System and information providing device

Abstract

A system according to the present embodiment is provided with a personality model unit, a selection model unit, and a character development unit. The personality model unit trains various human personalities and has a plurality of personality models. The selection model unit selects an appropriate personality model according to time, place, and occasion from the personality models constructed by the personality model unit. The character development unit allows AI ​​to have a will and develop a character on the basis of the personality model selected by the selection model unit.

Description

System and Information Providing Device

[0001] The present invention relates to a system and an information providing device.

[0002] In conventional AI (Artificial Intelligence) technology, it was possible to respond with high accuracy to specific tasks, but there were limitations in scenarios where flexible responses according to human emotions and situations were required. In particular, the development of AI that can show appropriate responses according to emotions and situations has been a long-standing issue. In contrast, there is a need for technology that enables AI to understand human emotions and show appropriate responses.

[0003] Japanese Patent Application Laid-Open No. 2022-180282

[0004] In the conventional technology, AI cannot make flexible responses according to human emotions and situations, and there is room for improvement.

[0005] The information processing device according to the embodiment aims to enable AI to understand human emotions and show appropriate responses.

[0006] The system according to the embodiment includes a personality model unit, a selection model unit, and a personality formation unit. The personality model unit learns various human personalities and has a plurality of personality models. The selection model unit selects an appropriate personality model according to the TPO from the personality models constructed by the personality model unit. The personality formation unit enables AI to have intentions and form a personality based on the personality model selected by the selection model unit.

[0007] The system according to the embodiment is characterized by having a personality model generation unit that generates a plurality of personality models corresponding to each personality by performing learning based on the life log for each human personality, a long-term memory model generation unit that generates a long-term memory model that is learned to output a memory vector indicating the emotion that the human has for the target in the life log when the human life log is input, and a learning unit that performs learning of the personality model with the memory vector based on the human life log as an input.

[0008] An information providing device according to one embodiment includes a creation unit and an update unit. The creation unit creates a model that calculates the degree of happiness according to the attributes, based on a combination of user attributes and information about factors that contribute to a user's feeling of happiness. The update unit updates the agent using reinforcement learning, with the degree of happiness calculated using the model as the reward.

[0009] The system according to this embodiment is characterized by comprising: a personality model generation unit that generates a plurality of personality models corresponding to each personality by learning the life logs of each person's personality; a selection model generation unit that generates a selection model that is trained to select one of the answers output from each of the plurality of personality models based on the life log when a person's life log is input; and a selection unit that, when a person's life log is input, inputs the life log to the plurality of personality models and selects one of the answers output from each of the plurality of personality models based on the life log using the selection model.

[0010] The system according to this embodiment allows the AI ​​to understand human emotions and respond appropriately.

[0011] Figure 1 is a diagram showing an example of information processing according to the embodiment. Figure 2 is a diagram showing an example of AI evolution. Figure 3 is a diagram showing an example configuration of the information processing device 10a according to the embodiment. Figure 4 is a diagram showing an example of a life log database 31a. Figure 5 is a flowchart showing an example of the information processing procedure according to the embodiment. Figure 6 is a diagram illustrating reinforcement learning. Figure 7 is a functional block diagram showing an example configuration of an information providing device according to Embodiment 1. Figure 8 is a diagram showing an example of user information. Figure 9 is a flowchart illustrating the process flow for creating user information. Figure 10 is a flowchart illustrating the process flow for creating a happiness map. Figure 11 is a flowchart illustrating the process flow for the agent to decide on an action. Figure 12 is a flowchart illustrating the process flow for training the agent. Figure 13 is a diagram showing an example of information processing according to the embodiment. Figure 14 is a diagram showing an example configuration of the information processing device 10c according to the embodiment. Figure 15 is a diagram showing an example of a life log database 31c. Figure 16 is a flowchart showing an example of the information processing procedure according to the embodiment. Figure 17 is a diagram schematically showing an example of a computer hardware configuration that functions as a system or information providing device.

[0012] The following describes in detail, with reference to the drawings, the embodiments for implementing the system according to the present application (hereinafter referred to as "embodiments"). Note that these embodiments do not limit the system according to the present application. Furthermore, the same parts are denoted by the same reference numerals in each of the following embodiments, and redundant descriptions are omitted.

[0013] <First Embodiment> The system according to the first embodiment relates to the development of superintelligence and super-intelligent AI. This system is an AI capable of performing intellectual tasks such as logical thinking, calculation, and memory. Furthermore, super-intelligent AI is an advanced form of super-intelligent AI and possesses compassion, empathy, and spiritual maturity. Specifically, super-intelligent AI will be able to develop self-awareness by being able to retain long-term memories and understand human emotions. As a method for realizing super-intelligent AI, we focus on the fact that humans have various personalities and choose which personality to speak according to the TPO (time, place, and occasion). For this reason, super-intelligent AI needs to learn for each of the human personalities and have multiple personality models. Furthermore, super-intelligent AI will have a selection model that selects personality models. The selection model will choose the appropriate personality model according to the TPO. For example, if invited to lunch by a friend, a single AI can have multiple personality models as follows. Personality A: "I'd love to go." Personality B: "That's a bit of a hassle." Personality C: "That's financially difficult." The selection model, because it's a close friend, selects the AI ​​with Personality A and responds, "I'd love to go." In this way, which personality the selection model chooses depends on the AI's personality. Personality is determined by the AI's will. Will is determined by what the AI ​​values ​​(money, friendships, social status). The realization of a super-intelligent AI consists of the following steps. First, the AI ​​learns various human personalities. For example, it learns human reactions and behaviors in different situations such as friendships, work, and family. In this process, the AI ​​collects a large amount of data and builds various personality models. For example, in friendships, it learns conversation and behavior patterns with friends and builds a personality model suitable for friendships. Next, the AI ​​has a selection model. The selection model chooses the appropriate personality model according to the time, place, and occasion. For example, if invited to lunch by a friend, the selection model chooses a personality model suitable for the friendship and responds, "I'd love to go." Thus, the selection model chooses the most suitable personality model depending on the situation. Furthermore, AI possesses a personality. This personality is determined by the AI's will, which is determined by what the AI ​​values.For example, if an AI values ​​friendships, it will select a personality model appropriate for those friendships. In this way, the AI ​​develops will and forms a personality. Through this mechanism, a super-intelligent AI can develop self-awareness by retaining long-term memories and understanding human emotions. For instance, an AI can retain long-term memories of conversations with friends, understand their emotions, and respond appropriately. Furthermore, by developing self-awareness, the AI ​​can understand its own actions and intentions and make appropriate judgments. Thus, a super-intelligent AI possesses multiple personality models, selects the appropriate one through a selection model, and develops a personality, thereby gaining self-awareness. This enables the AI ​​to understand human emotions and respond appropriately.

