers can have significant impacts, higher accuracy and reliability are required. For this reason, there was a need to develop a system in which multiple AI models cooperate and generate the most accurate and reliable answers through mutual evaluation and discussion.

[0659] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 1 is realized by the following means.

[0660] In this invention, the server includes means for multiple generative AI models to generate answers to questions, means for other generative AI models to evaluate the answers generated by each generative AI model, and means for selecting the most appropriate answer based on the evaluation results. This improves the accuracy and reliability of the generated answers, enabling reliable notifications to users, particularly when used as security alerts.

[0661] "Multiple generative artificial intelligence models" refers to multiple AI models that use different algorithms and datasets to generate answers in their own unique way.

[0662] "Means for generating answers to each question" refers to the process or function by which each AI model generates an individual answer to a question input by a user.

[0663] "Means for evaluation by other generative artificial intelligence models" refers to the process or function by which answers generated by one AI model are evaluated by other AI models to determine their accuracy and consistency.

[0664] "Means for selecting the most appropriate answer" refers to the process or function of selecting the answer that received the highest rating based on the results of multiple evaluations.

[0665] "Means for outputting the selected answer as the final answer" refers to the process or function of providing the selected most appropriate answer to the user as the final answer.

[0666] "Means for holding discussions, adding corrections, and updating answers" refers to the process or function of discussing the answers generated by each AI model and making necessary corrections to generate more accurate answers.

[0667] "A method of conducting a step-by-step evaluation process and selecting a final answer based on the results of a majority vote" refers to a process of going through a series of evaluation and selection steps in a step-by-step manner, ultimately selecting the answer supported by the majority of evaluators.

[0668] A "security alert" refers to a warning message that notifies the user of an abnormal security situation or abnormality.

[0669] "Means of notifying the user" refers to the process or function of sending the final selected answer or alert to the user's device and notifying the user by visual, auditory, or other means.

[0670] This invention relates to a system in which multiple generative artificial intelligence models cooperate to generate the most correct answer through mutual evaluation and discussion. Specifically, this invention describes an application for providing accurate alerts to users in security services.

[0671] Hardware and software used

[0672] Hardware:

[0673] 1. Server: Runs multiple AI models and manages the evaluation process.

[0674] 2. Security cameras: collect data and provide it for analysis.

[0675] 3. Smartphone: The device where the user enters the question and receives the final answer.

[0676] software:

[0677] 1. Python: A system-wide programming language.

[0678] 2. Requests library: Manages API communication.

[0679] 3. AI Models: Multiple generative AI models that use various algorithms to generate solutions to problems.

[0680] System processing flow

[0681] 1. User enters question:

[0682] Users use a smartphone app to enter specific questions, such as "Can you check if there's an intruder in my office?"

[0683] 2. The server receives the query:

[0684] The server receives this question and sends it to multiple generative AI models, each of which uses its own algorithm to generate an answer.

[0685] 3. The AI ​​model generates the answer:

[0686] Each generative AI model generates an answer to the question, for example:

[0687] Model 1: "Motion detected on camera, but it doesn't seem to be an intruder."

[0688] Model 2: "It looks like the door is open, but we need to check it out immediately."

[0689] Model 3: "Motion detected, but no further information available."

[0690] 4. Other models rate your answers:

[0691] The server sends these answers to other generative AI models, which then evaluate the answers of the other models based on the accuracy and consistency of the answers.

[0692] 5. Compilation of evaluation results and selection of best answers:

[0693] Based on the evaluation results of each AI model, the answer with the highest score will be selected. For example, the evaluation will be as follows:

[0694] "Motion detected on camera, but it doesn't appear to be an intruder" received the highest rating.

[0695] 6. Generate final answer and notify user:

[0696] The server then sends the final selected answer to the user's device and notifies them, allowing the user to receive accurate and reliable security alerts.

[0697] Prompt Sentence Examples

[0698] User input:

[0699] "Check if there's an intruder in the office."

[0700] This invention enables multiple generative AI models to cooperate to generate the most appropriate response and notify the user of it as a security alert, allowing the user to take appropriate action based on highly reliable information.

[0701] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[0702] Step 1:

[0703] A user uses a smartphone to input a question, for example, "Please check if there is an intruder in my office." This input is sent to the server through the user interface.

[0704] Step 2:

[0705] The server receives a user's question and sends it to multiple generative AI models. Each model receives the question and generates an answer using its own algorithm. A typical answer might be, "Motion was detected on the camera, but it doesn't appear to be an intruder."

[0706] Step 3:

[0707] The server sends the answers generated by each generative AI model to other generative AI models for evaluation. At this time, the input provided to each model is a list of generated answers, and each model evaluates this list.

[0708] Step 4:

[0709] Each generative AI model evaluates the answers generated by other models and assigns a score based on their accuracy, consistency, and completeness of information. The evaluation results are sent to the server in the form of a score. For example, model A may assign 90 points to the answer of model B and 80 points to the answer of model C.

[0710] Step 5:

[0711] The server then selects the answer with the highest score based on the aggregated evaluation results. For example, if "Motion was detected on the camera, but it doesn't appear to be an intruder" receives the highest score, it will be selected.

[0712] Step 6:

[0713] The server then sends the selected answer to the user's smartphone. This notification includes the most highly rated answer and a brief explanation of it. The user can then check the answer on their device.

[0714] Step 7:

[0715] If necessary, the generated answers may be further discussed and corrections added. Each generative AI model will be discussed again and generate a new answer with the necessary corrections. This process of evaluation and correction is repeated until the revised best answer is selected.

[0716] The above is a specific flow of processing by the program that realizes this invention, which enables users to receive accurate and reliable security alerts.

[0717] Furthermore, an emotion engine that estimates the user's emotion may be further combined. That is, the identification processing unit 290 may estimate the user's emotion using the emotion identification model 59, and perform identification processing using the user's emotion.

[0718] This invention relates to a generative AI platform that uses multiple generative AI models and an emotion engine to generate accurate and appropriate answers to user questions. Each generative AI model generates and modifies answers through mutual evaluation and discussion, and by combining it with an emotion engine that recognizes the user's emotions, it provides answers that are adapted to the user's emotional state. Below, we will explain the program processing of this system in natural language using concrete examples.

[0719] Program processing explanation

[0720] Enter your question

[0721] The user uses the terminal to input a question, for example, "Tell me what my friends think."

[0722] Emotion recognition

[0723] The device sends the user's input, voice, facial expression, and other data to the emotion engine, which analyzes this data and recognizes the user's emotions. For example, it can detect "anxiety."

[0724] Submit a Question

[0725] The device sends the question with the recognized emotion data to the server.

[0726] Generate answers

[0727] The server distributes the emotional data attached to the question to multiple generative AI models (AIM1, AIM2, AIM3). Each model takes the emotional data into account and generates an answer using its own algorithm. For example,

[0728] AIM1: "Your friends care about you."

[0729] AIM2: "Your friends are worried about you."

[0730] AIM3: "Your friends trust you."

[0731] Request for evaluation of answer

[0732] The server sends the generated answers to other generative AI models, allowing each model to evaluate the answers of the other generative AI models.

[0733] Evaluating answers

[0734] The generative AI model evaluates the answers generated by other models and assigns them a score based on the answer's accuracy, consistency, completeness of information, and fit to the perceived emotion.

[0735] Aggregation of evaluation results

[0736] The server aggregates the evaluation results from each model and analyzes which answer received the highest evaluation.

[0737] Notification of evaluation results and request for discussion

[0738] The server notifies each generative AI model of the evaluation results and requests them to revise or supplement their answers if necessary. Emotional data is also reflected.

[0739] Editing and discussing answers

[0740] Each generative AI model modifies its answers based on the evaluation results and emotional data, adding correction information. For example,

[0741] AIM1: "Friends care about you and always want to support you."

[0742] AIM2: "Your friend is very concerned about you and wants to talk more."

[0743] AIM3: "Your friends trust you and want to talk to you if they have a problem."

[0744] Final evaluation and majority vote

[0745] The server re-evaluates the updated answer and asks each model to re-evaluate.

[0746] Selection of the final answer

[0747] Based on the reevaluation results, the server selects a final answer, such as "Your friends care about you and always want to support you."

[0748] Submit your answer

[0749] The server sends the final selected answer to the terminal.

[0750] Show Answers

[0751] The terminal displays the final answer to the user.

[0752] This system enables multiple generative AI models to generate more accurate and reliable answers that adapt to the user's emotional state through mutual evaluation and discussion. As can be seen from the explanation using concrete examples, the present invention provides an innovative means for suppressing hallucination and providing information that takes into consideration the user's emotions.

[0753] The processing flow will be explained below.

[0754] Step 1:

[0755] The user inputs a question, for example, "What's the weather like in Tokyo tomorrow?" into the terminal.

[0756] Step 2:

[0757] The device sends the input question as data to the emotion engine, which analyzes the user's input, voice, facial expressions, and other data to recognize the user's emotions. For example, it can detect "anxiety."

[0758] Step 3:

[0759] The emotion engine returns the recognized emotion data to the terminal.

[0760] Step 4:

[0761] The device sends the question with the emotional data attached to it to the server.

[0762] Step 5:

[0763] The server receives questions and emotion data and distributes them to multiple generative AI models (AIM1, AIM2, AIM3).

[0764] Step 6:

[0765] Each generative AI model uses its own algorithm to generate answers to questions while taking into account emotional data. For example,

[0766] AIM1: "It's sunny."

[0767] AIM2: "Tomorrow will be sunny."

[0768] AIM3: "It's forecast to be sunny in Tokyo."

[0769] Step 7:

[0770] The server compiles the generated answers and sends them to other generative AI models, allowing each model to evaluate the answers of the other generative AI models.

[0771] Step 8:

[0772] Each generative AI model evaluates the answers sent by other generative AI models and assigns a score based on the answer's accuracy, consistency, completeness of information, and fit to the perceived emotion.

[0773] Step 9:

[0774] The server aggregates the evaluation results from each model and analyzes which answer received the highest evaluation.

[0775] Step 10:

[0776] The server notifies each generative AI model of the evaluation results and requests them to revise or supplement their answers as necessary. This emotional data is also reflected in the re-evaluation.

[0777] Step 11:

[0778] Each generative AI model modifies its answers based on the evaluation results and emotional data, adding correction information. For example,

[0779] AIM1: "Tomorrow will be sunny in Tokyo. The temperature will be around 25 degrees."

[0780] AIM2: "Tomorrow in Tokyo will be sunny with a maximum temperature of 25 degrees."

[0781] AIM3: "It'll be sunny and the temperature will be 25 degrees."

[0782] Step 12:

[0783] The server then distributes the updated answers to each generative AI model again and requests them to evaluate again.

[0784] Step 13:

[0785] Each generative AI model then evaluates the answers again, selects the most appropriate answer, and assigns it a score.

[0786] Step 14:

[0787] The server tally the results of the reevaluation and selects the answer with the highest score. For example, "Tomorrow in Tokyo, it will be sunny and the temperature will be around 25 degrees."

[0788] Step 15:

[0789] The server sends the final selected answer to the terminal.

[0790] Step 16:

[0791] The terminal displays the final answer to the user.

[0792] Example 2

[0793] Next, a description will be given of Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the smart glasses 214 will be referred to as a "terminal."

[0794] In conventional generative AI systems, the answers generated are not always adapted to the user's emotional state, which reduces user satisfaction. Furthermore, the evaluation process for multiple generative AI models is insufficient, resulting in answers that lack consistency and accuracy. Furthermore, the generated answers do not reflect correction information or discussions, which can result in the delivery of less than optimal answers.

[0795] The specific processing by the specific processing unit 290 of the data processing device 12 in the second embodiment is realized by the following means.

[0796] In this invention, the server includes means for multiple generative AI models to generate answers to questions, means for other generative AI models to evaluate the answers generated by each generative AI model, means for selecting the most appropriate answer based on the evaluation results, an emotion engine that recognizes the user's emotion using data input by the user, and means for attaching the emotion data recognized by the emotion engine to the question and transmitting it, thereby making it possible to generate and provide accurate and appropriate answers that are adapted to the user's emotional state.

[0797] "Multiple generative AI models" are multiple AI systems with different algorithms and settings, each capable of generating answers independently.

[0798] A "question" is any information or question entered by a user that requires an answer from a generative AI model.

[0799] The "means for generating an answer" refers to the process or algorithm that the generative artificial intelligence model uses to create an appropriate answer to a user's question.

[0800] "Means for evaluation by other generative artificial intelligence models" refers to the process by which each generative artificial intelligence model evaluates answers generated by other models and assigns them a score or ranking.

[0801] The "means of selecting the most appropriate answer based on the evaluation results" is a system that aggregates the evaluation results obtained from multiple generative AI models and selects the answer that receives the highest evaluation.

[0802] An "emotion engine" is a system that analyzes the user's input data, voice, facial expressions, etc. to recognize the user's emotional state.

[0803] "Emotion data" is data showing the results of analysis by the user's emotion engine, and represents an emotional state such as anxiety or relief.

[0804] The "means of transmission" is the process or method of transmitting data from the device to the server, often using a secure communications protocol.

[0805] "Correction information" is information that is added to the initially generated answer, and is used to improve the accuracy and appropriateness of the answer.

[0806] "Means for discussion" refers to a process in which multiple generative AI models discuss the generated answers and make optimal improvements or corrections.

[0807] "Majority vote" is a method of statistically aggregating the results of multiple evaluations and scoring and selecting the answer that receives the most support.

[0808] The system starts when a user uses a device to input a question. For example, "Tell me what my friend thinks." The user can use a keyboard or touch screen to input a question. The system may also collect voice and facial expression data, which are sent to the emotion engine.

[0809] The emotion engine analyzes the collected data and recognizes the user's emotions. Voice recognition technology and facial expression analysis algorithms are used to recognize emotions, such as "anxiety." The emotion data recognized in this way is attached to the user's question and sent from the device to the server. A secure communication protocol (e.g., HTTPS) is used to protect the data.

[0810] The server distributes the received question and emotion data to multiple generative AI models (e.g., AIM1, AIM2, AIM3). Each model generates an answer using its own natural language processing algorithm. Examples of generated answers include:

[0811] AIM1: "Your friends care about you."

[0812] AIM2: "Your friends are worried about you."

[0813] AIM3: "Your friends trust you."

[0814] The server then sends the answers generated by each generative AI model to other generative AI models, who evaluate each other's answers based on the accuracy, consistency, completeness of the information, and suitability to the emotional data. For example, AIM1 might give 70 points to AIM2's answer and 80 points to AIM3's answer.

[0815] The server aggregates these evaluation results and selects the answer with the highest rating. After selection, the server notifies each generative AI model of the evaluation results and requests them to revise or supplement the answer as necessary. Emotional data is also reflected.

[0816] Each generative AI model modifies its answers based on the evaluation results and emotional data, adding correction information. For example, it may be modified as follows:

[0817] AIM1: "Friends care about you and always want to support you."

[0818] AIM2: "Your friend is very concerned about you and wants to talk more."

[0819] AIM3: "Your friends trust you and want to talk to you if they have a problem."

[0820] Finally, the server re-evaluates the updated answers and selects the answer with the most votes. The final answer is sent to the device and displayed to the user, providing a reliable answer to the question.

[0821] For example, consider the following prompt:

[0822] "Please tell me what my friend thinks. I'm feeling anxious right now."

[0823] This system enables multiple generative AI models to provide more accurate and reliable answers that adapt to the user's emotional state through mutual evaluation and discussion, thereby improving user satisfaction and the quality of information provided.

[0824] The flow of the identification process in the second embodiment will be described with reference to FIG.

[0825] Step 1: The user enters a question using a terminal.

[0826] Input: Text data entered by the user using a keyboard or touchscreen.

[0827] Output: Question text data imported to the terminal.

[0828] Specific operation: The user inputs a question such as "Please tell me what my friends think." The device collects the text data of the question.

[0829] Step 2: The device sends the user's input data to the emotion engine and recognizes the user's emotion.

[0830] Input: Text data of the user's question acquired by the device, as well as voice and facial expression data.

[0831] Output: Emotion data parsed by the emotion engine.

[0832] Specific operation: The device sends the user's text data, voice, and facial expressions to the emotion engine, which analyzes them and recognizes emotions such as "anxiety."

[0833] Step 3: The device sends the question accompanied by the recognized emotion data to the server.

[0834] Input: User question text data and sentiment data.

[0835] Output: Questions and sentiment data sent to the server.

[0836] Specific operation: The device attaches emotional data obtained from the emotion engine to the question and sends it securely to the server using HTTPS.

[0837] Step 4: The server distributes the questions and emotion data to multiple generative AI models.

[0838] Input: Question and emotion data received by the server.

[0839] Output: The questions and sentiment data delivered to each generative AI model.

[0840] Specific operation: The server distributes the received question and emotion data to the generative AI models (e.g., AIM1, AIM2, AIM3), and each model generates an answer.

[0841] Step 5: Each generative AI model generates an answer to the question.

[0842] Input: Questions and sentiment data delivered to each generative AI model.

[0843] Output: The generated answer.

[0844] How it works: The generative AI model uses a proprietary natural language processing algorithm to generate answers. For example, AIM1 generates "Your friends care about you."

[0845] Step 6: The server sends the generated answer to other generative AI models for evaluation.

[0846] Input: The answers generated by each generative AI model.

[0847] Output: Answer evaluation request sent to other generative AI models.

[0848] Specific operation: The server collects the answers generated by each model and sends them to other generative AI models for evaluation.

[0849] Step 7: Each generative AI model evaluates the answers generated by other models and assigns a score.

[0850] Input: Answers received from other generative artificial intelligence models.

[0851] Output: Evaluation score.

[0852] How it works: Each generative AI model evaluates the answers of other models based on specified criteria and assigns a score based on, for example, accuracy or consistency.

[0853] Step 8: The server aggregates the evaluation results of each generative AI model and selects the answer with the highest evaluation.

[0854] Input: Evaluation scores from all generative AI models.

[0855] Output: The answer that received the most votes.

[0856] Specific operation: The server aggregates the evaluation results and selects the answer with the highest score.

[0857] Step 9: The server notifies each generative AI model of the selected answers and requests corrections or supplements if necessary.

[0858] Input: Screened answers and assessment results.

[0859] Output: Request for corrections and clarifications.

[0860] Specific operation: The server sends the selected answers and their evaluation scores to the model, requesting corrections or supplements as necessary.

[0861] Step 10: Each generative AI model modifies the answer and adds correction information.

[0862] Input: Correction request and original answer from the server.

[0863] Output: The revised answer and any corrections added.

[0864] Specific behavior: Each model modifies the answer based on the evaluation results, adding correction information such as, "Your friends care about you and always want to support you."

[0865] Step 11: The server re-evaluates the revised answers and selects the final answer based on the majority vote.

[0866] Input: Revised answer.

[0867] Output: The final selected answer.

[0868] Specific operation: The server has each generative AI model evaluate the revised answer again, and selects the final answer by majority vote based on the evaluation results.

[0869] Step 12: The server sends the final selected answer to the terminal and displays it to the user.

[0870] Input: Final selected answer.

[0871] Output: The final answer displayed on the terminal.

[0872] Specific operation: The server securely transmits the final selected answer to the terminal, which then displays it to the user, allowing the user to obtain a reliable answer.

[0873] (Application example 2)

[0874] Next, a description will be given of Application Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the smart glasses 214 will be referred to as a "terminal."

[0875] Conventional generative AI systems have difficulty generating responses that take user emotions into account, and in situations such as security support, it is necessary to respond appropriately to user anxieties and concerns. Therefore, a system that can recognize user emotions and provide appropriate responses based on those emotions is needed.

[0876] The identification process by the identification processing unit 290 of the data processing device 12 in Application Example 2 is realized by the following means. In this invention, the server includes means for sending a question entered by a user to an emotion engine and recognizing the user's emotion, means for sending the question accompanied by emotion data to the server, and means for the server to cause multiple generation year artificial intelligence models to generate answers taking the emotion data into consideration. This makes it possible to provide appropriate and reliable answers that take the user's emotions into consideration.

[0877] "Multiple Generation Year AI Models" refers to multiple AI systems that operate based on different algorithms and data sets and generate unique answers to questions.

[0878] "Means for generating answers to each question" refers to the function by which each AI model generates its own answer to a question entered by a user.

[0879] "Means for evaluation by other generation year artificial intelligence models" refers to the function by which answers generated by one generation year artificial intelligence model are evaluated by other generation year artificial intelligence models.

[0880] "Means for selecting the most appropriate answer" refers to the function of selecting the most appropriate answer based on the results of evaluations by multiple AI models.

[0881] "Means for outputting the selected answer as the final answer" refers to a function that provides the user with the selected most appropriate answer.

[0882] An "emotion engine" is a system that analyzes data such as user input, voice, and facial expressions to recognize the user's emotions.

[0883] The "means for transmitting a question with emotional data attached to it to a server" refers to a function for adding emotional data recognized by an emotion engine to question data and transmitting it to a server.

[0884] "Means for the server to generate answers taking into account emotional data" refers to the function of the server to have multiple artificial intelligence models generate answers based on emotional data.

[0885] This invention shows how to implement a system that recognizes a user's emotions and provides appropriate answers based on those emotions. This system is composed of multiple AI models, an emotion engine, and a server.

[0886] Program Generation

[0887] The system runs a program that includes the following functions:

[0888] Processing Description

[0889] 1. User enters question:

[0890] A user uses a terminal to input a question. For example, the user inputs the question, "My security camera isn't working. What should I do?"

[0891] 2. Emotion Recognition:

[0892] The device sends the text data entered by the user to the emotion engine, which uses natural language processing (NLP) technology to analyze the user's emotions and extract emotional data such as anxiety or impatience. The emotion engine can use, for example, Google Cloud's NLP API or Microsoft Azure's Text Analytics API.

[0893] 3. Sending questions with emotion data:

[0894] The terminal adds the emotion data extracted by the emotion engine and sends the question to the server.

[0895] 4. Generate answers:

[0896] The server distributes questions that take emotion data into account to multiple generative AI models, each of which uses its own algorithm to generate answers, including OpenAI's GPT-3 and GPT-4 models and the Hugging Face Transformers model.

[0897] 5. Peer assessment of answers:

[0898] The server sends the generated answers to other generative AI models, which then evaluate the answers based on accuracy, consistency, completeness of information, and suitability to the emotional data.

[0899] 6. Counting the evaluation results and selecting the responses:

[0900] The server tally the results and selects the answer with the highest rating.

[0901] 7. Final answer output:

[0902] The server sends the selected best answer to the user's terminal, which displays the final answer to the user.

[0903] Adding specific examples

[0904] As a concrete example, consider the case where a user inputs the question "My security camera isn't working." In this case, the emotion engine detects anxiety from the user's text data and sends the emotion data and question to the server. The server then delivers the question and emotion data to the generative AI model, which may generate the following answers:

[0905] Model 1: "Check the security camera cable. It may have a loose connection."

[0906] Model 2: "Don't be afraid. Try restarting your security cameras."

[0907] Model 3: "If you need your security camera repaired, please contact our support center."

[0908] These answers are then evaluated against each other, and the most appropriate answer, "Don't worry. Try restarting your security camera," is selected and provided to the user.

[0909] Prompt Sentence Examples

[0910] An example prompt sentence would be:

[0911] User Question: "My security camera isn't working. What should I do?"

[0912] Post-sentiment prompt: "In response to the question 'My security camera isn't working. What should I do?', which the user is feeling anxious about, each generative AI model should generate an answer that will ease the user's anxiety."

[0913] This system allows users to receive empathetic and reliable support, making it easier to deal with security-related issues.

[0914] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[0915] Step 1:

[0916] User enters a question:

[0917] The user inputs a question using the terminal. At the time the question is input, the terminal receives the question as text data. Specifically, the user inputs the question using a keyboard or voice input, and the question is saved as text data on the terminal.

[0918] Step 2:

[0919] Emotion Recognition:

[0920] The device sends the question text data entered by the user to the emotion engine. Specifically, it calls the emotion engine API (e.g., Google Cloud NLP API, Microsoft Azure Text Analytics API) to analyze the question text. The input data is the user's question text, and the output data is the emotion data returned by the emotion engine. The emotion engine performs text analysis and extracts emotions such as anxiety and impatience.

[0921] Step 3:

[0922] Sending questions with sentiment data:

[0923] The device adds the emotion data extracted by the emotion engine to the question text and sends it to the server. The input data is the emotion data and the question text, and the output data is a data packet that integrates them. This data packet is sent to the server and received by the server.

[0924] Step 4:

[0925] Generate an answer:

[0926] The server sends the received question and emotion data to multiple generative AI models in parallel. The server calls the API of the generative AI model (e.g., OpenAI GPT-3, Hugging Face Transformers), and each model generates an answer using its own algorithm. The input data is the question with the emotion data added, and the output data is the answer returned by each generative AI model.

[0927] Step 5:

[0928] Peer assessment of answers:

[0929] The server sends the answers received from each generative AI model back to other generative AI models to evaluate the answers. Specifically, the server requests each answer from other AI models for evaluation, and each model returns an evaluation score. The input data is the answer generated by each generative AI model, and the output data is the evaluation score.

[0930] Step 6:

[0931] Counting the assessment results and sorting the responses:

[0932] The server aggregates the evaluation results returned by each generative AI model and selects the answer with the highest evaluation. The input data is the evaluation score, and the output data is the optimal answer. The evaluation scores are aggregated and analyzed within the server.

[0933] Step 7:

[0934] Final answer output:

[0935] The server sends the selected optimal answer to the user's terminal. The terminal receives this final answer and displays it to the user. Specifically, the server sends the optimal answer in packet form to the terminal, and the terminal's user interface displays it. The input data is the optimal answer, and the output data is the answer displayed to the user.

[0936] These processing steps allow users to receive emotionally sensitive and reliable answers.

[0937] The specific processing unit 290 transmits the result of the specific processing to the smart glasses 214. In the smart glasses 214, the control unit 46A causes the speaker 240 to output the result of the specific processing. The microphone 238 acquires audio indicating a user input regarding the result of the specific processing. The control unit 46A transmits audio data indicating the user input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the audio data.

[0938] The data generation model 58 is a so-called generative AI (Artificial Intelligence). An example of the data generation model 58 is ChatGPT (Internet Search<URL: https: / / openai.com / blog / chatgpt> ), Gemini (Internet search <url: https: gemini.google.com ?hl="ja">) and other generation AIs. The data generation model 58 is obtained by performing deep learning on a neural network. A prompt including an instruction is input to the data generation model 58, and inference data such as voice data indicating voice, text data indicating text, and image data indicating an image is also input. The data generation model 58 performs inference on the input inference data in accordance with the instruction indicated by the prompt, and outputs the inference result in a data format such as voice data and text data. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization.

[0939] In the above embodiment, an example in which the specific processing is performed by the data processing device 12 has been given, but the technology of the present disclosure is not limited to this, and the specific processing may be performed by the smart glasses 214.

[0940] [Third embodiment]

[0941] FIG. 5 shows an example of the configuration of a data processing system 310 according to the third embodiment.

[0942] 5, the data processing system 310 includes the data processing device 12 and a headset type terminal 314. An example of the data processing device 12 is a server.

[0943] The data processing device 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" according to the technology of the present disclosure. The computer 22 includes a processor 28, a RAM 30, and a storage 32. The processor 28, the RAM 30, and the storage 32 are connected to a bus 34. The database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. Examples of the network 54 include a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0944] The headset type terminal 314 includes a computer 36, a microphone 238, a speaker 240, a camera 42, a communication I / F 44, and a display 343. The computer 36 includes a processor 46, a RAM 48, and a storage 50. The processor 46, the RAM 48, and the storage 50 are connected to a bus 52. The microphone 238, the speaker 240, the camera 42, and the display 343 are also connected to the bus 52.

[0945] The microphone 238 receives instructions and the like from the user 20 by receiving voice uttered by the user 20. The microphone 238 captures the voice uttered by the user 20, converts the captured voice into audio data, and outputs it to the processor 46. The speaker 240 outputs audio in accordance with instructions from the processor 46.

[0946] Camera 42 is a small digital camera equipped with an optical system including a lens, aperture, and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, and captures images of the surroundings of user 20 (for example, an imaging range defined by an angle of view equivalent to the field of vision of a typical healthy person).

[0947] The communication I / F 44 is connected to a network 54. The communication I / Fs 44 and 26 are responsible for the exchange of various information between the processor 46 and the processor 28 via the network 54. The exchange of various information between the processor 46 and the processor 28 using the communication I / Fs 44 and 26 is carried out in a secure state.

[0948] Fig. 6 shows an example of the main functions of the data processing device 12 and the headset type terminal 314. As shown in Fig. 6, in the data processing device 12, a specific process is performed by the processor 28. A specific process program 56 is stored in the storage 32.

[0949] The specific processing program 56 is an example of a "program" according to the technology of the present disclosure. The processor 28 reads the specific processing program 56 from the storage 32 and executes the read specific processing program 56 on the RAM 30. The specific processing is realized by the processor 28 operating as a specific processing unit 290 in accordance with the specific processing program 56 executed on the RAM 30.

[0950] The storage 32 stores a data generation model 58 and an emotion identification model 59. The data generation model 58 and the emotion identification model 59 are used by the identification processing unit 290.

[0951] In the headset type terminal 314, a reception output process is performed by the processor 46. A reception output program 60 is stored in the storage 50. The processor 46 reads the reception output program 60 from the storage 50 and executes the read reception output program 60 on the RAM 48. The reception output process is realized by the processor 46 operating as the control unit 46A in accordance with the reception output program 60 executed on the RAM 48.

[0952] Next, a description will be given of the identification process performed by the identification processing unit 290 of the data processing device 12. In the following description, the data processing device 12 will be referred to as the "server" and the headset type terminal 314 will be referred to as the "terminal."

[0953] This invention relates to a generative AI platform for improving the accuracy of answers generated by AI. This platform is a system in which multiple generative AI models cooperate to generate the most correct answers through mutual evaluation and discussion. Below, we will explain the program processing of this system in natural language using concrete examples.

[0954] Program processing explanation

[0955] Enter your question

[0956] A user uses an email client to type a question, for example, "What's the weather like in Tokyo tomorrow?"

[0957] Generate answers

[0958] The server receives this question and sends it to several different generative artificial intelligence models (AIM1, AIM2, AIM3, etc.), each of which uses its own algorithm to generate an answer.

[0959] AIM1: "It's sunny."

[0960] AIM2: "Tomorrow will be sunny."

[0961] AIM3: "It's forecast to be sunny in Tokyo."

[0962] Request for evaluation of answer

[0963] The server sends the generated answers to other generative AI models in the Validator Pool, allowing each model to evaluate the answers of the other generative AI models.

[0964] Evaluating answers

[0965] Each generative AI model evaluates the answers generated by other models and assigns them a score based on the accuracy, consistency, and completeness of the answers.

[0966] Aggregation of evaluation results

[0967] The server aggregates the results from each model and sees which answer received the highest rating.

[0968] Notification of evaluation results and request for discussion

[0969] The server notifies each generative AI model of the evaluation results and requests them to correct or supplement their answers as necessary.

[0970] Editing and discussing answers

[0971] Each model modifies its answer based on the discussion and adds any necessary corrections. For example, AIM1 can update its answer as follows:

[0972] AIM1: "Tomorrow will be sunny in Tokyo, with temperatures around 25 degrees."

[0973] AIM2: "Tomorrow in Tokyo will be sunny with a maximum temperature of 25 degrees."

[0974] AIM3: "It'll be sunny and 25 degrees."

[0975] Final evaluation and majority vote

[0976] The server then re-evaluates the updated answer against the models in the Validator Pool. Each model selects the most appropriate answer and a majority vote is held.

[0977] Selection of the final answer

[0978] Based on the evaluation results, the server selects the final answer. The final answer obtained through this process is "Tomorrow in Tokyo, it will be sunny and the temperature will be around 25 degrees."