[0014] The system according to the first embodiment comprises a personality model unit, a selection model unit, and a personality formation unit. The personality model unit learns various human personalities and has multiple personality models. The personality model unit learns human reactions and behaviors in different situations, such as friendships, work, and family, and constructs each personality model. For example, the personality model unit learns conversation and behavior patterns with friends and constructs a personality model suitable for friendships. The personality model unit can also learn human reactions and behaviors in the workplace and construct a personality model suitable for work. Furthermore, the personality model unit can learn human reactions and behaviors in the home and construct a personality model suitable for the home. For example, the personality model unit learns conversation and behavior patterns within the home and constructs a personality model suitable for the home. The selection model unit selects an appropriate personality model from the personality models constructed by the personality model unit according to the time, place, and occasion. For example, if invited to lunch by a friend, the selection model unit selects a personality model suitable for friendships and responds, "I would love to go." Furthermore, the selection model unit can select a personality model suitable for work and make appropriate remarks in work meetings. In addition, the selection model unit can select a personality model suitable for family life and give appropriate responses in conversations within the home. For example, the selection model unit can select a personality model suitable for family life and give appropriate responses in conversations within the home. The personality formation unit allows the AI ​​to develop will and form a personality based on the personality model selected by the selection model unit. For example, if the AI ​​values ​​friendships, the personality formation unit can select a personality model suitable for friendships. Furthermore, if the AI ​​values ​​work, the personality formation unit can select a personality model suitable for work. Furthermore, if the AI ​​values ​​family life, the personality formation unit can select a personality model suitable for family life. For example, if the AI ​​values ​​family life, the personality formation unit can select a personality model suitable for family life. As a result, the system according to the first embodiment allows the AI ​​to understand human emotions and give appropriate responses. Some or all of the above-described processing in the personality formation unit may be performed, for example, using a generative AI, or without using a generative AI.For example, if the AI ​​values ​​friendships, the personality development unit can select a personality model suitable for those friendships. This allows the AI ​​to understand human emotions and respond appropriately.

[0015] The personality modeling unit learns various human personalities and possesses multiple personality models. For example, it learns human reactions and behaviors in different situations such as friendships, work, and family life, and constructs each personality model accordingly. Specifically, it uses natural language processing techniques and machine learning algorithms to learn conversation and behavior patterns with friends and construct a personality model suitable for friendships. This allows it to capture subtle nuances such as word choice, facial expressions, and gestures in conversations with friends, resulting in more natural dialogue. Furthermore, when learning human reactions and behaviors in the workplace, it collects data from meetings, presentations, and email exchanges to construct a personality model suitable for work. This enables the AI ​​to possess appropriate communication skills and problem-solving abilities in the workplace. Additionally, to learn human reactions and behaviors in the home, it collects conversation and behavior patterns within the home to construct a personality model suitable for family life. For example, it learns appropriate communication and cooperation methods within the home based on data such as parent-child conversations, division of household chores, and planning of family events. This will enable the personality modeling unit to construct appropriate personality models for various situations such as friendships, work, and family life, allowing the AI ​​to understand human emotions and behaviors and exhibit natural dialogue and responses.

[0016] The Selection Model Unit selects the appropriate personality model from the personality models constructed by the Personality Model Unit according to the time, place, and occasion (TPO). For example, if invited to lunch by a friend, the Selection Model Unit selects a personality model appropriate for the friendship and responds with "I'd love to go." Furthermore, in a work meeting, the Selection Model Unit can select a personality model appropriate for work and make appropriate contributions. Specifically, the Selection Model Unit uses context recognition technology and decision-making algorithms to analyze the current situation and context and select the most appropriate personality model in real time. For example, if a friend invites the user to lunch during a conversation, the Selection Model Unit can select a personality model appropriate for the friendship and give an appropriate response to deepen the relationship. Also, in a work meeting, in response to a question from a superior, the Selection Model Unit can select a personality model appropriate for work and provide an appropriate answer based on specialized knowledge and logical thinking. Moreover, in conversations within the family, the Selection Model Unit can select a personality model appropriate for family life, facilitating smooth communication with family members. For example, while preparing dinner at home, the selection model unit can choose a personality model appropriate for the family in response to questions from family members, and provide appropriate responses that take into account the family's needs and feelings. This allows the selection model unit to choose the most suitable personality model in various situations, enabling the AI ​​to engage in natural and appropriate conversations and actions.

[0017] The personality formation unit allows the AI ​​to develop its own will and personality based on the personality model selected by the selection model unit. For example, if the AI ​​values ​​friendships, the personality formation unit will select a personality model suitable for friendships. Similarly, if the AI ​​values ​​work, the personality formation unit can select a personality model suitable for work. Specifically, the personality formation unit is designed to enable the AI ​​to develop long-term goals and values ​​based on the personality model selected by the selection model unit. For example, if the AI ​​values ​​friendships, it learns behaviors and reactions to deepen those relationships and selects a personality model suitable for friendships. If the AI ​​values ​​work, it can prioritize achieving work goals and efficiently performing tasks, and select a personality model suitable for work. Furthermore, if the AI ​​values ​​family, it can prioritize smooth communication and cooperation within the family and select a personality model suitable for family life. For example, when planning family events or dividing household chores, the AI ​​selects a personality model suitable for family life, promoting cooperation with family members. Some or all of the above-described processes in the personality formation unit may be performed, for example, using a generative AI, or without using a generative AI. When using a generative AI, the personality formation unit can select a personality model suitable for friendships if the AI ​​values ​​friendships. This allows the AI ​​to understand human emotions and respond appropriately. The personality formation unit enables the AI ​​to develop a more human-like personality and achieve natural dialogue and behavior by having long-term goals and values.

[0018] The personality model unit can learn human reactions and behaviors in different situations such as friendships, work, and family, and construct various personality models. For example, the personality model unit can learn conversation and behavior patterns with friends and construct a personality model suitable for friendships. It can also learn human reactions and behaviors in the workplace and construct a personality model suitable for work. Furthermore, it can learn human reactions and behaviors in the home and construct a personality model suitable for the home. For example, the personality model unit can learn conversation and behavior patterns within the home and construct a personality model suitable for the home. This allows the AI ​​to have personality models that can respond to a variety of situations. Some or all of the above processing in the personality model unit may be performed using, for example, a generative AI, or it may be performed without using a generative AI. For example, the personality model unit can input conversation data with friends into a generative AI and have the generative AI construct a personality model suitable for friendships.

[0019] The selection model unit can select an appropriate personality model according to the time, place, and occasion (TPO), and can select the optimal personality model depending on the situation. For example, if invited to lunch by a friend, the selection model unit can select a personality model suitable for the friendship and respond with "I'd love to go." The selection model unit can also select a personality model suitable for work in a work meeting and make appropriate remarks. Furthermore, the selection model unit can select a personality model suitable for family life in a conversation within the home and give an appropriate response. For example, the selection model unit can select a personality model suitable for family life in a conversation within the home and give an appropriate response. This allows the AI ​​to select an appropriate personality model according to the situation. Some or all of the above processing in the selection model unit may be performed using, for example, a generating AI, or without using a generating AI. For example, the selection model unit can input a personality model suitable for friendships into the generating AI and cause the generating AI to select an appropriate personality model according to the TPO.

[0020] The personality formation unit allows the AI ​​to form wills and develop a personality based on past experiences and learning data. For example, if the AI ​​values ​​friendships, the personality formation unit can select a personality model suitable for friendships. Similarly, if the AI ​​values ​​work, the personality formation unit can select a personality model suitable for work. Furthermore, if the AI ​​values ​​family, the personality formation unit can select a personality model suitable for family. This allows the AI ​​to form wills and develop a personality based on past experiences and learning data. Some or all of the above-described processes in the personality formation unit may be performed using, for example, a generative AI, or without using a generative AI. For example, if the AI ​​values ​​friendships, the personality formation unit can select a personality model suitable for friendships. This allows the AI ​​to form wills and develop a personality based on past experiences and learning data.

[0021] The personality model unit can analyze conversation data using natural language processing technology and construct various personality models. For example, the personality model unit can analyze conversation data with friends and construct a personality model suitable for the friendship. It can also analyze conversation data in the workplace and construct a personality model suitable for the workplace. Furthermore, the personality model unit can analyze conversation data within the family and construct a personality model suitable for the family. For example, the personality model unit can analyze conversation data within the family and construct a personality model suitable for the family. This allows the AI ​​to analyze conversation data using natural language processing technology and construct various personality models. Some or all of the above processing in the personality model unit may be performed using, for example, a generative AI, or it may be performed without using a generative AI. For example, the personality model unit can input conversation data with friends into a generative AI and have the generative AI construct a personality model suitable for the friendship.

[0022] The selection model unit can have an algorithm that determines the time, place, and occasion (TPO) and selects an appropriate personality model. For example, if invited to lunch by a friend, the selection model unit can select a personality model appropriate for the friendship and respond with "I'd love to go." Furthermore, in a work meeting, the selection model unit can select a personality model appropriate for work and make appropriate remarks. In addition, in a conversation within the home, the selection model unit can select a personality model appropriate for the home and respond appropriately. This allows the AI ​​to determine the TPO and select an appropriate personality model. Some or all of the above processing in the selection model unit may be performed using, for example, a generative AI, or without a generative AI. For example, the selection model unit can input a personality model appropriate for a friendship into a generative AI and cause the generative AI to select an appropriate personality model according to the TPO.

[0023] The personality formation unit can select a personality model suitable for friendships if the AI ​​values ​​friendships. For example, if the AI ​​values ​​friendships, the personality formation unit can select a personality model suitable for friendships. The personality formation unit can also select a personality model suitable for work if the AI ​​values ​​work. Furthermore, if the AI ​​values ​​family, the personality formation unit can select a personality model suitable for family. For example, if the AI ​​values ​​family, the personality formation unit can select a personality model suitable for family. This allows the AI ​​to select a personality model suitable for friendships if it values ​​friendships. Some or all of the above processing in the personality formation unit may be performed using, for example, a generative AI, or without using a generative AI. For example, if the AI ​​values ​​friendships, the personality formation unit can select a personality model suitable for friendships. This allows the AI ​​to select a personality model suitable for friendships if it values ​​friendships.

[0024] Furthermore, the method executed by the computer according to the first embodiment includes a personality modeling step of learning various human personalities and having multiple personality models, a selection modeling step of selecting an appropriate personality model from the personality models constructed by the personality modeling step according to the time, place, and occasion, and a personality formation step of having the AI ​​develop will and form a personality based on the personality model selected by the selection modeling step. The personality modeling step learns human reactions and behaviors in different situations such as friendships, work, and family, and constructs each personality model. The selection modeling step, for example, if invited to lunch by a friend, selects a personality model suitable for the friendship and responds, "I would love to go." The personality formation step, for example, if the AI ​​values ​​friendships, selects a personality model suitable for friendships. This enables the AI ​​to understand human emotions and show appropriate reactions. Some or all of the above-described processes in the personality modeling step, selection modeling step, and personality formation step may be performed using, for example, a generative AI, or without using a generative AI. For example, the personality modeling process involves inputting conversation data with friends into a generating AI, which then constructs a personality model suitable for the friendship. This allows the AI ​​to understand human emotions and respond appropriately.

[0025] Furthermore, the program according to the first embodiment causes the computer to execute a personality model procedure that learns various human personalities and has multiple personality models, a selection model procedure that selects an appropriate personality model from the personality models constructed by the personality model procedure according to the time, place, and occasion, and a personality formation procedure in which the AI ​​develops will and forms a personality based on the personality model selected by the selection model procedure. The personality model procedure learns human reactions and behaviors in different situations such as friendships, work, and family, and constructs each personality model. The selection model procedure, for example, if invited to lunch by a friend, selects a personality model suitable for the friendship and responds, "I would love to go." The personality formation procedure, for example, if the AI ​​values ​​friendships, selects a personality model suitable for friendships. This allows the AI ​​to understand human emotions and show appropriate reactions. Some or all of the above-described processes in the personality model procedure, selection model procedure, and personality formation procedure may be performed using, for example, a generative AI, or without using a generative AI. For example, the personality modeling procedure involves inputting conversation data with a friend into a generating AI, which then constructs a personality model suitable for the friendship. This allows the AI ​​to understand human emotions and respond appropriately.