[0979] Submit your answer

[0980] The server sends the final selected answer to the terminal.

[0981] Show Answers

[0982] The terminal displays the final answer to the user.

[0983] This system allows multiple generative AI models to cooperate and generate more accurate and reliable answers through mutual evaluation and discussion. As can be seen from this explanation with concrete examples, the present invention provides an effective means for suppressing hallucination and improving the accuracy of information.

[0984] The processing flow will be explained below.

[0985] Step 1:

[0986] The user inputs a question, for example, "What's the weather like in Tokyo tomorrow?" into the terminal.

[0987] Step 2:

[0988] The terminal transmits the entered question to the server.

[0989] Step 3:

[0990] The server receives questions and distributes them to multiple generative artificial intelligence models (AIM1, AIM2, AIM3).

[0991] Step 4:

[0992] Each generative AI model generates an answer to a question sent by the server using its own algorithm. For example,

[0993] AIM1: "It's sunny."

[0994] AIM2: "Tomorrow will be sunny."

[0995] AIM3: "It's forecast to be sunny in Tokyo."

[0996] Step 5:

[0997] The server sends the generated answer to each of the other generative AI models in the Validator Pool for evaluation.

[0998] Step 6:

[0999] Each generative AI model evaluates the answers it receives from other models and assigns a score based on the accuracy, consistency, and completeness of the answers.

[1000] Step 7:

[1001] The server aggregates the evaluation results from each model and analyzes which answer received the highest evaluation.

[1002] Step 8:

[1003] The server notifies each generative AI model of the evaluation results and, if necessary, requests them to correct or supplement their answers.

[1004] Step 9:

[1005] Each generative AI model modifies the answer based on the evaluation results and adds correction information. For example,

[1006] AIM1: "Tomorrow will be sunny in Tokyo, with temperatures around 25 degrees."

[1007] AIM2: "Tomorrow in Tokyo will be sunny with a maximum temperature of 25 degrees."

[1008] AIM3: "It'll be sunny and the temperature will be 25 degrees."

[1009] Step 10:

[1010] The server then sends the updated answers to the Validator Pool again, asking each model to re-evaluate.

[1011] Step 11:

[1012] Each generative AI model then evaluates the answers again, selects the most appropriate answer, and assigns it a score.

[1013] Step 12:

[1014] The server tally the results of the reevaluation and selects the answer with the highest score. For example, "Tomorrow in Tokyo, it will be sunny and the temperature will be around 25 degrees."

[1015] Step 13:

[1016] The server sends the final selected answer to the terminal.

[1017] Step 14:

[1018] The terminal displays the final answer to the user.

[1019] Example 1

[1020] Next, a description will be given of Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the headset type terminal 314 will be referred to as a "terminal."

[1021] In existing generative AI models, relying on a single model's answer can reduce the accuracy and reliability of the answer. Furthermore, when the answers generated by each model differ, it is difficult to determine which answer is the most accurate. There is a need to solve this problem and provide more accurate and reliable answers.

[1022] The specific processing by the specific processing unit 290 of the data processing device 12 in the first embodiment is realized by the following means.

[1023] In this invention, the server includes means for inputting a question, means for multiple generative AI models to each generate an answer to the question, means for other generative AI models to evaluate the answers generated by each generative AI model, means for selecting the most appropriate answer based on the evaluation results, means for correcting or supplementing the selected answer, means for re-evaluating the corrected answer, means for performing the evaluation process in stages and selecting a final answer based on the result of majority voting, and means for transmitting and displaying the selected final answer to a user terminal. This makes it possible to provide users with more accurate and reliable final answers through mutual evaluation and discussion among multiple generative AI models.

[1024] The "means for inputting a question" provides an interface for a user to input a question to the system.

[1025] A "generative artificial intelligence model" refers to an algorithm or machine learning model that generates answers in natural language to input questions.

[1026] The "means for generating an answer" is what enables the generative artificial intelligence model to generate an answer to an input question.

[1027] "Means for evaluating answers" refers to a function in which answers generated by a generative AI model are evaluated by other generative AI models and scored for accuracy and consistency.

[1028] The "means for selecting the most appropriate answer" includes a function for selecting the answer that has received the highest evaluation based on the evaluation results of each generative artificial intelligence model.

[1029] "Means for correcting or supplementing answers" refers to the functionality for adding necessary corrections or additional information to selected answers.

[1030] "Means for re-evaluation" refers to a function that allows each generative artificial intelligence model to re-evaluate an answer that has been corrected or supplemented.

[1031] "Means for carrying out the evaluation process in stages" includes a function for dividing the evaluation into multiple stages and carrying out a final evaluation based on the evaluation results at each stage.

[1032] "Means for selecting a final answer based on the results of a majority vote" refers to a function for selecting the answer that receives the most support as the final answer based on the evaluation results of each generative AI model.

[1033] The "means for transmitting and displaying the final answer" provides a function for transmitting the selected final answer to the user terminal and displaying it.

[1034] This invention relates to a generative AI platform for improving the accuracy of answers generated by AI. This platform is a system in which multiple generative artificial intelligence models (generative AI models) cooperate to generate the most correct answers through mutual evaluation and discussion.

[1035] Hardware and software used

[1036] The hardware required to implement this system includes a high-performance server and user terminals (e.g., PCs and smartphones). The server has high computing power and storage capacity, and a cloud platform for running the generative AI models can also be used. The software includes each generative AI model, an email client, a web form, etc.

[1037] Data processing and calculation

[1038] When a user enters a question using an email client or web form, the question is stored in a database and sent to a server, which parses the question and converts it into a prompt that the generative AI model can understand.

[1039] For example, in response to the question "What's the weather like in Tokyo tomorrow?", the server generates a prompt sentence, "Please tell me the weather forecast for Tokyo tomorrow," and sends this to multiple generative AI models (AIM1, AIM2, AIM3, etc.).

[1040] Each generative AI model receives a prompt and uses its own algorithms and datasets to generate an answer.

[1041] AIM1: "It's sunny."

[1042] AIM2: "Tomorrow will be sunny."

[1043] AIM3: "It's forecast to be sunny in Tokyo."

[1044] The generated answers are sent by the server to other generative AI models in the Validator Pool, where each model evaluates the answers of the other models based on the accuracy, consistency, and completeness of the answers.

[1045] The evaluation results are tallied, and the server checks which answer received the highest rating. For example, if AIM1 received the highest rating, that score is recorded. The evaluation results are notified to each generative AI model, which corrects or supplements the answer as necessary. For example, if AIM1 is evaluated as having insufficient temperature information, additional information is provided.

[1046] Each generative AI model modifies its answer based on the discussion and adds any necessary corrections. Answers may be updated as follows:

[1047] AIM1: "Tomorrow will be sunny in Tokyo, with temperatures around 25 degrees."

[1048] AIM2: "Tomorrow in Tokyo will be sunny with a maximum temperature of 25 degrees."

[1049] AIM3: "It'll be sunny and 25 degrees."

[1050] The server then evaluates the updated answers again using the models in the Validator Pool, and each generative AI model selects the most appropriate answer through a majority vote. Based on the evaluation results, the server selects the final answer: "Tomorrow in Tokyo will be sunny, with a temperature of around 25 degrees."

[1051] Examples of specific examples and prompts

[1052] As a specific example, the following shows how a generative AI model generates and evaluates answers to the input, "What's the weather like in Tokyo tomorrow?"

[1053] User input: "What's the weather like in Tokyo tomorrow?"

[1054] Server prompt: "What's the weather forecast for Tokyo tomorrow?"

[1055] Example answer from a generative AI model:

[1056] AIM1: "It's sunny."

[1057] AIM2: "Tomorrow will be sunny."

[1058] AIM3: "It's forecast to be sunny in Tokyo."

[1059] Example of an answer after evaluation and revision:

[1060] AIM1: "Tomorrow will be sunny in Tokyo, with temperatures around 25 degrees."

[1061] AIM2: "Tomorrow in Tokyo will be sunny with a maximum temperature of 25 degrees."

[1062] AIM3: "It'll be sunny and 25 degrees."

[1063] Ultimately, the answer selected by the server based on the majority vote was: "Tomorrow in Tokyo, it will be sunny, and the temperature will be around 25 degrees."

[1064] Through the collaboration and mutual evaluation of multiple generative AI models, this system enables the provision of more accurate and reliable information to users.

[1065] The flow of the identification process in the first embodiment will be described with reference to FIG.

[1066] Step 1:

[1067] A user enters a question using an email client or web form, for example, "What's the weather like in Tokyo tomorrow?" This input is sent to the server.

[1068] Input: User question: "What's the weather like in Tokyo tomorrow?"

[1069] Output: Query data sent to the server

[1070] Step 2:

[1071] The server analyzes the received question and converts it into a prompt that the generative AI model can understand. For example, it generates a prompt such as, "Please tell me the weather forecast for Tokyo tomorrow."

[1072] Input: Question data "What's the weather like in Tokyo tomorrow?"

[1073] Output: Prompt "What is the weather forecast for Tokyo tomorrow?"

[1074] Step 3:

[1075] The server sends the generated prompt to multiple generative AI models (AIM1, AIM2, AIM3, etc.), each of which receives the prompt and generates an answer using its own algorithm and dataset.

[1076] Input: Prompt "What is the weather forecast for Tokyo tomorrow?"

[1077] Output: Answer of each generative AI model (e.g., AIM1: "It's sunny," AIM2: "It will be sunny tomorrow," AIM3: "It is forecast to be sunny in Tokyo.")

[1078] Step 4:

[1079] The server sends the generated answers to other generative AI models in the Validator Pool, asking each model to evaluate the answers of the other models.

[1080] Input: The generative AI model's answer

[1081] Output: Evaluation request data

[1082] Step 5:

[1083] Generative AI models evaluate the answers generated by other generative AI models. For example, AIM2 looks at the answers of AIM1 and AIM3 and assigns a score to each. The evaluation criteria are the accuracy, consistency, and completeness of the information.

[1084] Input: Answers from other generative AI models

[1085] Output: Evaluation score (e.g., AIM2 scoring)

[1086] Step 6:

[1087] The server aggregates the results from each model and checks which answer received the highest score. For example, if AIM1 received the highest score, that score is recorded.

[1088] Input: Rating score

[1089] Output: The answer from the model with the highest score

[1090] Step 7:

[1091] The server notifies each generative AI model of the evaluation results and asks them to correct or supplement their answers as necessary. For example, if AIM1 receives an evaluation that "temperature information is insufficient," it will ask them to provide additional information.

[1092] Input: Evaluation result

[1093] Output: Notification of correction request

[1094] Step 8:

[1095] The generative AI model then modifies the answer based on the discussion and adds any necessary corrections. For example, AIM1 updates the answer to "Tomorrow will be sunny in Tokyo, with the temperature around 25 degrees."

[1096] Input: Notification of correction request

[1097] Output: Updated answer (e.g. AIM1: "Tomorrow will be sunny in Tokyo, with the temperature around 25 degrees.")

[1098] Step 9:

[1099] The server then re-evaluates the updated answers with the models in the Validator Pool. Each generative AI model selects the most appropriate answer and a majority vote is held.

[1100] Input: Updated answer

[1101] Output: Reassessment score and majority vote result

[1102] Step 10:

[1103] The server selects the final answer based on the evaluation results. For example, it selects "Tomorrow in Tokyo will be sunny, with the temperature around 25 degrees" as the most popular answer.

[1104] Input: Reassessment score and majority vote result

[1105] Output: Final answer "Tomorrow in Tokyo it will be sunny, with the temperature around 25 degrees."

[1106] Step 11:

[1107] The server sends the final selected answer to the terminal.

[1108] Input: Final answer

[1109] Output: Data sent to the terminal

[1110] Step 12:

[1111] The terminal displays the final answer to the user, who can view it through their email client or web interface.

[1112] Input: Final response data

[1113] Output: What is displayed to the user

[1114] This series of processes enables the provision of more accurate and reliable information to users through cooperation and mutual evaluation of multiple generative AI models.

[1115] (Application example 1)

[1116] Next, a description will be given of Application Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the headset type terminal 314 will be referred to as a "terminal."

[1117] Conventional answer generation systems using generative AI models often lack consistency and reliability, with the accuracy of each model varying. Furthermore, especially in the security field, where incorrect answers can have significant impacts, higher accuracy and reliability are required. For this reason, there was a need to develop a system in which multiple AI models cooperate and generate the most accurate and reliable answers through mutual evaluation and discussion.

[1118] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 1 is realized by the following means.

[1119] In this invention, the server includes means for multiple generative AI models to generate answers to questions, means for other generative AI models to evaluate the answers generated by each generative AI model, and means for selecting the most appropriate answer based on the evaluation results. This improves the accuracy and reliability of the generated answers, enabling reliable notifications to users, particularly when used as security alerts.

[1120] "Multiple generative artificial intelligence models" refers to multiple AI models that use different algorithms and datasets to generate answers in their own unique way.

[1121] "Means for generating answers to each question" refers to the process or function by which each AI model generates an individual answer to a question input by a user.

[1122] "Means for evaluation by other generative artificial intelligence models" refers to the process or function by which answers generated by one AI model are evaluated by other AI models to determine their accuracy and consistency.

[1123] "Means for selecting the most appropriate answer" refers to the process or function of selecting the answer that received the highest rating based on the results of multiple evaluations.

[1124] "Means for outputting the selected answer as the final answer" refers to the process or function of providing the selected most appropriate answer to the user as the final answer.

[1125] "Means for holding discussions, adding corrections, and updating answers" refers to the process or function of discussing the answers generated by each AI model and making necessary corrections to generate more accurate answers.

[1126] "A method of conducting a step-by-step evaluation process and selecting a final answer based on the results of a majority vote" refers to a process of going through a series of evaluation and selection steps in a step-by-step manner, ultimately selecting the answer supported by the majority of evaluators.

[1127] A "security alert" refers to a warning message that notifies the user of an abnormal security situation or abnormality.

[1128] "Means of notifying the user" refers to the process or function of sending the final selected answer or alert to the user's device and notifying the user by visual, auditory, or other means.

[1129] This invention relates to a system in which multiple generative artificial intelligence models cooperate to generate the most correct answer through mutual evaluation and discussion. Specifically, this invention describes an application for providing accurate alerts to users in security services.

[1130] Hardware and software used

[1131] Hardware:

[1132] 1. Server: Runs multiple AI models and manages the evaluation process.

[1133] 2. Security cameras: collect data and provide it for analysis.

[1134] 3. Smartphone: The device where the user enters the question and receives the final answer.

[1135] software:

[1136] 1. Python: A system-wide programming language.

[1137] 2. Requests library: Manages API communication.

[1138] 3. AI Models: Multiple generative AI models that use various algorithms to generate solutions to problems.

[1139] System processing flow

[1140] 1. User enters question:

[1141] Users use a smartphone app to enter specific questions, such as "Can you check if there's an intruder in my office?"

[1142] 2. The server receives the query:

[1143] The server receives this question and sends it to multiple generative AI models, each of which uses its own algorithm to generate an answer.

[1144] 3. The AI ​​model generates the answer:

[1145] Each generative AI model generates an answer to the question, for example:

[1146] Model 1: "Motion detected on camera, but it doesn't seem to be an intruder."

[1147] Model 2: "It looks like the door is open, but we need to check it out immediately."

[1148] Model 3: "Motion detected, but no further information available."

[1149] 4. Other models rate your answers:

[1150] The server sends these answers to other generative AI models, which then evaluate the answers of the other models based on the accuracy and consistency of the answers.

[1151] 5. Compilation of evaluation results and selection of best answers:

[1152] Based on the evaluation results of each AI model, the answer with the highest score will be selected. For example, the evaluation will be as follows:

[1153] "Motion detected on camera, but it doesn't appear to be an intruder" received the highest rating.

[1154] 6. Generate final answer and notify user:

[1155] The server then sends the final selected answer to the user's device and notifies them, allowing the user to receive accurate and reliable security alerts.

[1156] Prompt Sentence Examples

[1157] User input:

[1158] "Check if there's an intruder in the office."

[1159] This invention enables multiple generative AI models to cooperate to generate the most appropriate response and notify the user of it as a security alert, allowing the user to take appropriate action based on highly reliable information.

[1160] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[1161] Step 1:

[1162] A user uses a smartphone to input a question, for example, "Please check if there is an intruder in my office." This input is sent to the server through the user interface.

[1163] Step 2:

[1164] The server receives a user's question and sends it to multiple generative AI models. Each model receives the question and generates an answer using its own algorithm. A typical answer might be, "Motion was detected on the camera, but it doesn't appear to be an intruder."

[1165] Step 3:

[1166] The server sends the answers generated by each generative AI model to other generative AI models for evaluation. At this time, the input provided to each model is a list of generated answers, and each model evaluates this list.

[1167] Step 4:

[1168] Each generative AI model evaluates the answers generated by other models and assigns a score based on their accuracy, consistency, and completeness of information. The evaluation results are sent to the server in the form of a score. For example, model A may assign 90 points to the answer of model B and 80 points to the answer of model C.

[1169] Step 5:

[1170] The server then selects the answer with the highest score based on the aggregated evaluation results. For example, if "Motion was detected on the camera, but it doesn't appear to be an intruder" receives the highest score, it will be selected.

[1171] Step 6:

[1172] The server then sends the selected answer to the user's smartphone. This notification includes the most highly rated answer and a brief explanation of it. The user can then check the answer on their device.

[1173] Step 7:

[1174] If necessary, the generated answers may be further discussed and corrections added. Each generative AI model will be discussed again and generate a new answer with the necessary corrections. This process of evaluation and correction is repeated until the revised best answer is selected.

[1175] The above is a specific flow of processing by the program that realizes this invention, which enables users to receive accurate and reliable security alerts.

[1176] Furthermore, an emotion engine that estimates the user's emotion may be further combined. That is, the identification processing unit 290 may estimate the user's emotion using the emotion identification model 59, and perform identification processing using the user's emotion.

[1177] This invention relates to a generative AI platform that uses multiple generative AI models and an emotion engine to generate accurate and appropriate answers to user questions. Each generative AI model generates and modifies answers through mutual evaluation and discussion, and by combining it with an emotion engine that recognizes the user's emotions, it provides answers that are adapted to the user's emotional state. Below, we will explain the program processing of this system in natural language using concrete examples.

[1178] Program processing explanation

[1179] Enter your question

[1180] The user uses the terminal to input a question, for example, "Tell me what my friends think."

[1181] Emotion recognition

[1182] The device sends the user's input, voice, facial expression, and other data to the emotion engine, which analyzes this data and recognizes the user's emotions. For example, it can detect "anxiety."

[1183] Submit a Question

[1184] The device sends the question with the recognized emotion data to the server.

[1185] Generate answers

[1186] The server distributes the emotional data attached to the question to multiple generative AI models (AIM1, AIM2, AIM3). Each model takes the emotional data into account and generates an answer using its own algorithm. For example,

[1187] AIM1: "Your friends care about you."

[1188] AIM2: "Your friends are worried about you."

[1189] AIM3: "Your friends trust you."

[1190] Request for evaluation of answer

[1191] The server sends the generated answers to other generative AI models, allowing each model to evaluate the answers of the other generative AI models.

[1192] Evaluating answers

[1193] The generative AI model evaluates the answers generated by other models and assigns them a score based on the answer's accuracy, consistency, completeness of information, and fit to the perceived emotion.

[1194] Aggregation of evaluation results

[1195] The server aggregates the evaluation results from each model and analyzes which answer received the highest evaluation.

[1196] Notification of evaluation results and request for discussion

[1197] The server notifies each generative AI model of the evaluation results and requests them to revise or supplement their answers if necessary. Emotional data is also reflected.

[1198] Editing and discussing answers

[1199] Each generative AI model modifies its answers based on the evaluation results and emotional data, adding correction information. For example,

[1200] AIM1: "Friends care about you and always want to support you."

[1201] AIM2: "Your friend is very concerned about you and wants to talk more."

[1202] AIM3: "Your friends trust you and want to talk to you if they have a problem."

[1203] Final evaluation and majority vote

[1204] The server re-evaluates the updated answer and asks each model to re-evaluate.

[1205] Selection of the final answer

[1206] Based on the reevaluation results, the server selects a final answer, such as "Your friends care about you and always want to support you."

[1207] Submit your answer

[1208] The server sends the final selected answer to the terminal.

[1209] Show Answers

[1210] The terminal displays the final answer to the user.

[1211] This system enables multiple generative AI models to generate more accurate and reliable answers that adapt to the user's emotional state through mutual evaluation and discussion. As can be seen from the explanation using concrete examples, the present invention provides an innovative means for suppressing hallucination and providing information that takes into consideration the user's emotions.

[1212] The processing flow will be explained below.

[1213] Step 1:

[1214] The user inputs a question, for example, "What's the weather like in Tokyo tomorrow?" into the terminal.

[1215] Step 2:

[1216] The device sends the input question as data to the emotion engine, which analyzes the user's input, voice, facial expressions, and other data to recognize the user's emotions. For example, it can detect "anxiety."

[1217] Step 3:

[1218] The emotion engine returns the recognized emotion data to the terminal.

[1219] Step 4:

[1220] The device sends the question with the emotional data attached to it to the server.

[1221] Step 5:

[1222] The server receives questions and emotion data and distributes them to multiple generative AI models (AIM1, AIM2, AIM3).

[1223] Step 6:

[1224] Each generative AI model uses its own algorithm to generate answers to questions while taking into account emotional data. For example,

[1225] AIM1: "It's sunny."

[1226] AIM2: "Tomorrow will be sunny."

[1227] AIM3: "It's forecast to be sunny in Tokyo."

[1228] Step 7:

[1229] The server compiles the generated answers and sends them to other generative AI models, allowing each model to evaluate the answers of the other generative AI models.

[1230] Step 8:

[1231] Each generative AI model evaluates the answers sent by other generative AI models and assigns a score based on the answer's accuracy, consistency, completeness of information, and fit to the perceived emotion.

[1232] Step 9:

[1233] The server aggregates the evaluation results from each model and analyzes which answer received the highest evaluation.

[1234] Step 10:

[1235] The server notifies each generative AI model of the evaluation results and requests them to revise or supplement their answers as necessary. This emotional data is also reflected in the re-evaluation.

[1236] Step 11:

[1237] Each generative AI model modifies its answers based on the evaluation results and emotional data, adding correction information. For example,

[1238] AIM1: "Tomorrow will be sunny in Tokyo. The temperature will be around 25 degrees."

[1239] AIM2: "Tomorrow in Tokyo will be sunny with a maximum temperature of 25 degrees."

[1240] AIM3: "It'll be sunny and the temperature will be 25 degrees."

[1241] Step 12:

[1242] The server then distributes the updated answers to each generative AI model again and requests them to evaluate again.

[1243] Step 13:

[1244] Each generative AI model then evaluates the answers again, selects the most appropriate answer, and assigns it a score.

[1245] Step 14:

[1246] The server tally the results of the reevaluation and selects the answer with the highest score. For example, "Tomorrow in Tokyo, it will be sunny and the temperature will be around 25 degrees."

[1247] Step 15:

[1248] The server sends the final selected answer to the terminal.

[1249] Step 16:

[1250] The terminal displays the final answer to the user.

[1251] Example 2

[1252] Next, a description will be given of Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the headset type terminal 314 will be referred to as a "terminal."

[1253] In conventional generative AI systems, the answers generated are not always adapted to the user's emotional state, which reduces user satisfaction. Furthermore, the evaluation process for multiple generative AI models is insufficient, resulting in answers that lack consistency and accuracy. Furthermore, the generated answers do not reflect correction information or discussions, which can result in the delivery of less than optimal answers.

[1254] The specific processing by the specific processing unit 290 of the data processing device 12 in the second embodiment is realized by the following means.

[1255] In this invention, the server includes means for multiple generative AI models to generate answers to questions, means for other generative AI models to evaluate the answers generated by each generative AI model, means for selecting the most appropriate answer based on the evaluation results, an emotion engine that recognizes the user's emotion using data input by the user, and means for attaching the emotion data recognized by the emotion engine to the question and transmitting it, thereby making it possible to generate and provide accurate and appropriate answers that are adapted to the user's emotional state.

[1256] "Multiple generative AI models" are multiple AI systems with different algorithms and settings, each capable of generating answers independently.

[1257] A "question" is any information or question entered by a user that requires an answer from a generative AI model.

[1258] The "means for generating an answer" refers to the process or algorithm that the generative artificial intelligence model uses to create an appropriate answer to a user's question.

[1259] "Means for evaluation by other generative artificial intelligence models" refers to the process by which each generative artificial intelligence model evaluates answers generated by other models and assigns them a score or ranking.

[1260] The "means of selecting the most appropriate answer based on the evaluation results" is a system that aggregates the evaluation results obtained from multiple generative AI models and selects the answer that receives the highest evaluation.

[1261] An "emotion engine" is a system that analyzes the user's input data, voice, facial expressions, etc. to recognize the user's emotional state.

[1262] "Emotion data" is data showing the results of analysis by the user's emotion engine, and represents an emotional state such as anxiety or relief.

[1263] The "means of transmission" is the process or method of transmitting data from the device to the server, often using a secure communications protocol.

[1264] "Correction information" is information that is added to the initially generated answer, and is used to improve the accuracy and appropriateness of the answer.

[1265] "Means for discussion" refers to a process in which multiple generative AI models discuss the generated answers and make optimal improvements or corrections.

[1266] "Majority vote" is a method of statistically aggregating the results of multiple evaluations and scoring and selecting the answer that receives the most support.

[1267] The system starts when a user uses a device to input a question. For example, "Tell me what my friend thinks." The user can use a keyboard or touch screen to input a question. The system may also collect voice and facial expression data, which are sent to the emotion engine.

[1268] The emotion engine analyzes the collected data and recognizes the user's emotions. Voice recognition technology and facial expression analysis algorithms are used to recognize emotions, such as "anxiety." The emotion data recognized in this way is attached to the user's question and sent from the device to the server. A secure communication protocol (e.g., HTTPS) is used to protect the data.

[1269] The server distributes the received question and emotion data to multiple generative AI models (e.g., AIM1, AIM2, AIM3). Each model generates an answer using its own natural language processing algorithm. Examples of generated answers include:

[1270] AIM1: "Your friends care about you."

[1271] AIM2: "Your friends are worried about you."

[1272] AIM3: "Your friends trust you."

[1273] The server then sends the answers generated by each generative AI model to other generative AI models, who evaluate each other's answers based on the accuracy, consistency, completeness of the information, and suitability to the emotional data. For example, AIM1 might give 70 points to AIM2's answer and 80 points to AIM3's answer.

[1274] The server aggregates these evaluation results and selects the answer with the highest rating. After selection, the server notifies each generative AI model of the evaluation results and requests them to revise or supplement the answer as necessary. Emotional data is also reflected.

[1275] Each generative AI model modifies its answers based on the evaluation results and emotional data, adding correction information. For example, it may be modified as follows:

[1276] AIM1: "Friends care about you and always want to support you."

[1277] AIM2: "Your friend is very concerned about you and wants to talk more."

[1278] AIM3: "Your friends trust you and want to talk to you if they have a problem."

[1279] Finally, the server re-evaluates the updated answers and selects the answer with the most votes. The final answer is sent to the device and displayed to the user, providing a reliable answer to the question.

[1280] For example, consider the following prompt:

[1281] "Please tell me what my friend thinks. I'm feeling anxious right now."

[1282] This system enables multiple generative AI models to provide more accurate and reliable answers that adapt to the user's emotional state through mutual evaluation and discussion, thereby improving user satisfaction and the quality of information provided.

[1283] The flow of the identification process in the second embodiment will be described with reference to FIG.

[1284] Step 1: The user enters a question using a terminal.

[1285] Input: Text data entered by the user using a keyboard or touchscreen.

[1286] Output: Question text data imported to the terminal.

[1287] Specific operation: The user inputs a question such as "Please tell me what my friends think." The device collects the text data of the question.

[1288] Step 2: The device sends the user's input data to the emotion engine and recognizes the user's emotion.

[1289] Input: Text data of the user's question acquired by the device, as well as voice and facial expression data.

[1290] Output: Emotion data parsed by the emotion engine.

[1291] Specific operation: The device sends the user's text data, voice, and facial expressions to the emotion engine, which analyzes them and recognizes emotions such as "anxiety."

[1292] Step 3: The device sends the question accompanied by the recognized emotion data to the server.

[1293] Input: User question text data and sentiment data.

[1294] Output: Questions and sentiment data sent to the server.

[1295] Specific operation: The device attaches emotional data obtained from the emotion engine to the question and sends it securely to the server using HTTPS.

[1296] Step 4: The server distributes the questions and emotion data to multiple generative AI models.

[1297] Input: Question and emotion data received by the server.

[1298] Output: The questions and sentiment data delivered to each generative AI model.

[1299] Specific operation: The server distributes the received question and emotion data to the generative AI models (e.g., AIM1, AIM2, AIM3), and each model generates an answer.

[1300] Step 5: Each generative AI model generates an answer to the question.

[1301] Input: Questions and sentiment data delivered to each generative AI model.

[1302] Output: The generated answer.

[1303] How it works: The generative AI model uses a proprietary natural language processing algorithm to generate answers. For example, AIM1 generates "Your friends care about you."

[1304] Step 6: The server sends the generated answer to other generative AI models for evaluation.

[1305] Input: The answers generated by each generative AI model.

[1306] Output: Answer evaluation request sent to other generative AI models.

[1307] Specific operation: The server collects the answers generated by each model and sends them to other generative AI models for evaluation.

[1308] Step 7: Each generative AI model evaluates the answers generated by other models and assigns a score.

[1309] Input: Answers received from other generative artificial intelligence models.

[1310] Output: Evaluation score.

[1311] How it works: Each generative AI model evaluates the answers of other models based on specified criteria and assigns a score based on, for example, accuracy or consistency.

[1312] Step 8: The server aggregates the evaluation results of each generative AI model and selects the answer with the highest evaluation.

[1313] Input: Evaluation scores from all generative AI models.

[1314] Output: The answer that received the most votes.

[1315] Specific operation: The server aggregates the evaluation results and selects the answer with the highest score.

[1316] Step 9: The server notifies each generative AI model of the selected answers and requests corrections or supplements if necessary.

[1317] Input: Screened answers and assessment results.

[1318] Output: Request for corrections and clarifications.

[1319] Specific operation: The server sends the selected answers and their evaluation scores to the model, requesting corrections or supplements as necessary.

[1320] Step 10: Each generative AI model modifies the answer and adds correction information.

[1321] Input: Correction request and original answer from the server.

[1322] Output: The revised answer and any corrections added.

[1323] Specific behavior: Each model modifies the answer based on the evaluation results, adding correction information such as, "Your friends care about you and always want to support you."

[1324] Step 11: The server re-evaluates the revised answers and selects the final answer based on the majority vote.

[1325] Input: Revised answer.

[1326] Output: The final selected answer.

[1327] Specific operation: The server has each generative AI model evaluate the revised answer again, and selects the final answer by majority vote based on the evaluation results.

[1328] Step 12: The server sends the final selected answer to the terminal and displays it to the user.

[1329] Input: Final selected answer.

[1330] Output: The final answer displayed on the terminal.

[1331] Specific operation: The server securely transmits the final selected answer to the terminal, which then displays it to the user, allowing the user to obtain a reliable answer.

[1332] (Application example 2)

[1333] Next, a description will be given of Application Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the headset type terminal 314 will be referred to as a "terminal."

[1334] Conventional generative AI systems have difficulty generating responses that take user emotions into account, and in situations such as security support, it is necessary to respond appropriately to user anxieties and concerns. Therefore, a system that can recognize user emotions and provide appropriate responses based on those emotions is needed.

[1335] The identification process by the identification processing unit 290 of the data processing device 12 in Application Example 2 is realized by the following means. In this invention, the server includes means for sending a question entered by a user to an emotion engine and recognizing the user's emotion, means for sending the question accompanied by emotion data to the server, and means for the server to cause multiple generation year artificial intelligence models to generate answers taking the emotion data into consideration. This makes it possible to provide appropriate and reliable answers that take the user's emotions into consideration.