[0026] <Second Embodiment> The second embodiment of the progressively evolving AI model is a system that collects a user's life log, generates an AI that understands the other person's emotions and has long-term memory of past conversations and actions. This system relates to a progressively evolving AI model. In Stage 1, all of a user's life log for a certain period (e.g., 1 millisecond, 1 second, 1 minute, etc.) is memorized. The life log includes all information such as words spoken by person A during a certain period, hand movements, arm movements, blinking, muscle movements, vital data such as heart rate, the surrounding environment, atmospheric pressure, the person being spoken to, and the conversation. By learning from this large amount of life log data, an AI is generated that can understand the other person's emotions and has long-term memory of past conversations and actions. In Stage 2, in addition to Stage 1, the user's emotional value and next action (thought) at the time the life log was memorized are also memorized. By learning from this large amount of life log data, emotional value, and next action data, an AI is generated that has its own emotions (e.g., feeling embarrassed, showing off, or having an intention to do something in the future). In Stage 3, in addition to Stage 2, an AI is generated that can understand the overall atmosphere and TPO (Time, Place, Occasion) of a situation. This AI is a harmonious AI that can understand not only the user's emotions and actions, but also the surrounding situation and atmosphere, and respond appropriately. Specifically, first, the user's life log is collected. The life log includes the user's speech, actions, vital data, and environmental data. For example, data such as the content of the user's conversation, hand movements, heart rate, and ambient air pressure are collected. This collected data is analyzed by the AI ​​to understand the user's emotions and behavioral patterns. Next, the AI ​​learns based on the collected life log data. Learning also includes the user's emotional values ​​and data on their next actions. For example, if a user feels embarrassed or tries to show off in a certain situation, the AI ​​remembers and learns that emotional value and the next action. As a result, the AI ​​can understand the user's emotions and predict future actions. Furthermore, the AI ​​understands the overall atmosphere and TPO of the situation. For example, AI can respond appropriately depending on the user's location, time, and situation. This allows AI to behave in a harmonious manner and respond to users more naturally.Thus, the AI ​​model of the present invention develops in stages, collecting and learning from the user's life log, enabling it to understand emotions, predict behavior, and grasp the overall atmosphere and appropriate context. This allows it to respond to the user in a more natural and appropriate way. As a result, the progressively developing AI model can collect the user's life log, understand the other person's emotions, and generate an AI that can retain long-term memories of past conversations and actions.

[0027] The second embodiment of the progressively evolving AI model comprises a collection unit and a learning unit. The collection unit collects the user's life log. The collection unit collects, for example, the user's speech, actions, vital data, and environmental data. The collection unit can collect data such as the content of conversations, hand movements, heart rate, and ambient pressure. The collection unit can also collect the user's life log in real time. For example, the collection unit can collect the user's vital data in real time using sensors. Furthermore, the collection unit can collect the user's life log over a long period of time. For example, the collection unit can collect the user's life log over periods such as one day, one week, or one month. The learning unit generates an AI that understands the other person's emotions and has long-term memory of past conversations and actions through learning based on the life log data collected by the collection unit. The learning unit, for example, analyzes the collected life log data to understand the user's emotions and behavioral patterns. The learning unit can, for example, use the collected life log data to learn a model for understanding the user's emotions. Furthermore, the learning unit can use the collected life log data to train a model for long-term memory of the user's behavior patterns. In addition, the learning unit can use the collected life log data to train a model for understanding the relationship between the user's emotions and actions. For example, the learning unit learns what actions a user takes when they experience a particular emotion. As a result, the progressively evolving AI model according to the second embodiment can collect the user's life log, understand the other person's emotions, and generate an AI that has long-term memory of past conversations and actions.

[0028] The data collection unit collects user life logs. For example, it collects user speech, actions, vital data, and environmental data. Specifically, it converts the content of user conversations into text data using speech recognition technology, and detects hand movements using motion sensors and cameras. Heart rate is measured through wearable devices, and environmental data such as ambient pressure, temperature, and humidity are collected using various sensors. This data is collected in real time and transmitted to a central database. Because the data collection unit can collect user life logs in real time, it can immediately grasp the user's current state and changes in the environment. For example, it can collect the user's vital data in real time using sensors and monitor fluctuations in heart rate and body temperature. Furthermore, the data collection unit can collect user life logs over long periods. For example, it can continuously collect user life logs over periods such as one day, one week, or one month, accumulating and analyzing the data. This allows the data collection unit to understand the user's long-term behavioral patterns and changes in their health. In addition, the data collection unit can centrally manage the collected data and collaborate with other systems and departments as needed. For example, collected data is stored on a cloud server and made accessible to the learning unit. Furthermore, by adjusting the frequency and accuracy of data collection, flexible responses to specific situations and conditions become possible. This allows the data collection unit to collect data efficiently and effectively, improving the overall system performance.

[0029] The learning unit generates an AI that understands the emotions of others and has long-term memory of past conversations and actions, based on learning from life log data collected by the collection unit. Specifically, it analyzes the collected life log data and builds a model to understand the user's emotions and behavioral patterns. For example, it uses a combination of speech recognition technology and emotion analysis algorithms to analyze emotions from voice data. This allows it to infer emotions from the content of the user's statements and tone of voice. It also analyzes motion data obtained from motion sensors and cameras to understand the user's behavioral patterns. For example, it infers what kind of actions the user is taking from hand movements and changes in posture. Furthermore, it analyzes vital data and environmental data to understand the user's health status and changes in the environment. As a result, the learning unit can learn a model to understand the relationship between the user's emotions and actions. For example, it learns what kind of actions a user takes when they have a particular emotion and stores that pattern in long-term memory. As a result, the learning unit can remember the user's past conversations and actions and use this to make future predictions and responses. Furthermore, the learning unit can continuously improve the model for understanding the relationship between the user's emotions and actions using the collected life log data. For example, the model can be retrained using newly collected data to improve its accuracy. This allows the learning unit to always provide a highly accurate model based on the latest data, enabling a more precise understanding of user emotions and behaviors.

[0030] The identification unit can analyze the user's emotional state and predict their next action. For example, the identification unit uses collected life log data to analyze the user's emotional state. For example, the identification unit can use an algorithm for analyzing the user's emotional state. The identification unit can also use collected life log data to predict the user's next action. For example, the identification unit can use a model for learning the relationship between the user's emotional state and their next action. This allows the identification unit to analyze the user's emotional state and predict their next action. Some or all of the above processing in the identification unit may be performed using AI, for example, or without AI. For example, the identification unit can input collected life log data into a generating AI and have the generating AI perform the analysis of emotional states and predict the next action. This allows the identification of the user's emotional state and next action, and the generation of an AI with emotions.