[1336] "Multiple Generation Year AI Models" refers to multiple AI systems that operate based on different algorithms and data sets and generate unique answers to questions.

[1337] "Means for generating answers to each question" refers to the function by which each AI model generates its own answer to a question entered by a user.

[1338] "Means for evaluation by other generation year artificial intelligence models" refers to the function by which answers generated by one generation year artificial intelligence model are evaluated by other generation year artificial intelligence models.

[1339] "Means for selecting the most appropriate answer" refers to the function of selecting the most appropriate answer based on the results of evaluations by multiple AI models.

[1340] "Means for outputting the selected answer as the final answer" refers to a function that provides the user with the selected most appropriate answer.

[1341] An "emotion engine" is a system that analyzes data such as user input, voice, and facial expressions to recognize the user's emotions.

[1342] The "means for transmitting a question with emotional data attached to it to a server" refers to a function for adding emotional data recognized by an emotion engine to question data and transmitting it to a server.

[1343] "Means for the server to generate answers taking into account emotional data" refers to the function of the server to have multiple artificial intelligence models generate answers based on emotional data.

[1344] This invention shows how to implement a system that recognizes a user's emotions and provides appropriate answers based on those emotions. This system is composed of multiple AI models, an emotion engine, and a server.

[1345] Program Generation

[1346] The system runs a program that includes the following functions:

[1347] Processing Description

[1348] 1. User enters question:

[1349] A user uses a terminal to input a question. For example, the user inputs the question, "My security camera isn't working. What should I do?"

[1350] 2. Emotion Recognition:

[1351] The device sends the text data entered by the user to the emotion engine, which uses natural language processing (NLP) technology to analyze the user's emotions and extract emotional data such as anxiety or impatience. The emotion engine can use, for example, Google Cloud's NLP API or Microsoft Azure's Text Analytics API.

[1352] 3. Sending questions with emotion data:

[1353] The terminal adds the emotion data extracted by the emotion engine and sends the question to the server.

[1354] 4. Generate answers:

[1355] The server distributes questions that take emotion data into account to multiple generative AI models, each of which uses its own algorithm to generate answers, including OpenAI's GPT-3 and GPT-4 models and the Hugging Face Transformers model.

[1356] 5. Peer assessment of answers:

[1357] The server sends the generated answers to other generative AI models, which then evaluate the answers based on accuracy, consistency, completeness of information, and suitability to the emotional data.

[1358] 6. Counting the evaluation results and selecting the responses:

[1359] The server tally the results and selects the answer with the highest rating.

[1360] 7. Final answer output:

[1361] The server sends the selected best answer to the user's terminal, which displays the final answer to the user.

[1362] Adding specific examples

[1363] As a concrete example, consider the case where a user inputs the question "My security camera isn't working." In this case, the emotion engine detects anxiety from the user's text data and sends the emotion data and question to the server. The server then delivers the question and emotion data to the generative AI model, which may generate the following answers:

[1364] Model 1: "Check the security camera cable. It may have a loose connection."

[1365] Model 2: "Don't be afraid. Try restarting your security cameras."

[1366] Model 3: "If you need your security camera repaired, please contact our support center."

[1367] These answers are then evaluated against each other, and the most appropriate answer, "Don't worry. Try restarting your security camera," is selected and provided to the user.

[1368] Prompt Sentence Examples

[1369] An example prompt sentence would be:

[1370] User Question: "My security camera isn't working. What should I do?"

[1371] Post-sentiment prompt: "In response to the question 'My security camera isn't working. What should I do?', which the user is feeling anxious about, each generative AI model should generate an answer that will ease the user's anxiety."

[1372] This system allows users to receive empathetic and reliable support, making it easier to deal with security-related issues.

[1373] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[1374] Step 1:

[1375] User enters a question:

[1376] The user inputs a question using the terminal. At the time the question is input, the terminal receives the question as text data. Specifically, the user inputs the question using a keyboard or voice input, and the question is saved as text data on the terminal.

[1377] Step 2:

[1378] Emotion Recognition:

[1379] The device sends the question text data entered by the user to the emotion engine. Specifically, it calls the emotion engine API (e.g., Google Cloud NLP API, Microsoft Azure Text Analytics API) to analyze the question text. The input data is the user's question text, and the output data is the emotion data returned by the emotion engine. The emotion engine performs text analysis and extracts emotions such as anxiety and impatience.

[1380] Step 3:

[1381] Sending questions with sentiment data:

[1382] The device adds the emotion data extracted by the emotion engine to the question text and sends it to the server. The input data is the emotion data and the question text, and the output data is a data packet that integrates them. This data packet is sent to the server and received by the server.

[1383] Step 4:

[1384] Generate an answer:

[1385] The server sends the received question and emotion data to multiple generative AI models in parallel. The server calls the API of the generative AI model (e.g., OpenAI GPT-3, Hugging Face Transformers), and each model generates an answer using its own algorithm. The input data is the question with the emotion data added, and the output data is the answer returned by each generative AI model.

[1386] Step 5:

[1387] Peer assessment of answers:

[1388] The server sends the answers received from each generative AI model back to other generative AI models to evaluate the answers. Specifically, the server requests each answer from other AI models for evaluation, and each model returns an evaluation score. The input data is the answer generated by each generative AI model, and the output data is the evaluation score.

[1389] Step 6:

[1390] Counting the assessment results and sorting the responses:

[1391] The server aggregates the evaluation results returned by each generative AI model and selects the answer with the highest evaluation. The input data is the evaluation score, and the output data is the optimal answer. The evaluation scores are aggregated and analyzed within the server.

[1392] Step 7:

[1393] Final answer output:

[1394] The server sends the selected optimal answer to the user's terminal. The terminal receives this final answer and displays it to the user. Specifically, the server sends the optimal answer in packet form to the terminal, and the terminal's user interface displays it. The input data is the optimal answer, and the output data is the answer displayed to the user.

[1395] These processing steps allow users to receive emotionally sensitive and reliable answers.

[1396] The specific processing unit 290 transmits the result of the specific processing to the headset type terminal 314. In the headset type terminal 314, the control unit 46A causes the speaker 240 and the display 343 to output the result of the specific processing. The microphone 238 acquires audio indicating a user input regarding the result of the specific processing. The control unit 46A transmits audio data indicating the user input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the audio data.

[1397] The data generation model 58 is a so-called generative AI (Artificial Intelligence). An example of the data generation model 58 is ChatGPT (Internet Search<URL: https: / / openai.com / blog / chatgpt> ), Gemini (Internet search <url: https: gemini.google.com ?hl="ja">) and other generation AIs. The data generation model 58 is obtained by performing deep learning on a neural network. A prompt including an instruction is input to the data generation model 58, and inference data such as voice data indicating voice, text data indicating text, and image data indicating an image is also input. The data generation model 58 performs inference on the input inference data in accordance with the instruction indicated by the prompt, and outputs the inference result in a data format such as voice data and text data. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization.

[1398] In the above embodiment, an example was given in which the specific processing is performed by the data processing device 12, but the technology of the present disclosure is not limited to this, and the specific processing may be performed by the headset type terminal 314.

[1399] [Fourth embodiment]

[1400] FIG. 7 shows an example of the configuration of a data processing system 410 according to the fourth embodiment.

[1401] 7, a data processing system 410 includes a data processing device 12 and a robot 414. An example of the data processing device 12 is a server.

[1402] The data processing device 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" according to the technology of the present disclosure. The computer 22 includes a processor 28, a RAM 30, and a storage 32. The processor 28, the RAM 30, and the storage 32 are connected to a bus 34. The database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. Examples of the network 54 include a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[1403] The robot 414 includes a computer 36, a microphone 238, a speaker 240, a camera 42, a communication I / F 44, and a control target 443. The computer 36 includes a processor 46, a RAM 48, and a storage 50. The processor 46, the RAM 48, and the storage 50 are connected to a bus 52. The microphone 238, the speaker 240, the camera 42, and the control target 443 are also connected to the bus 52.

[1404] The microphone 238 receives instructions and the like from the user 20 by receiving voice uttered by the user 20. The microphone 238 captures the voice uttered by the user 20, converts the captured voice into audio data, and outputs it to the processor 46. The speaker 240 outputs audio in accordance with instructions from the processor 46.

[1405] Camera 42 is a small digital camera equipped with an optical system including a lens, aperture, and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, and captures images of the surroundings of user 20 (for example, an imaging range defined by an angle of view equivalent to the field of vision of a typical healthy person).

[1406] The communication I / F 44 is connected to a network 54. The communication I / Fs 44 and 26 are responsible for the exchange of various information between the processor 46 and the processor 28 via the network 54. The exchange of various information between the processor 46 and the processor 28 using the communication I / Fs 44 and 26 is carried out in a secure state.

[1407] The control object 443 includes a display device, LEDs in the eyes, and motors for driving the arms, hands, and feet. The posture and gestures of the robot 414 are controlled by controlling the motors of the arms, hands, and feet. Some of the emotions of the robot 414 can be expressed by controlling these motors. In addition, the facial expressions of the robot 414 can also be expressed by controlling the light emission state of the LEDs in the eyes of the robot 414.

[1408] Fig. 8 shows an example of the main functions of the data processing device 12 and the robot 414. As shown in Fig. 8, in the data processing device 12, a specific process is performed by the processor 28. A specific process program 56 is stored in the storage 32.

[1409] The specific processing program 56 is an example of a "program" according to the technology of the present disclosure. The processor 28 reads the specific processing program 56 from the storage 32 and executes the read specific processing program 56 on the RAM 30. The specific processing is realized by the processor 28 operating as a specific processing unit 290 in accordance with the specific processing program 56 executed on the RAM 30.

[1410] The storage 32 stores a data generation model 58 and an emotion identification model 59. The data generation model 58 and the emotion identification model 59 are used by the identification processing unit 290.

[1411] In the robot 414, the processor 46 performs the reception output process. A reception output program 60 is stored in the storage 50. The processor 46 reads the reception output program 60 from the storage 50 and executes the read reception output program 60 on the RAM 48. The reception output process is realized by the processor 46 operating as the control unit 46A in accordance with the reception output program 60 executed on the RAM 48.

[1412] Next, a description will be given of the specific processing performed by the specific processing unit 290 of the data processing device 12. In the following description, the data processing device 12 will be referred to as a "server" and the robot 414 will be referred to as a "terminal."

[1413] This invention relates to a generative AI platform for improving the accuracy of answers generated by AI. This platform is a system in which multiple generative AI models cooperate to generate the most correct answers through mutual evaluation and discussion. Below, we will explain the program processing of this system in natural language using concrete examples.

[1414] Program processing explanation

[1415] Enter your question

[1416] A user uses an email client to type a question, for example, "What's the weather like in Tokyo tomorrow?"

[1417] Generate answers

[1418] The server receives this question and sends it to several different generative artificial intelligence models (AIM1, AIM2, AIM3, etc.), each of which uses its own algorithm to generate an answer.

[1419] AIM1: "It's sunny."

[1420] AIM2: "Tomorrow will be sunny."

[1421] AIM3: "It's forecast to be sunny in Tokyo."

[1422] Request for evaluation of answer

[1423] The server sends the generated answers to other generative AI models in the Validator Pool, allowing each model to evaluate the answers of the other generative AI models.

[1424] Evaluating answers

[1425] Each generative AI model evaluates the answers generated by other models and assigns them a score based on the accuracy, consistency, and completeness of the answers.

[1426] Aggregation of evaluation results

[1427] The server aggregates the results from each model and sees which answer received the highest rating.

[1428] Notification of evaluation results and request for discussion

[1429] The server notifies each generative AI model of the evaluation results and requests them to correct or supplement their answers as necessary.

[1430] Editing and discussing answers

[1431] Each model modifies its answer based on the discussion and adds any necessary corrections. For example, AIM1 can update its answer as follows:

[1432] AIM1: "Tomorrow will be sunny in Tokyo, with temperatures around 25 degrees."

[1433] AIM2: "Tomorrow in Tokyo will be sunny with a maximum temperature of 25 degrees."

[1434] AIM3: "It'll be sunny and 25 degrees."

[1435] Final evaluation and majority vote

[1436] The server then re-evaluates the updated answer against the models in the Validator Pool. Each model selects the most appropriate answer and a majority vote is held.

[1437] Selection of the final answer

[1438] Based on the evaluation results, the server selects the final answer. The final answer obtained through this process is "Tomorrow in Tokyo, it will be sunny and the temperature will be around 25 degrees."

[1439] Submit your answer

[1440] The server sends the final selected answer to the terminal.

[1441] Show Answers

[1442] The terminal displays the final answer to the user.

[1443] This system allows multiple generative AI models to cooperate and generate more accurate and reliable answers through mutual evaluation and discussion. As can be seen from this explanation with concrete examples, the present invention provides an effective means for suppressing hallucination and improving the accuracy of information.

[1444] The processing flow will be explained below.

[1445] Step 1:

[1446] The user inputs a question, for example, "What's the weather like in Tokyo tomorrow?" into the terminal.

[1447] Step 2:

[1448] The terminal transmits the entered question to the server.

[1449] Step 3:

[1450] The server receives questions and distributes them to multiple generative artificial intelligence models (AIM1, AIM2, AIM3).

[1451] Step 4:

[1452] Each generative AI model generates an answer to a question sent by the server using its own algorithm. For example,

[1453] AIM1: "It's sunny."

[1454] AIM2: "Tomorrow will be sunny."

[1455] AIM3: "It's forecast to be sunny in Tokyo."

[1456] Step 5:

[1457] The server sends the generated answer to each of the other generative AI models in the Validator Pool for evaluation.

[1458] Step 6:

[1459] Each generative AI model evaluates the answers it receives from other models and assigns a score based on the accuracy, consistency, and completeness of the answers.

[1460] Step 7:

[1461] The server aggregates the evaluation results from each model and analyzes which answer received the highest evaluation.

[1462] Step 8:

[1463] The server notifies each generative AI model of the evaluation results and, if necessary, requests them to correct or supplement their answers.

[1464] Step 9:

[1465] Each generative AI model modifies the answer based on the evaluation results and adds correction information. For example,

[1466] AIM1: "Tomorrow will be sunny in Tokyo, with temperatures around 25 degrees."

[1467] AIM2: "Tomorrow in Tokyo will be sunny with a maximum temperature of 25 degrees."

[1468] AIM3: "It'll be sunny and the temperature will be 25 degrees."

[1469] Step 10:

[1470] The server then sends the updated answers to the Validator Pool again, asking each model to re-evaluate.

[1471] Step 11:

[1472] Each generative AI model then evaluates the answers again, selects the most appropriate answer, and assigns it a score.

[1473] Step 12:

[1474] The server tally the results of the reevaluation and selects the answer with the highest score. For example, "Tomorrow in Tokyo, it will be sunny and the temperature will be around 25 degrees."

[1475] Step 13:

[1476] The server sends the final selected answer to the terminal.

[1477] Step 14:

[1478] The terminal displays the final answer to the user.

[1479] Example 1

[1480] Next, a description will be given of Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the robot 414 will be referred to as a "terminal."

[1481] In existing generative AI models, relying on a single model's answer can reduce the accuracy and reliability of the answer. Furthermore, when the answers generated by each model differ, it is difficult to determine which answer is the most accurate. There is a need to solve this problem and provide more accurate and reliable answers.

[1482] The specific processing by the specific processing unit 290 of the data processing device 12 in the first embodiment is realized by the following means.

[1483] In this invention, the server includes means for inputting a question, means for multiple generative AI models to each generate an answer to the question, means for other generative AI models to evaluate the answers generated by each generative AI model, means for selecting the most appropriate answer based on the evaluation results, means for correcting or supplementing the selected answer, means for re-evaluating the corrected answer, means for performing the evaluation process in stages and selecting a final answer based on the result of majority voting, and means for transmitting and displaying the selected final answer to a user terminal. This makes it possible to provide users with more accurate and reliable final answers through mutual evaluation and discussion among multiple generative AI models.