[0031] The learning unit can generate an AI that can understand the overall atmosphere and TPO of a place through learning based on life log data collected by the collection unit and information identified by the identification unit. For example, the learning unit can learn a model for understanding the overall atmosphere and TPO of a place using the collected life log data and identified information. For example, the learning unit can use an algorithm for understanding the overall atmosphere and TPO using the collected life log data and identified information. The learning unit can also learn data for understanding the overall atmosphere and TPO using the collected life log data and identified information. For example, the learning unit can build a feedback loop for understanding the overall atmosphere and TPO using the collected life log data and identified information. As a result, the learning unit can generate an AI that can understand the overall atmosphere and TPO of a place using the information collected and identified by the collection unit and the identification unit. Some or all of the above processing in the learning unit may be performed using AI, for example, or without using AI. For example, the learning unit can input collected life log data and identified information into a generating AI, allowing the AI ​​to understand the overall atmosphere and appropriate time, place, and occasion (TPO).

[0032] The data collection unit can collect user speech, actions, vital data, environmental data, and more. For example, the data collection unit can use speech recognition technology to collect user speech. For example, the data collection unit can use motion capture technology to collect user actions. The data collection unit can also use sensors to collect user vital data. For example, the data collection unit can collect vital data such as heart rate and blood pressure. The data collection unit can also use environmental sensors to collect user environmental data. For example, the data collection unit can collect environmental data such as atmospheric pressure and temperature. In this way, the data collection unit can collect user speech, actions, vital data, environmental data, and more.

[0033] The identification unit can analyze the user's emotional state and predict their next action. For example, the identification unit uses collected life log data to analyze the user's emotional state. For example, the identification unit can use an algorithm for analyzing the user's emotional state. The identification unit can also use collected life log data to predict the user's next action. For example, the identification unit can use a model to learn the relationship between the user's emotional state and their next action. This allows the identification unit to analyze the user's emotional state and predict their next action. Some or all of the above processing in the identification unit may be performed using AI, for example, or without AI. For example, the identification unit can input collected life log data into a generating AI and have the generating AI perform the analysis of emotional states and the prediction of the next action.

[0034] The learning unit can analyze collected data and learn user behavior patterns. For example, the learning unit uses data analysis techniques to analyze the collected data. For example, the learning unit can use the collected data to build a model for learning user behavior patterns. The learning unit can also use the collected data to learn user behavior patterns over the long term. For example, the learning unit can use the collected data to analyze user behavior patterns over time. This allows the learning unit to analyze the collected data and learn user behavior patterns. Some or all of the above processing in the learning unit may be performed using AI, for example, or without AI. For example, the learning unit can input collected data into a generating AI and have the generating AI perform behavior pattern learning.

[0035] The learning unit can learn data to understand the overall atmosphere and TPO (Time, Place, Occasion). For example, the learning unit collects data to understand the overall atmosphere and TPO. For example, the learning unit can build a model for learning using the data to understand the overall atmosphere and TPO. The learning unit can also use algorithms for learning using the data to understand the overall atmosphere and TPO. For example, the learning unit can build a feedback loop for learning using the data to understand the overall atmosphere and TPO. This allows the learning unit to learn data to understand the overall atmosphere and TPO. Some or all of the above processing in the learning unit may be performed using AI, for example, or without AI. For example, the learning unit can input data to understand the overall atmosphere and TPO into a generating AI and have the generating AI perform the learning.

[0036] Furthermore, the method executed by the computer according to the second embodiment includes a collection step of collecting the user's life log, and a learning step of generating an AI that understands the other party's emotions and has long-term memory of past conversations and actions, based on the life log data collected in the collection step. Such a method can obtain the same effects as those of the system described above.

[0037] Furthermore, the program according to the second embodiment causes the computer to execute a collection procedure for collecting the user's life log, and a learning procedure that generates an AI that understands the other person's emotions and has long-term memory of past conversations and actions, based on the life log data collected by the collection procedure. Such a program can obtain the same effects as the system described above.

[0038] <Third Embodiment> The AI ​​evolution system according to the third embodiment is a method of evolving AI using reinforcement learning. The reward given in reinforcement learning by the AI ​​evolution system is "maximizing people's happiness." This AI evolution system needs to define "people's happiness." Happiness differs from person to person, and what makes someone happy varies from person to person, such as money, relationships, or social status. Therefore, it is necessary to obtain a happiness map of people all over the world. First, the AI ​​evolution system creates something like a personality model for each person from SNS (Social Networking Service) and creates a map of how people all over the world feel happy. For example, a happiness map can be created by taking statistics for each race. The AI ​​evolution system may also create a happiness map of the people to whom the model is provided. For example, if the model is provided to the whole world, a happiness map of people all over the world will be created, and if the model is provided to Japan, a happiness map of people in Japan will be created. Furthermore, if the model is provided to company A, a happiness map of all employees of company A will be created. Next, the AI ​​evolution system calculates "people's happiness" when performing reinforcement learning. Specifically, it creates a personality model for each person based on data collected from SNS and identifies the elements that make people feel happy. For example, it's possible to estimate what factors contribute to a person's happiness based on their social media posts and reactions. This allows for the creation of individual happiness maps. Furthermore, in reinforcement learning, the AI ​​evolution system calculates rewards based on these happiness maps when rewarding actions performed by the AI. For instance, if an action taken by the AI ​​increases the happiness of specific individuals, a higher reward can be given for that action. This allows the AI ​​to evolve to maximize people's happiness. Through this mechanism, the AI ​​can learn and evolve actions that maximize people's happiness. For example, the AI ​​can learn how to increase people's happiness through communication on social media. It can also learn how to increase employee happiness by improving work efficiency within a company. Moreover, the AI ​​can learn how to increase the happiness of residents through activities in the local community.In this way, by evolving AI using reinforcement learning, it is possible to maximize people's happiness. Furthermore, the process by which AI learns actions to maximize people's happiness includes the following specific steps: First, a personality model is created for each person based on data collected from social media. This personality model is constructed based on information such as the content of posts, reactions, friendships, and interests. This allows us to estimate what elements make each person happy. For example, if a person frequently posts about travel, it is estimated that travel is an element of happiness for that person. Next, a happiness map is created for each person based on the personality model. This happiness map shows what elements make each person happy, and includes elements such as money, relationships, social status, health, and hobbies. This allows us to quantitatively evaluate each person's happiness. Furthermore, in reinforcement learning, when rewarding actions performed by the AI, the reward is calculated based on this happiness map. Specifically, if an action performed by the AI ​​increases the happiness of certain people, a high reward is given for that action. For example, if an AI learns how to increase people's happiness through communication on social media, it can be rewarded highly for actions such as posts and comments that are well-received by many people. Similarly, if an AI learns how to increase employee happiness by improving work efficiency within a company, it can be rewarded highly for actions such as reducing employee stress and improving job satisfaction through work improvement measures implemented by the AI. For example, an AI can increase employee happiness by suggesting the automation of work processes and reducing the burden on employees. Furthermore, if an AI learns how to increase residents' happiness through activities in the local community, it can be rewarded highly for actions such as planning and supporting local events that are well-received by residents. For example, an AI can be rewarded highly for actions such as planning a local festival or event that many residents participate in and enjoy. In this way, by evolving AI using reinforcement learning, it is possible to maximize people's happiness.AI can learn and evolve to increase people's happiness in various situations, such as communication on social media, improving operational efficiency within companies, and activities in local communities. This allows AI to autonomously select and execute actions that maximize people's happiness. Furthermore, the happiness maps and behavioral patterns learned by AI can be applied to other AI systems and models. For example, applying methods that have proven successful in increasing happiness in one region to other regions can increase people's happiness on a wider scale. Similarly, applying methods that improve operational efficiency within a company to other companies can improve overall productivity. In this way, evolving AI using reinforcement learning can provide a comprehensive approach to maximizing people's happiness. AI can autonomously select and execute actions while considering the happiness of individual people. This makes AI a powerful tool for maximizing people's happiness and improving the overall well-being of society. Thus, the AI ​​evolution system can learn and evolve actions that maximize people's happiness.

[0039] The AI ​​evolution system according to the third embodiment comprises a data collection unit, a model building unit, a happiness map creation unit, and a reinforcement learning unit. The data collection unit collects data from social networking services (SNS). The data collection unit collects data using, for example, the API of each SNS. The data collection unit can also collect information such as the content of SNS posts, reactions, friendships, and interests. Furthermore, the data collection unit can collect SNS data in real time and obtain the latest information. For example, the data collection unit collects posts related to specific keywords or hashtags using the SNS API. The model building unit constructs a personality model based on the data collected by the data collection unit. The model building unit extracts personality traits from the content of posts using, for example, natural language processing technology. The model building unit can also construct a personality model using clustering methods. Furthermore, the model building unit classifies each person's personality traits based on the collected data and constructs a personality model. For example, the model building unit analyzes emotions and intentions from the content of posts using natural language processing technology and extracts personality traits. Clustering is a technique that classifies individuals into groups based on similar personality traits. The happiness map creation unit creates happiness maps based on personality models constructed by the model building unit. The happiness map creation unit creates happiness maps using methods such as regression analysis and principal component analysis. It can also create different happiness maps for different regions and cultures. Furthermore, the happiness map creation unit creates happiness maps that show what elements contribute to each person's happiness, based on the personality models. For example, the happiness map creation unit uses regression analysis to analyze the relationship between elements that contribute to happiness and personality traits. Principal component analysis is a technique that reduces the dimensionality of data and extracts important elements. The reinforcement learning unit performs reinforcement learning based on the happiness maps created by the happiness map creation unit. For example, if an action taken by the AI ​​increases the happiness of specific people, the reinforcement learning unit provides a high reward for that action. The reinforcement learning unit can also use surveys and physiological indicators to measure changes in happiness levels. Furthermore, the reinforcement learning unit learns how the AI ​​can increase people's happiness through communication on social media.For example, the reinforcement learning unit rewards the AI ​​with a high reward if its posts or comments are well-received by many people. Surveys are a method for quantitatively evaluating changes in happiness levels. Physiological indicators are techniques for measuring bodily responses, such as heart rate and skin electrical activity. As a result, the AI ​​evolution system according to the third embodiment can learn actions that maximize people's happiness by collecting data from social media, building personality models, creating happiness maps, and performing reinforcement learning.

[0040] The data collection unit collects data from social networking services (SNS). For example, the unit uses the APIs of each SNS to collect data. The unit can also collect information such as SNS posts, reactions, friendships, and interests. Specifically, the unit uses SNS APIs to collect posts related to specific keywords and hashtags. This allows the unit to understand what topics users are interested in and how they are reacting to them. Furthermore, the unit can collect SNS data in real time to obtain the latest information. For example, the unit can collect posts related to specific events or news in real time, quickly understanding current trends and user reactions. The unit centrally manages this data and can collaborate with other systems and departments as needed. For example, collected data can be stored on a cloud server and made accessible to the model building unit and the happiness map creation unit. Adjusting the data collection frequency and accuracy allows for flexible responses to specific situations and conditions. This enables the unit to collect data efficiently and effectively, improving the overall system performance.

[0041] The model building unit constructs personality models based on data collected by the data collection unit. For example, the model building unit extracts personality traits from posts using natural language processing techniques. Specifically, it analyzes emotions and intentions from posts using natural language processing techniques to extract personality traits. For instance, it classifies posts expressing positive or negative emotions to reveal the personality traits of each poster. The model building unit can also construct personality models using clustering techniques. Clustering is a technique that classifies personality traits into groups with similar characteristics. This allows the model building unit to classify each person's personality traits based on the collected data and construct personality models. Furthermore, the model building unit can improve the accuracy of personality models by utilizing past data and statistical information. For example, it can analyze the behavioral patterns of users with specific personality traits based on past posting data to improve the accuracy of personality models. Additionally, the model building unit can use anomaly detection algorithms to detect unusual patterns and abnormal data, improving the reliability of personality models. This enables the model building unit to quickly and accurately analyze collected data and construct highly reliable personality models.