[1484] The "means for inputting a question" provides an interface for a user to input a question to the system.

[1485] A "generative artificial intelligence model" refers to an algorithm or machine learning model that generates answers in natural language to input questions.

[1486] The "means for generating an answer" is what enables the generative artificial intelligence model to generate an answer to an input question.

[1487] "Means for evaluating answers" refers to a function in which answers generated by a generative AI model are evaluated by other generative AI models and scored for accuracy and consistency.

[1488] The "means for selecting the most appropriate answer" includes a function for selecting the answer that has received the highest evaluation based on the evaluation results of each generative artificial intelligence model.

[1489] "Means for correcting or supplementing answers" refers to the functionality for adding necessary corrections or additional information to selected answers.

[1490] "Means for re-evaluation" refers to a function that allows each generative artificial intelligence model to re-evaluate an answer that has been corrected or supplemented.

[1491] "Means for carrying out the evaluation process in stages" includes a function for dividing the evaluation into multiple stages and carrying out a final evaluation based on the evaluation results at each stage.

[1492] "Means for selecting a final answer based on the results of a majority vote" refers to a function for selecting the answer that receives the most support as the final answer based on the evaluation results of each generative AI model.

[1493] The "means for transmitting and displaying the final answer" provides a function for transmitting the selected final answer to the user terminal and displaying it.

[1494] This invention relates to a generative AI platform for improving the accuracy of answers generated by AI. This platform is a system in which multiple generative artificial intelligence models (generative AI models) cooperate to generate the most correct answers through mutual evaluation and discussion.

[1495] Hardware and software used

[1496] The hardware required to implement this system includes a high-performance server and user terminals (e.g., PCs and smartphones). The server has high computing power and storage capacity, and a cloud platform for running the generative AI models can also be used. The software includes each generative AI model, an email client, a web form, etc.

[1497] Data processing and calculation

[1498] When a user enters a question using an email client or web form, the question is stored in a database and sent to a server, which parses the question and converts it into a prompt that the generative AI model can understand.

[1499] For example, in response to the question "What's the weather like in Tokyo tomorrow?", the server generates a prompt sentence, "Please tell me the weather forecast for Tokyo tomorrow," and sends this to multiple generative AI models (AIM1, AIM2, AIM3, etc.).

[1500] Each generative AI model receives a prompt and uses its own algorithms and datasets to generate an answer.

[1501] AIM1: "It's sunny."

[1502] AIM2: "Tomorrow will be sunny."

[1503] AIM3: "It's forecast to be sunny in Tokyo."

[1504] The generated answers are sent by the server to other generative AI models in the Validator Pool, where each model evaluates the answers of the other models based on the accuracy, consistency, and completeness of the answers.

[1505] The evaluation results are tallied, and the server checks which answer received the highest rating. For example, if AIM1 received the highest rating, that score is recorded. The evaluation results are notified to each generative AI model, which corrects or supplements the answer as necessary. For example, if AIM1 is evaluated as having insufficient temperature information, additional information is provided.

[1506] Each generative AI model modifies its answer based on the discussion and adds any necessary corrections. Answers may be updated as follows:

[1507] AIM1: "Tomorrow will be sunny in Tokyo, with temperatures around 25 degrees."

[1508] AIM2: "Tomorrow in Tokyo will be sunny with a maximum temperature of 25 degrees."

[1509] AIM3: "It'll be sunny and 25 degrees."

[1510] The server then evaluates the updated answers again using the models in the Validator Pool, and each generative AI model selects the most appropriate answer through a majority vote. Based on the evaluation results, the server selects the final answer: "Tomorrow in Tokyo will be sunny, with a temperature of around 25 degrees."

[1511] Examples of specific examples and prompts

[1512] As a specific example, the following shows how a generative AI model generates and evaluates answers to the input, "What's the weather like in Tokyo tomorrow?"

[1513] User input: "What's the weather like in Tokyo tomorrow?"

[1514] Server prompt: "What's the weather forecast for Tokyo tomorrow?"

[1515] Example answer from a generative AI model:

[1516] AIM1: "It's sunny."

[1517] AIM2: "Tomorrow will be sunny."

[1518] AIM3: "It's forecast to be sunny in Tokyo."

[1519] Example of an answer after evaluation and revision:

[1520] AIM1: "Tomorrow will be sunny in Tokyo, with temperatures around 25 degrees."

[1521] AIM2: "Tomorrow in Tokyo will be sunny with a maximum temperature of 25 degrees."

[1522] AIM3: "It'll be sunny and 25 degrees."

[1523] Ultimately, the answer selected by the server based on the majority vote was: "Tomorrow in Tokyo, it will be sunny, and the temperature will be around 25 degrees."

[1524] Through the collaboration and mutual evaluation of multiple generative AI models, this system enables the provision of more accurate and reliable information to users.

[1525] The flow of the identification process in the first embodiment will be described with reference to FIG.

[1526] Step 1:

[1527] A user enters a question using an email client or web form, for example, "What's the weather like in Tokyo tomorrow?" This input is sent to the server.

[1528] Input: User question: "What's the weather like in Tokyo tomorrow?"

[1529] Output: Query data sent to the server

[1530] Step 2:

[1531] The server analyzes the received question and converts it into a prompt that the generative AI model can understand. For example, it generates a prompt such as, "Please tell me the weather forecast for Tokyo tomorrow."

[1532] Input: Question data "What's the weather like in Tokyo tomorrow?"

[1533] Output: Prompt "What is the weather forecast for Tokyo tomorrow?"

[1534] Step 3:

[1535] The server sends the generated prompt to multiple generative AI models (AIM1, AIM2, AIM3, etc.), each of which receives the prompt and generates an answer using its own algorithm and dataset.

[1536] Input: Prompt "What is the weather forecast for Tokyo tomorrow?"

[1537] Output: Answer of each generative AI model (e.g., AIM1: "It's sunny," AIM2: "It will be sunny tomorrow," AIM3: "It is forecast to be sunny in Tokyo.")

[1538] Step 4:

[1539] The server sends the generated answers to other generative AI models in the Validator Pool, asking each model to evaluate the answers of the other models.

[1540] Input: The generative AI model's answer

[1541] Output: Evaluation request data

[1542] Step 5:

[1543] Generative AI models evaluate the answers generated by other generative AI models. For example, AIM2 looks at the answers of AIM1 and AIM3 and assigns a score to each. The evaluation criteria are the accuracy, consistency, and completeness of the information.

[1544] Input: Answers from other generative AI models

[1545] Output: Evaluation score (e.g., AIM2 scoring)

[1546] Step 6:

[1547] The server aggregates the results from each model and checks which answer received the highest score. For example, if AIM1 received the highest score, that score is recorded.

[1548] Input: Rating score

[1549] Output: The answer from the model with the highest score

[1550] Step 7:

[1551] The server notifies each generative AI model of the evaluation results and asks them to correct or supplement their answers as necessary. For example, if AIM1 receives an evaluation that "temperature information is insufficient," it will ask them to provide additional information.

[1552] Input: Evaluation result

[1553] Output: Notification of correction request

[1554] Step 8:

[1555] The generative AI model then modifies the answer based on the discussion and adds any necessary corrections. For example, AIM1 updates the answer to "Tomorrow will be sunny in Tokyo, with the temperature around 25 degrees."

[1556] Input: Notification of correction request

[1557] Output: Updated answer (e.g. AIM1: "Tomorrow will be sunny in Tokyo, with the temperature around 25 degrees.")

[1558] Step 9:

[1559] The server then re-evaluates the updated answers with the models in the Validator Pool. Each generative AI model selects the most appropriate answer and a majority vote is held.

[1560] Input: Updated answer

[1561] Output: Reassessment score and majority vote result

[1562] Step 10:

[1563] The server selects the final answer based on the evaluation results. For example, it selects "Tomorrow in Tokyo will be sunny, with the temperature around 25 degrees" as the most popular answer.

[1564] Input: Reassessment score and majority vote result

[1565] Output: Final answer "Tomorrow in Tokyo it will be sunny, with the temperature around 25 degrees."

[1566] Step 11:

[1567] The server sends the final selected answer to the terminal.

[1568] Input: Final answer

[1569] Output: Data sent to the terminal

[1570] Step 12:

[1571] The terminal displays the final answer to the user, who can view it through their email client or web interface.

[1572] Input: Final response data

[1573] Output: What is displayed to the user

[1574] This series of processes enables the provision of more accurate and reliable information to users through cooperation and mutual evaluation of multiple generative AI models.

[1575] (Application example 1)

[1576] Next, a description will be given of Application Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the robot 414 will be referred to as a "terminal."

[1577] Conventional answer generation systems using generative AI models often lack consistency and reliability, with the accuracy of each model varying. Furthermore, especially in the security field, where incorrect answers can have significant impacts, higher accuracy and reliability are required. For this reason, there was a need to develop a system in which multiple AI models cooperate and generate the most accurate and reliable answers through mutual evaluation and discussion.

[1578] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 1 is realized by the following means.

[1579] In this invention, the server includes means for multiple generative AI models to generate answers to questions, means for other generative AI models to evaluate the answers generated by each generative AI model, and means for selecting the most appropriate answer based on the evaluation results. This improves the accuracy and reliability of the generated answers, enabling reliable notifications to users, particularly when used as security alerts.

[1580] "Multiple generative artificial intelligence models" refers to multiple AI models that use different algorithms and datasets to generate answers in their own unique way.

[1581] "Means for generating answers to each question" refers to the process or function by which each AI model generates an individual answer to a question input by a user.

[1582] "Means for evaluation by other generative artificial intelligence models" refers to the process or function by which answers generated by one AI model are evaluated by other AI models to determine their accuracy and consistency.

[1583] "Means for selecting the most appropriate answer" refers to the process or function of selecting the answer that received the highest rating based on the results of multiple evaluations.

[1584] "Means for outputting the selected answer as the final answer" refers to the process or function of providing the selected most appropriate answer to the user as the final answer.

[1585] "Means for holding discussions, adding corrections, and updating answers" refers to the process or function of discussing the answers generated by each AI model and making necessary corrections to generate more accurate answers.

[1586] "A method of conducting a step-by-step evaluation process and selecting a final answer based on the results of a majority vote" refers to a process of going through a series of evaluation and selection steps in a step-by-step manner, ultimately selecting the answer supported by the majority of evaluators.

[1587] A "security alert" refers to a warning message that notifies the user of an abnormal security situation or abnormality.

[1588] "Means of notifying the user" refers to the process or function of sending the final selected answer or alert to the user's device and notifying the user by visual, auditory, or other means.

[1589] This invention relates to a system in which multiple generative artificial intelligence models cooperate to generate the most correct answer through mutual evaluation and discussion. Specifically, this invention describes an application for providing accurate alerts to users in security services.

[1590] Hardware and software used

[1591] Hardware:

[1592] 1. Server: Runs multiple AI models and manages the evaluation process.

[1593] 2. Security cameras: collect data and provide it for analysis.

[1594] 3. Smartphone: The device where the user enters the question and receives the final answer.

[1595] software:

[1596] 1. Python: A system-wide programming language.

[1597] 2. Requests library: Manages API communication.

[1598] 3. AI Models: Multiple generative AI models that use various algorithms to generate solutions to problems.

[1599] System processing flow

[1600] 1. User enters question:

[1601] Users use a smartphone app to enter specific questions, such as "Can you check if there's an intruder in my office?"

[1602] 2. The server receives the query:

[1603] The server receives this question and sends it to multiple generative AI models, each of which uses its own algorithm to generate an answer.

[1604] 3. The AI ​​model generates the answer:

[1605] Each generative AI model generates an answer to the question, for example:

[1606] Model 1: "Motion detected on camera, but it doesn't seem to be an intruder."

[1607] Model 2: "It looks like the door is open, but we need to check it out immediately."

[1608] Model 3: "Motion detected, but no further information available."

[1609] 4. Other models rate your answers:

[1610] The server sends these answers to other generative AI models, which then evaluate the answers of the other models based on the accuracy and consistency of the answers.

[1611] 5. Compilation of evaluation results and selection of best answers:

[1612] Based on the evaluation results of each AI model, the answer with the highest score will be selected. For example, the evaluation will be as follows:

[1613] "Motion detected on camera, but it doesn't appear to be an intruder" received the highest rating.

[1614] 6. Generate final answer and notify user:

[1615] The server then sends the final selected answer to the user's device and notifies them, allowing the user to receive accurate and reliable security alerts.

[1616] Prompt Sentence Examples

[1617] User input:

[1618] "Check if there's an intruder in the office."

[1619] This invention enables multiple generative AI models to cooperate to generate the most appropriate response and notify the user of it as a security alert, allowing the user to take appropriate action based on highly reliable information.

[1620] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[1621] Step 1:

[1622] A user uses a smartphone to input a question, for example, "Please check if there is an intruder in my office." This input is sent to the server through the user interface.

[1623] Step 2:

[1624] The server receives a user's question and sends it to multiple generative AI models. Each model receives the question and generates an answer using its own algorithm. A typical answer might be, "Motion was detected on the camera, but it doesn't appear to be an intruder."

[1625] Step 3:

[1626] The server sends the answers generated by each generative AI model to other generative AI models for evaluation. At this time, the input provided to each model is a list of generated answers, and each model evaluates this list.

[1627] Step 4:

[1628] Each generative AI model evaluates the answers generated by other models and assigns a score based on their accuracy, consistency, and completeness of information. The evaluation results are sent to the server in the form of a score. For example, model A may assign 90 points to the answer of model B and 80 points to the answer of model C.

[1629] Step 5:

[1630] The server then selects the answer with the highest score based on the aggregated evaluation results. For example, if "Motion was detected on the camera, but it doesn't appear to be an intruder" receives the highest score, it will be selected.

[1631] Step 6:

[1632] The server then sends the selected answer to the user's smartphone. This notification includes the most highly rated answer and a brief explanation of it. The user can then check the answer on their device.

[1633] Step 7:

[1634] If necessary, the generated answers may be further discussed and corrections added. Each generative AI model will be discussed again and generate a new answer with the necessary corrections. This process of evaluation and correction is repeated until the revised best answer is selected.

[1635] The above is a specific flow of processing by the program that realizes this invention, which enables users to receive accurate and reliable security alerts.

[1636] Furthermore, an emotion engine that estimates the user's emotion may be further combined. That is, the identification processing unit 290 may estimate the user's emotion using the emotion identification model 59, and perform identification processing using the user's emotion.

[1637] This invention relates to a generative AI platform that uses multiple generative AI models and an emotion engine to generate accurate and appropriate answers to user questions. Each generative AI model generates and modifies answers through mutual evaluation and discussion, and by combining it with an emotion engine that recognizes the user's emotions, it provides answers that are adapted to the user's emotional state. Below, we will explain the program processing of this system in natural language using concrete examples.

[1638] Program processing explanation

[1639] Enter your question

[1640] The user uses the terminal to input a question, for example, "Tell me what my friends think."

[1641] Emotion recognition

[1642] The device sends the user's input, voice, facial expression, and other data to the emotion engine, which analyzes this data and recognizes the user's emotions. For example, it can detect "anxiety."

[1643] Submit a Question

[1644] The device sends the question with the recognized emotion data to the server.

[1645] Generate answers

[1646] The server distributes the emotional data attached to the question to multiple generative AI models (AIM1, AIM2, AIM3). Each model takes the emotional data into account and generates an answer using its own algorithm. For example,

[1647] AIM1: "Your friends care about you."

[1648] AIM2: "Your friends are worried about you."

[1649] AIM3: "Your friends trust you."

[1650] Request for evaluation of answer

[1651] The server sends the generated answers to other generative AI models, allowing each model to evaluate the answers of the other generative AI models.

[1652] Evaluating answers

[1653] The generative AI model evaluates the answers generated by other models and assigns them a score based on the answer's accuracy, consistency, completeness of information, and fit to the perceived emotion.

[1654] Aggregation of evaluation results

[1655] The server aggregates the evaluation results from each model and analyzes which answer received the highest evaluation.