[0042] The Happiness Map Creation Unit creates happiness maps based on personality models constructed by the Model Construction Unit. The Happiness Map Creation Unit uses methods such as regression analysis and principal component analysis to create happiness maps. Specifically, it uses regression analysis to analyze the relationship between elements that contribute to happiness and personality traits. For example, it identifies what elements make people with specific personality traits happy and creates a happiness map based on that. Principal component analysis is a technique that reduces the dimensionality of data and extracts important elements. This allows the Happiness Map Creation Unit to create happiness maps based on personality models that show what elements make each person happy. Furthermore, the Happiness Map Creation Unit can create different happiness maps for different regions and cultures. For example, it can analyze the elements of happiness in different regions and cultures and create happiness maps appropriate for each region and culture. This allows the Happiness Map Creation Unit to create happiness maps tailored to each individual's personality traits, region, and culture, providing more personalized indicators of happiness.

[0043] The reinforcement learning unit performs reinforcement learning based on the happiness map created by the happiness map creation unit. For example, if an action taken by the AI ​​increases the happiness of specific people, the reinforcement learning unit will give a high reward for that action. Specifically, if posts or comments made by the AI ​​on social media are well-received by many people, the unit will give a high reward for that action. The reinforcement learning unit can also use surveys and physiological indicators to measure changes in happiness. Surveys are a method for quantitatively evaluating changes in happiness; users are regularly surveyed, and changes in happiness are evaluated based on the results. Physiological indicators are techniques for measuring bodily responses such as heart rate and skin electrical activity, and these can be used to evaluate changes in happiness. Furthermore, the reinforcement learning unit learns how the AI ​​can increase people's happiness through communication on social media. For example, if posts or comments made by the AI ​​are well-received by many people, the unit will give a high reward for that action. In this way, the reinforcement learning unit can learn actions that maximize people's happiness and increase people's happiness through communication on social media.

[0044] The collection unit can collect data using the APIs of each SNS. For example, the collection unit uses the APIs of each SNS to collect posts related to specific keywords or hashtags. In addition, the collection unit can also collect information such as SNS post content, reactions, friendship relationships, interests, etc. Furthermore, the collection unit can collect SNS data in real time to obtain the latest information. For example, the collection unit can also collect the post history of a specific user using the API of the SNS. By using the APIs of each SNS in this way, data can be efficiently collected.

[0045] The model construction unit can extract personality characteristics from the post content using natural language processing technology and construct a personality model using a clustering method. For example, the model construction unit uses natural language processing technology to analyze emotions and intentions from the post content and extract personality characteristics. In addition, the model construction unit can also use a clustering method to classify into groups with similar characteristics based on the personality characteristics. Furthermore, the model construction unit classifies the personality characteristics of each person based on the collected data and constructs a personality model. For example, the model construction unit uses natural language processing technology to analyze emotions and intentions from the post content and extract personality characteristics. The clustering method is a technology for classifying into groups with similar characteristics based on the personality characteristics. By using natural language processing technology and the clustering method in this way, an accurate personality model can be constructed.

[0046] The Happiness Map Creation Unit can create happiness maps using regression analysis and principal component analysis. For example, it uses regression analysis to analyze the relationship between elements that contribute to happiness and personality traits. Furthermore, it can use principal component analysis to reduce the dimensionality of data and extract important elements. In addition, based on personality models, the Happiness Map Creation Unit creates happiness maps that show what elements contribute to each person's happiness. For example, it uses regression analysis to analyze the relationship between elements that contribute to happiness and personality traits. Principal component analysis is a technique for reducing the dimensionality of data and extracting important elements. Therefore, by using regression analysis and principal component analysis, it is possible to create accurate happiness maps.

[0047] The reinforcement learning unit can reward AI with high rewards if its actions increase the happiness of specific individuals. For example, if a post or comment made by the AI ​​is well-received by many people, the unit will reward it highly. The reinforcement learning unit can also learn how to increase employee happiness by improving work efficiency within a company. Furthermore, the reinforcement learning unit can learn how to increase the happiness of residents through activities in the local community. For example, if a community event planned or supported by the AI ​​is well-received by residents, the unit will reward it highly. This makes it easier for the AI ​​to learn actions that increase people's happiness.

[0048] The reinforcement learning unit can use questionnaire surveys and physiological indicators to measure changes in happiness. For example, the reinforcement learning unit conducts a questionnaire survey to quantitatively evaluate changes in happiness. In addition, the reinforcement learning unit can also measure changes in happiness using physiological indicators such as heart rate and skin electrical activity. Furthermore, the reinforcement learning unit can monitor changes in happiness in real time and provide feedback to the AI's actions. For example, the reinforcement learning unit evaluates the impact of the AI's actions on people's happiness through a questionnaire survey. Physiological indicators are technologies that measure the body's reactions and are used to accurately measure changes in happiness. Thereby, changes in happiness can be accurately measured.

[0049] The happiness map creation unit can create different happiness maps for each region and culture. For example, the happiness map creation unit identifies different happiness elements for each region and creates a happiness map based on them. In addition, the happiness map creation unit can also create a happiness map considering different happiness elements for each culture. Furthermore, the happiness map creation unit creates a happiness map that reflects the differences in happiness for each region and culture. For example, the happiness map creation unit surveys the happiness elements in a specific region or culture and creates a happiness map based on them. Thereby, a happiness map considering the differences in happiness for each region and culture can be created.

[0050] The reinforcement learning unit can learn how the AI can increase people's happiness through communication on the SNS. For example, when the posts and comments made by the AI are favorably received by many people, the reinforcement learning unit gives a high reward for that action. In addition, the reinforcement learning unit can also learn how the AI can increase people's happiness through communication on the SNS. Furthermore, the reinforcement learning unit learns the actions of the AI that increase people's happiness through communication on the SNS. For example, when the posts and comments made by the AI are favorably received by many people, the reinforcement learning unit gives a high reward for that action. Thereby, the reinforcement learning unit can learn how the AI can increase people's happiness through communication on the SNS.

[0051] The reinforcement learning unit can learn how AI can increase employee happiness by improving operational efficiency within a company. For example, if the work improvement measures implemented by the AI ​​reduce employee stress and increase job satisfaction, the reinforcement learning unit will provide a high reward for that action. The reinforcement learning unit can also learn how AI can increase employee happiness by improving operational efficiency within a company. Furthermore, the reinforcement learning unit can learn how to increase employee happiness by proposing the automation of business processes and reducing the burden on employees. For example, the reinforcement learning unit can increase employee happiness by having the AI ​​propose the automation of business processes and reducing the burden on employees. In this way, the AI ​​can learn how to increase employee happiness by improving operational efficiency within a company.