[1656] Notification of evaluation results and request for discussion

[1657] The server notifies each generative AI model of the evaluation results and requests them to revise or supplement their answers if necessary. Emotional data is also reflected.

[1658] Editing and discussing answers

[1659] Each generative AI model modifies its answers based on the evaluation results and emotional data, adding correction information. For example,

[1660] AIM1: "Friends care about you and always want to support you."

[1661] AIM2: "Your friend is very concerned about you and wants to talk more."

[1662] AIM3: "Your friends trust you and want to talk to you if they have a problem."

[1663] Final evaluation and majority vote

[1664] The server re-evaluates the updated answer and asks each model to re-evaluate.

[1665] Selection of the final answer

[1666] Based on the reevaluation results, the server selects a final answer, such as "Your friends care about you and always want to support you."

[1667] Submit your answer

[1668] The server sends the final selected answer to the terminal.

[1669] Show Answers

[1670] The terminal displays the final answer to the user.

[1671] This system enables multiple generative AI models to generate more accurate and reliable answers that adapt to the user's emotional state through mutual evaluation and discussion. As can be seen from the explanation using concrete examples, the present invention provides an innovative means for suppressing hallucination and providing information that takes into consideration the user's emotions.

[1672] The processing flow will be explained below.

[1673] Step 1:

[1674] The user inputs a question, for example, "What's the weather like in Tokyo tomorrow?" into the terminal.

[1675] Step 2:

[1676] The device sends the input question as data to the emotion engine, which analyzes the user's input, voice, facial expressions, and other data to recognize the user's emotions. For example, it can detect "anxiety."

[1677] Step 3:

[1678] The emotion engine returns the recognized emotion data to the terminal.

[1679] Step 4:

[1680] The device sends the question with the emotional data attached to it to the server.

[1681] Step 5:

[1682] The server receives questions and emotion data and distributes them to multiple generative AI models (AIM1, AIM2, AIM3).

[1683] Step 6:

[1684] Each generative AI model uses its own algorithm to generate answers to questions while taking into account emotional data. For example,

[1685] AIM1: "It's sunny."

[1686] AIM2: "Tomorrow will be sunny."

[1687] AIM3: "It's forecast to be sunny in Tokyo."

[1688] Step 7:

[1689] The server compiles the generated answers and sends them to other generative AI models, allowing each model to evaluate the answers of the other generative AI models.

[1690] Step 8:

[1691] Each generative AI model evaluates the answers sent by other generative AI models and assigns a score based on the answer's accuracy, consistency, completeness of information, and fit to the perceived emotion.

[1692] Step 9:

[1693] The server aggregates the evaluation results from each model and analyzes which answer received the highest evaluation.

[1694] Step 10:

[1695] The server notifies each generative AI model of the evaluation results and requests them to revise or supplement their answers as necessary. This emotional data is also reflected in the re-evaluation.

[1696] Step 11:

[1697] Each generative AI model modifies its answers based on the evaluation results and emotional data, adding correction information. For example,

[1698] AIM1: "Tomorrow will be sunny in Tokyo. The temperature will be around 25 degrees."

[1699] AIM2: "Tomorrow in Tokyo will be sunny with a maximum temperature of 25 degrees."

[1700] AIM3: "It'll be sunny and the temperature will be 25 degrees."

[1701] Step 12:

[1702] The server then distributes the updated answers to each generative AI model again and requests them to evaluate again.

[1703] Step 13:

[1704] Each generative AI model then evaluates the answers again, selects the most appropriate answer, and assigns it a score.

[1705] Step 14:

[1706] The server tally the results of the reevaluation and selects the answer with the highest score. For example, "Tomorrow in Tokyo, it will be sunny and the temperature will be around 25 degrees."

[1707] Step 15:

[1708] The server sends the final selected answer to the terminal.

[1709] Step 16:

[1710] The terminal displays the final answer to the user.

[1711] Example 2

[1712] Next, a description will be given of Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the robot 414 will be referred to as a "terminal."

[1713] In conventional generative AI systems, the answers generated are not always adapted to the user's emotional state, which reduces user satisfaction. Furthermore, the evaluation process for multiple generative AI models is insufficient, resulting in answers that lack consistency and accuracy. Furthermore, the generated answers do not reflect correction information or discussions, which can result in the delivery of less than optimal answers.

[1714] The specific processing by the specific processing unit 290 of the data processing device 12 in the second embodiment is realized by the following means.

[1715] In this invention, the server includes means for multiple generative AI models to generate answers to questions, means for other generative AI models to evaluate the answers generated by each generative AI model, means for selecting the most appropriate answer based on the evaluation results, an emotion engine that recognizes the user's emotion using data input by the user, and means for attaching the emotion data recognized by the emotion engine to the question and transmitting it, thereby making it possible to generate and provide accurate and appropriate answers that are adapted to the user's emotional state.

[1716] "Multiple generative AI models" are multiple AI systems with different algorithms and settings, each capable of generating answers independently.

[1717] A "question" is any information or question entered by a user that requires an answer from a generative AI model.

[1718] The "means for generating an answer" refers to the process or algorithm that the generative artificial intelligence model uses to create an appropriate answer to a user's question.

[1719] "Means for evaluation by other generative artificial intelligence models" refers to the process by which each generative artificial intelligence model evaluates answers generated by other models and assigns them a score or ranking.

[1720] The "means of selecting the most appropriate answer based on the evaluation results" is a system that aggregates the evaluation results obtained from multiple generative AI models and selects the answer that receives the highest evaluation.

[1721] An "emotion engine" is a system that analyzes the user's input data, voice, facial expressions, etc. to recognize the user's emotional state.

[1722] "Emotion data" is data showing the results of analysis by the user's emotion engine, and represents an emotional state such as anxiety or relief.

[1723] The "means of transmission" is the process or method of transmitting data from the device to the server, often using a secure communications protocol.

[1724] "Correction information" is information that is added to the initially generated answer, and is used to improve the accuracy and appropriateness of the answer.

[1725] "Means for discussion" refers to a process in which multiple generative AI models discuss the generated answers and make optimal improvements or corrections.

[1726] "Majority vote" is a method of statistically aggregating the results of multiple evaluations and scoring and selecting the answer that receives the most support.

[1727] The system starts when a user uses a device to input a question. For example, "Tell me what my friend thinks." The user can use a keyboard or touch screen to input a question. The system may also collect voice and facial expression data, which are sent to the emotion engine.

[1728] The emotion engine analyzes the collected data and recognizes the user's emotions. Voice recognition technology and facial expression analysis algorithms are used to recognize emotions, such as "anxiety." The emotion data recognized in this way is attached to the user's question and sent from the device to the server. A secure communication protocol (e.g., HTTPS) is used to protect the data.

[1729] The server distributes the received question and emotion data to multiple generative AI models (e.g., AIM1, AIM2, AIM3). Each model generates an answer using its own natural language processing algorithm. Examples of generated answers include:

[1730] AIM1: "Your friends care about you."

[1731] AIM2: "Your friends are worried about you."

[1732] AIM3: "Your friends trust you."

[1733] The server then sends the answers generated by each generative AI model to other generative AI models, who evaluate each other's answers based on the accuracy, consistency, completeness of the information, and suitability to the emotional data. For example, AIM1 might give 70 points to AIM2's answer and 80 points to AIM3's answer.

[1734] The server aggregates these evaluation results and selects the answer with the highest rating. After selection, the server notifies each generative AI model of the evaluation results and requests them to revise or supplement the answer as necessary. Emotional data is also reflected.

[1735] Each generative AI model modifies its answers based on the evaluation results and emotional data, adding correction information. For example, it may be modified as follows:

[1736] AIM1: "Friends care about you and always want to support you."

[1737] AIM2: "Your friend is very concerned about you and wants to talk more."

[1738] AIM3: "Your friends trust you and want to talk to you if they have a problem."

[1739] Finally, the server re-evaluates the updated answers and selects the answer with the most votes. The final answer is sent to the device and displayed to the user, providing a reliable answer to the question.

[1740] For example, consider the following prompt:

[1741] "Please tell me what my friend thinks. I'm feeling anxious right now."

[1742] This system enables multiple generative AI models to provide more accurate and reliable answers that adapt to the user's emotional state through mutual evaluation and discussion, thereby improving user satisfaction and the quality of information provided.

[1743] The flow of the identification process in the second embodiment will be described with reference to FIG.

[1744] Step 1: The user enters a question using a terminal.

[1745] Input: Text data entered by the user using a keyboard or touchscreen.

[1746] Output: Question text data imported to the terminal.

[1747] Specific operation: The user inputs a question such as "Please tell me what my friends think." The device collects the text data of the question.

[1748] Step 2: The device sends the user's input data to the emotion engine and recognizes the user's emotion.

[1749] Input: Text data of the user's question acquired by the device, as well as voice and facial expression data.

[1750] Output: Emotion data parsed by the emotion engine.

[1751] Specific operation: The device sends the user's text data, voice, and facial expressions to the emotion engine, which analyzes them and recognizes emotions such as "anxiety."

[1752] Step 3: The device sends the question accompanied by the recognized emotion data to the server.

[1753] Input: User question text data and sentiment data.

[1754] Output: Questions and sentiment data sent to the server.

[1755] Specific operation: The device attaches emotional data obtained from the emotion engine to the question and sends it securely to the server using HTTPS.

[1756] Step 4: The server distributes the questions and emotion data to multiple generative AI models.

[1757] Input: Question and emotion data received by the server.

[1758] Output: The questions and sentiment data delivered to each generative AI model.

[1759] Specific operation: The server distributes the received question and emotion data to the generative AI models (e.g., AIM1, AIM2, AIM3), and each model generates an answer.

[1760] Step 5: Each generative AI model generates an answer to the question.

[1761] Input: Questions and sentiment data delivered to each generative AI model.

[1762] Output: The generated answer.

[1763] How it works: The generative AI model uses a proprietary natural language processing algorithm to generate answers. For example, AIM1 generates "Your friends care about you."

[1764] Step 6: The server sends the generated answer to other generative AI models for evaluation.

[1765] Input: The answers generated by each generative AI model.

[1766] Output: Answer evaluation request sent to other generative AI models.

[1767] Specific operation: The server collects the answers generated by each model and sends them to other generative AI models for evaluation.

[1768] Step 7: Each generative AI model evaluates the answers generated by other models and assigns a score.

[1769] Input: Answers received from other generative artificial intelligence models.

[1770] Output: Evaluation score.

[1771] How it works: Each generative AI model evaluates the answers of other models based on specified criteria and assigns a score based on, for example, accuracy or consistency.

[1772] Step 8: The server aggregates the evaluation results of each generative AI model and selects the answer with the highest evaluation.

[1773] Input: Evaluation scores from all generative AI models.

[1774] Output: The answer that received the most votes.

[1775] Specific operation: The server aggregates the evaluation results and selects the answer with the highest score.

[1776] Step 9: The server notifies each generative AI model of the selected answers and requests corrections or supplements if necessary.

[1777] Input: Screened answers and assessment results.

[1778] Output: Request for corrections and clarifications.

[1779] Specific operation: The server sends the selected answers and their evaluation scores to the model, requesting corrections or supplements as necessary.

[1780] Step 10: Each generative AI model modifies the answer and adds correction information.

[1781] Input: Correction request and original answer from the server.

[1782] Output: The revised answer and any corrections added.

[1783] Specific behavior: Each model modifies the answer based on the evaluation results, adding correction information such as, "Your friends care about you and always want to support you."

[1784] Step 11: The server re-evaluates the revised answers and selects the final answer based on the majority vote.

[1785] Input: Revised answer.

[1786] Output: The final selected answer.

[1787] Specific operation: The server has each generative AI model evaluate the revised answer again, and selects the final answer by majority vote based on the evaluation results.

[1788] Step 12: The server sends the final selected answer to the terminal and displays it to the user.

[1789] Input: Final selected answer.

[1790] Output: The final answer displayed on the terminal.

[1791] Specific operation: The server securely transmits the final selected answer to the terminal, which then displays it to the user, allowing the user to obtain a reliable answer.

[1792] (Application example 2)

[1793] Next, a description will be given of Application Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the robot 414 will be referred to as a "terminal."

[1794] Conventional generative AI systems have difficulty generating responses that take user emotions into account, and in situations such as security support, it is necessary to respond appropriately to user anxieties and concerns. Therefore, a system that can recognize user emotions and provide appropriate responses based on those emotions is needed.

[1795] The identification process by the identification processing unit 290 of the data processing device 12 in Application Example 2 is realized by the following means. In this invention, the server includes means for sending a question entered by a user to an emotion engine and recognizing the user's emotion, means for sending the question accompanied by emotion data to the server, and means for the server to cause multiple generation year artificial intelligence models to generate answers taking the emotion data into consideration. This makes it possible to provide appropriate and reliable answers that take the user's emotions into consideration.

[1796] "Multiple Generation Year AI Models" refers to multiple AI systems that operate based on different algorithms and data sets and generate unique answers to questions.

[1797] "Means for generating answers to each question" refers to the function by which each AI model generates its own answer to a question entered by a user.

[1798] "Means for evaluation by other generation year artificial intelligence models" refers to the function by which answers generated by one generation year artificial intelligence model are evaluated by other generation year artificial intelligence models.

[1799] "Means for selecting the most appropriate answer" refers to the function of selecting the most appropriate answer based on the results of evaluations by multiple AI models.

[1800] "Means for outputting the selected answer as the final answer" refers to a function that provides the user with the selected most appropriate answer.

[1801] An "emotion engine" is a system that analyzes data such as user input, voice, and facial expressions to recognize the user's emotions.

[1802] The "means for transmitting a question with emotional data attached to it to a server" refers to a function for adding emotional data recognized by an emotion engine to question data and transmitting it to a server.

[1803] "Means for the server to generate answers taking into account emotional data" refers to the function of the server to have multiple artificial intelligence models generate answers based on emotional data.

[1804] This invention shows how to implement a system that recognizes a user's emotions and provides appropriate answers based on those emotions. This system is composed of multiple AI models, an emotion engine, and a server.

[1805] Program Generation

[1806] The system runs a program that includes the following functions:

[1807] Processing Description

[1808] 1. User enters question:

[1809] A user uses a terminal to input a question. For example, the user inputs the question, "My security camera isn't working. What should I do?"

[1810] 2. Emotion Recognition:

[1811] The device sends the text data entered by the user to the emotion engine, which uses natural language processing (NLP) technology to analyze the user's emotions and extract emotional data such as anxiety or impatience. The emotion engine can use, for example, Google Cloud's NLP API or Microsoft Azure's Text Analytics API.

[1812] 3. Sending questions with emotion data:

[1813] The terminal adds the emotion data extracted by the emotion engine and sends the question to the server.

[1814] 4. Generate answers:

[1815] The server distributes questions that take emotion data into account to multiple generative AI models, each of which uses its own algorithm to generate answers, including OpenAI's GPT-3 and GPT-4 models and the Hugging Face Transformers model.

[1816] 5. Peer assessment of answers:

[1817] The server sends the generated answers to other generative AI models, which then evaluate the answers based on accuracy, consistency, completeness of information, and suitability to the emotional data.

[1818] 6. Counting the evaluation results and selecting the responses:

[1819] The server tally the results and selects the answer with the highest rating.

[1820] 7. Final answer output:

[1821] The server sends the selected best answer to the user's terminal, which displays the final answer to the user.

[1822] Adding specific examples

[1823] As a concrete example, consider the case where a user inputs the question "My security camera isn't working." In this case, the emotion engine detects anxiety from the user's text data and sends the emotion data and question to the server. The server then delivers the question and emotion data to the generative AI model, which may generate the following answers:

[1824] Model 1: "Check the security camera cable. It may have a loose connection."

[1825] Model 2: "Don't be afraid. Try restarting your security cameras."

[1826] Model 3: "If you need your security camera repaired, please contact our support center."

[1827] These answers are then evaluated against each other, and the most appropriate answer, "Don't worry. Try restarting your security camera," is selected and provided to the user.

[1828] Prompt Sentence Examples

[1829] An example prompt sentence would be:

[1830] User Question: "My security camera isn't working. What should I do?"

[1831] Post-sentiment prompt: "In response to the question 'My security camera isn't working. What should I do?', which the user is feeling anxious about, each generative AI model should generate an answer that will ease the user's anxiety."

[1832] This system allows users to receive empathetic and reliable support, making it easier to deal with security-related issues.

[1833] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[1834] Step 1:

[1835] User enters a question:

[1836] The user inputs a question using the terminal. At the time the question is input, the terminal receives the question as text data. Specifically, the user inputs the question using a keyboard or voice input, and the question is saved as text data on the terminal.

[1837] Step 2:

[1838] Emotion Recognition:

[1839] The device sends the question text data entered by the user to the emotion engine. Specifically, it calls the emotion engine API (e.g., Google Cloud NLP API, Microsoft Azure Text Analytics API) to analyze the question text. The input data is the user's question text, and the output data is the emotion data returned by the emotion engine. The emotion engine performs text analysis and extracts emotions such as anxiety and impatience.

[1840] Step 3:

[1841] Sending questions with sentiment data:

[1842] The device adds the emotion data extracted by the emotion engine to the question text and sends it to the server. The input data is the emotion data and the question text, and the output data is a data packet that integrates them. This data packet is sent to the server and received by the server.

[1843] Step 4:

[1844] Generate an answer:

[1845] The server sends the received question and emotion data to multiple generative AI models in parallel. The server calls the API of the generative AI model (e.g., OpenAI GPT-3, Hugging Face Transformers), and each model generates an answer using its own algorithm. The input data is the question with the emotion data added, and the output data is the answer returned by each generative AI model.

[1846] Step 5:

[1847] Peer assessment of answers:

[1848] The server sends the answers received from each generative AI model back to other generative AI models to evaluate the answers. Specifically, the server requests each answer from other AI models for evaluation, and each model returns an evaluation score. The input data is the answer generated by each generative AI model, and the output data is the evaluation score.

[1849] Step 6:

[1850] Counting the assessment results and sorting the responses:

[1851] The server aggregates the evaluation results returned by each generative AI model and selects the answer with the highest evaluation. The input data is the evaluation score, and the output data is the optimal answer. The evaluation scores are aggregated and analyzed within the server.

[1852] Step 7:

[1853] Final answer output:

[1854] The server sends the selected optimal answer to the user's terminal. The terminal receives this final answer and displays it to the user. Specifically, the server sends the optimal answer in packet form to the terminal, and the terminal's user interface displays it. The input data is the optimal answer, and the output data is the answer displayed to the user.

[1855] These processing steps allow users to receive emotionally sensitive and reliable answers.

[1856] The specific processing unit 290 transmits the result of the specific processing to the robot 414. In the robot 414, the control unit 46A causes the speaker 240 and the control target 443 to output the result of the specific processing. The microphone 238 acquires voice indicating a user input regarding the result of the specific processing. The control unit 46A transmits voice data indicating the user input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the voice data.

[1857] The data generation model 58 is a so-called generative AI (Artificial Intelligence). An example of the data generation model 58 is ChatGPT (Internet Search<URL: https: / / openai.com / blog / chatgpt> ), Gemini (Internet search <url: https: gemini.google.com ?hl="ja">) and other generation AIs. The data generation model 58 is obtained by performing deep learning on a neural network. A prompt including an instruction is input to the data generation model 58, and inference data such as voice data indicating voice, text data indicating text, and image data indicating an image is also input. The data generation model 58 performs inference on the input inference data in accordance with the instruction indicated by the prompt, and outputs the inference result in a data format such as voice data and text data. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization.

[1858] In the above embodiment, an example was given in which the specific processing is performed by the data processing device 12, but the technology of the present disclosure is not limited to this, and the specific processing may be performed by the robot 414.

[1859] The emotion identification model 59 as an emotion engine may determine the user's emotion according to a specific mapping. Specifically, the emotion identification model 59 may determine the user's emotion according to an emotion map (see FIG. 9), which is a specific mapping. Similarly, the emotion identification model 59 may determine the robot's emotion, and the identification processing unit 290 may perform identification processing using the robot's emotion.

[1860] FIG. 9 is a diagram illustrating an emotion map 400 on which multiple emotions are mapped. In the emotion map 400, emotions are arranged in concentric circles radiating from the center. Emotions closer to the center of the concentric circles are more primitive. Emotions representing states and actions arising from a state of mind are arranged on the outer edges of the concentric circles. The concept of emotion includes both affect and mental states. Emotions generally generated from reactions occurring in the brain are arranged on the left side of the concentric circles. Emotions generally induced by situational judgment are arranged on the right side of the concentric circles. Emotions generally generated from reactions occurring in the brain and induced by situational judgment are arranged on the upper and lower sides of the concentric circles. Furthermore, the emotion of "pleasure" is arranged on the upper side of the concentric circles, and the emotion of "discomfort" is arranged on the lower side. In this way, in the emotion map 400, multiple emotions are mapped based on the structure by which emotions are generated, and emotions that tend to occur simultaneously are mapped close to each other.

[1861] These emotions are distributed in the 3 o'clock direction on emotion map 400, and typically fluctuate between relief and anxiety. In the right half of emotion map 400, situational awareness dominates over internal sensations, resulting in a sense of calm.

[1862] The inside of emotion map 400 represents what is going on in the mind, and the outside of emotion map 400 represents behavior, so the further you go outside emotion map 400, the more visible the emotions become (the more they are expressed in behavior).

[1863] Human emotions are based on various balances, such as posture and blood sugar levels. When these balances deviate from the ideal, a state of discomfort is indicated, and when they approach the ideal, a state of pleasure is indicated. Emotions can also be created for robots, automobiles, and motorcycles, based on various balances, such as posture and remaining battery life. When these balances deviate from the ideal, a state of discomfort is indicated, and when they approach the ideal, a state of pleasure is indicated. An emotion map can be generated, for example, based on Dr. Mitsuyoshi's emotion map (Research on Voice Emotion Recognition and Emotional Brain Physiological Signal Analysis Systems, Tokushima University, Doctoral Dissertation: https: / / ci.nii.ac.jp / naid / 500000375379). The left half of the emotion map lists emotions belonging to the "reaction" domain, where sensation is dominant. The right half of the emotion map lists emotions belonging to the "situation" domain, where situational awareness is dominant.

[1864] The emotion map defines two emotions that promote learning. One is a negative emotion on the situation side, around the middle of "repentance" or "reflection." In other words, this occurs when the robot experiences negative emotions such as "I never want to feel this way again" or "I don't want to be scolded again." The other is a positive emotion on the response side, around "desire." In other words, this occurs when the robot experiences positive feelings such as "I want more" or "I want to know more."

[1865] The emotion identification model 59 inputs user input into a pre-trained neural network, obtains emotion values ​​indicating each emotion shown in the emotion map 400, and determines the user's emotion. This neural network is pre-trained based on multiple pieces of training data that are combinations of user input and emotion values ​​indicating each emotion shown in the emotion map 400. Furthermore, this neural network is trained so that emotions that are located close to each other have similar values, as in the emotion map 900 shown in FIG. 10. FIG. 10 shows an example in which multiple emotions, "relieved," "calm," and "reassuring," have similar emotion values.

[1866] The system according to the present disclosure has been described above mainly with respect to the functions of the data processing device 12, but the system according to the present disclosure is not necessarily implemented on a server. The system according to the present disclosure may be implemented as a general information processing system. The present disclosure may be implemented, for example, as a software program running on a personal computer or an application running on a smartphone, etc. The method according to the present disclosure may be provided to users in the form of SaaS (Software as a Service).

[1867] In the above embodiment, an example was given in which the specific processing is performed by one computer 22, but the technology of the present disclosure is not limited to this, and the specific processing may be distributed and performed by a plurality of computers including the computer 22. For example, the data generation model 58 may be provided in an external device of the data processing device 12, and data may be generated in the external device in accordance with input data.

[1868] In the above embodiment, an example in which the specific processing program 56 is stored in the storage 32 has been described, but the technology of the present disclosure is not limited to this. For example, the specific processing program 56 may be stored in a portable, computer-readable, non-transitory storage medium such as a USB (Universal Serial Bus) memory. The specific processing program 56 stored in the non-transitory storage medium is installed in the computer 22 of the data processing device 12. The processor 28 executes the specific processing in accordance with the specific processing program 56.

[1869] Alternatively, the specific processing program 56 may be stored in a storage device such as a server connected to the data processing device 12 via the network 54, and the specific processing program 56 may be downloaded and installed on the computer 22 in response to a request from the data processing device 12.

[1870] It is not necessary to store all of the specific processing program 56 in a storage device such as a server connected to the data processing device 12 via the network 54, or to store all of the specific processing program 56 in the storage 32; only a portion of the specific processing program 56 may be stored.

[1871] The hardware resource for executing a specific process can be any of the following processors: An example of a processor is a CPU, which is a general-purpose processor that functions as a hardware resource for executing a specific process by executing software, i.e., a program. Another example of a processor is a dedicated electrical circuit, such as an FPGA (Field-Programmable Gate Array), a PLD (Programmable Logic Device), or an ASIC (Application Specific Integrated Circuit), which is a processor with a circuit configuration designed specifically for executing a specific process. Each processor has built-in or connected memory, and each processor uses the memory to execute the specific process.

[1872] The hardware resource that executes the specific processing may be configured with one of these various processors, or may be configured with a combination of two or more processors of the same or different types (for example, a combination of multiple FPGAs, or a combination of a CPU and an FPGA). Also, the hardware resource that executes the specific processing may be a single processor.

[1873] As an example of a system configured with a single processor, first, one processor is configured by combining one or more CPUs and software, and this processor functions as a hardware resource that executes a specific process. Second, there is a system that uses a processor that realizes the functions of an entire system including multiple hardware resources that execute a specific process on a single IC chip, as typified by SoC (System-on-a-chip). In this way, a specific process is realized using one or more of the above-mentioned various processors as hardware resources.

[1874] Furthermore, the hardware structure of these various processors can be, more specifically, an electric circuit that combines circuit elements such as semiconductor devices. The specific processing described above is merely an example. Therefore, it goes without saying that unnecessary steps may be deleted, new steps may be added, or the processing order may be rearranged, without departing from the spirit of the invention.

[1875] The above-described description and illustrations are a detailed explanation of the parts related to the technology of the present disclosure and are merely an example of the technology of the present disclosure. For example, the above description of the configuration, functions, actions, and effects is an explanation of an example of the configuration, functions, actions, and effects of the parts related to the technology of the present disclosure. Therefore, it goes without saying that unnecessary parts may be deleted, new elements may be added, or replacements may be made to the above-described description and illustrations within the scope of the gist of the technology of the present disclosure. Furthermore, to avoid confusion and facilitate understanding of the parts related to the technology of the present disclosure, the above-described description and illustrations omit explanations of common technical knowledge that do not require particular explanation to enable the implementation of the technology of the present disclosure.

[1876] All publications, patent applications, and technical standards mentioned in this specification are herein incorporated by reference to the same extent as if each individual publication, patent application, or technical standard was specifically and individually indicated to be incorporated by reference.

[1877] The following is further disclosed regarding the above embodiment.

[1878] (Claim 1)

[1879] A means for generating answers to questions by a plurality of generative artificial intelligence models;

[1880] A means for other generative artificial intelligence models to evaluate the answers generated by each generative artificial intelligence model;

[1881] A means of selecting the most appropriate answer based on the evaluation results;

[1882] a means for outputting the selected answer as a final answer;

[1883] A system including:

[1884] (Claim 2)

[1885] 10. The system of claim 1, further comprising means for conducting a discussion and adding corrections to update the answer.

[1886] (Claim 3)

[1887] 10. The system of claim 1, further comprising means for cascading the evaluation process with multiple generative artificial intelligence models and selecting a final answer based on a majority vote.

[1888] "Example 1"

[1889] (Claim 1)

[1890] a means for inputting a question;

[1891] A means for generating answers to questions by a plurality of generative artificial intelligence models;

[1892] A means for other generative artificial intelligence models to evaluate the answers generated by each generative artificial intelligence model;

[1893] A means of selecting the most appropriate answer based on the evaluation results;

[1894] A means to correct or supplement selected responses;

[1895] a means of reassessing revised answers;

[1896] A step-by-step evaluation process with a final answer selected based on a majority vote;

[1897] means for transmitting the selected final answer to a user terminal and displaying it;

[1898] A system including:

[1899] (Claim 2)

[1900] 10. The system of claim 1, further comprising means for conducting a discussion and adding corrections to update the answer.

[1901] (Claim 3)

[1902] 10. The system of claim 1, further comprising means for performing the evaluation process in stages and selecting a final answer based on the results of a majority vote.

[1903] "Application Example 1"

[1904] (Claim 1)

[1905] A means for generating answers to questions by a plurality of generative artificial intelligence models;

[1906] A means for other generative artificial intelligence models to evaluate the answers generated by each generative artificial intelligence model;

[1907] A means of selecting the most appropriate answer based on the evaluation results;

[1908] a means for outputting the selected answer as a final answer;

[1909] A means of notifying the user of the final answer as a security alert;

[1910] A system including:

[1911] (Claim 2)

[1912] 10. The system of claim 1, further comprising means for conducting a discussion and adding corrections to update the answer.

[1913] (Claim 3)

[1914] 10. The system of claim 1, further comprising means for cascading the evaluation process with multiple generative artificial intelligence models and selecting a final answer based on a majority vote.

[1915] "Example 2: Combining Emotion Engines"

[1916] (Claim 1)

[1917] A means for generating answers to questions by a plurality of generative artificial intelligence models;

[1918] A means for other generative artificial intelligence models to evaluate the answers generated by each generative artificial intelligence model;

[1919] A means of selecting the most appropriate answer based on the evaluation results;

[1920] a means for outputting the selected answer as a final answer;

[1921] an emotion engine that recognizes a user's emotion using the user's input data;

[1922] means for assigning emotion data recognized by the emotion engine to a question and transmitting the question;

[1923] A system including:

[1924] (Claim 2)

[1925] 10. The system of claim 1, further comprising means for conducting a discussion and adding corrections to update the answer.

[1926] (Claim 3)

[1927] 10. The system of claim 1, further comprising means for cascading the evaluation process with multiple generative artificial intelligence models and selecting a final answer based on a majority vote.

[1928] "Application example 2 when combining emotion engines"

[1929] (Claim 1)

[1930] A means for generating answers to questions by a plurality of artificial intelligence models;

[1931] A means for other generation year artificial intelligence models to evaluate the answers generated by each generation year artificial intelligence model;

[1932] A means of selecting the most appropriate answer based on the evaluation results;

[1933] a means for outputting the selected answer as a final answer;

[1934] a means for transmitting a question input by a user to an emotion engine and recognizing the emotion of the user;

[1935] means for transmitting the question to which emotion data is added to a server;

[1936] A means for the server to cause a plurality of artificial intelligence models to generate answers in consideration of the emotion data;

[1937] A system including:

[1938] (Claim 2)

[1939] 10. The system of claim 1, further comprising means for conducting a discussion and adding corrections to update the answer.

[1940] (Claim 3)

[1941] 10. The system of claim 1, further comprising means for performing a step-by-step evaluation process using a plurality of generation-year artificial intelligence models and selecting a final answer based on the result of a majority vote. [Explanation of symbols]

[1942] 10, 210, 310, 410 Data Processing Systems 12 Data Processing Device 14 Smart Devices 214 Smart Glasses 314 Headset-type terminal 414 Robot< / url:> < / url:> < / url:> < / url:>

Claims

1. A means for generating answers to questions by a plurality of generative artificial intelligence models; A means for other generative artificial intelligence models to evaluate the answers generated by each generative artificial intelligence model; A means of selecting the most appropriate answer based on the evaluation results; a means for outputting the selected answer as a final answer; A system including:

2. The system of claim 1 further comprising means for providing a discussion and adding corrections to update the answer.

3. The system of claim 1 further comprising means for performing a step-by-step evaluation process using a plurality of generative artificial intelligence models and selecting a final answer based on the results of a majority vote.

Images