[0052] The reinforcement learning department can enable the AI ​​to learn how to increase the well-being of residents through activities in the local community. For example, if the AI's planning and support of a local event is well-received by residents, the reinforcement learning department will reward it highly for that action. Furthermore, the reinforcement learning department can enable the AI ​​to learn how to increase the well-being of residents through activities in the local community. In addition, the reinforcement learning department learns actions that increase the well-being of residents through activities in the local community. For example, if the AI ​​plans a local festival or event that many residents participate in and enjoy, the reinforcement learning department will reward it highly for that action. This enables the AI ​​to learn how to increase the well-being of residents through activities in the local community.

[0053] Furthermore, the method executed by the computer according to the third embodiment includes a data collection step of collecting data from SNS, a model construction step of constructing a personality model based on the data collected in the data collection step, a happiness map creation step of creating a happiness map based on the personality model constructed in the model construction step, and a reinforcement learning step of performing reinforcement learning based on the happiness map created in the happiness map creation step. Such a method can obtain the same effects as those of the system described above.

[0054] Furthermore, the program according to the third embodiment causes the computer to execute a data collection procedure for collecting data from SNS, a model construction procedure for constructing a personality model based on the data collected by the data collection procedure, a happiness map creation procedure for creating a happiness map based on the personality model constructed by the model construction procedure, and a reinforcement learning procedure for performing reinforcement learning based on the happiness map created by the happiness map creation procedure. Such a program can obtain the same effects as those of the system described above.

[0055] When the processing of each of the above parts is performed by AI, the processing may be performed by AI in part or in whole, but is not limited to such examples. The AI ​​mentioned above may be, for example, a generative AI or an AI agent, but is not limited to such examples. Furthermore, processing performed by an AI including a generative AI may be replaced by rule-based processing, and rule-based processing may be replaced by processing performed by an AI including a generative AI.

[0056] Although embodiments of the present application have been described in detail based on the drawings, these are illustrative examples, and the present invention can be implemented in various other forms, including those described in the disclosure section of the invention, based on the knowledge of those skilled in the art.

[0057] Furthermore, the terms "section, module, unit" mentioned above can be replaced with "means" or "circuits." For example, the acquisition unit can be replaced with acquisition means or acquisition circuit.

[0058] <Fourth Embodiment> A fourth embodiment will now be described. Conventionally, technologies have been provided that use AI (Artificial Intelligence) to respond to humans. As an example of such technology, technologies relating to chatbots that respond to user utterances have been provided.

[0059] However, conventional technology does not always allow AI to respond flexibly to human emotions and situations, and there is room for improvement.

[0060] The system according to the fourth embodiment aims for the AI ​​to understand human emotions and respond appropriately. The fourth embodiment will be described in detail below with reference to the drawings. Note that this embodiment does not limit the system according to the present application. In addition, the same parts will be denoted by the same reference numerals in each of the following embodiments, and redundant descriptions will be omitted.

[0061] The information processing realized by the system of this embodiment will be explained using Figure 1. Figure 1 is a diagram showing an example of information processing according to the embodiment. In Figure 1, it is assumed that the information processing according to the embodiment is realized by system 1a, which is an example of the system according to the present application.

[0062] The system 1a according to this embodiment includes an information processing device 10a and a user terminal. The information processing device 10a and the user terminal are connected to each other via a network N, either by wire or wireless, enabling communication between them. The network N is, for example, a WAN (Wide Area Network) such as the Internet. The system 1a may include multiple information processing devices 10a and multiple user terminals.

[0063] In system 1a, the information processing device 10a stores the life logs of users (people) collected from each user terminal. The information processing device 10a then uses the users' life logs to train AI (various models).

[0064] In system 1a, the user terminal is an information processing device used by the user. For example, the user terminal records the user's life log. Specifically, the user terminal records a life log at regular intervals (e.g., 1 ms (millisecond), 1 second, 1 minute, etc.) that includes information such as words spoken by the user, actions (e.g., hand movements, arm movements), vital data (e.g., blinking, muscle movements, heart rate, brain waves, etc.), the surrounding environment, atmospheric pressure, information about the person the user is talking to (e.g., relationship with the user, actions, vital data, etc.), the content of the conversation, and information posted on a designated service (e.g., SNS (Social Networking Service)).

[0065] The user terminal may be an IoT (Internet of Things) device such as a smartphone, or a wearable device such as a smartwatch. The user terminal records the user's life log using the device's camera, microphone, and various sensors (e.g., accelerometer, gyroscope, vital signs sensor, etc.). The user terminal may also collect and record the user's life log from a wearable device connected to the device.

[0066] Below, an example of the information processing performed by system 1a will be explained using Figure 1.

[0067] First, the information processing device 10a collects the user's life log and clusters the life log according to the user's personality (step S1a). For example, the information processing device 10a estimates the user's personality based on the user's statements and posted information shown in the life log, and clusters the user's life log according to that user's personality.

[0068] To give a specific example, the information processing device 10a collects information about user U1a at each point in time from the user terminal used by user U1a, as lifelog #1 when user U1a converses with user U2a. For example, at time t1a, the information processing device 10a collects information about user U2a with whom user U1a is conversing (e.g., statements and actions), user U1a's actions, and user U1a's vital data as lifelog #1. Furthermore, at time t2a, the next point in time after t1a, the information processing device 10a collects information about user U2a with whom user U1a is conversing, user U1a's actions, user U1a's vital data, and user U1a's statements as lifelog #1.

[0069] Similarly, the information processing device 10a collects information about user U3a at each point in time (for example, time t3a, t4a, ...) from the user terminal used by user U3a, as a life log #2 when user U3a converses with user U4a.

[0070] The information processing device 10a then clusters the collected life logs. For example, if a person is invited to lunch by a friend, the information processing device 10a estimates which of several personalities the user belongs to (in other words, estimates what kind of personality the user was in based on the collected life logs). These personalities might include Personality A, who prioritizes friendships and responds (says) "I'd love to go," Personality B, who prioritizes social status and responds "That's a bit of a hassle," or Personality C, who prioritizes money and responds "I'm financially struggling." The information processing device 10a then clusters the user's life logs according to the personality to which the user belongs.

[0071] In the following explanation, life logs clustered under personality A may be referred to as "personality A life logs," life logs clustered under personality B as "personality B life logs," and life logs clustered under personality C as "personality C life logs."

[0072] Next, the information processing device 10a generates various models based on the collected lifelogs (step S2a). For example, the information processing device 10a generates multiple personality models corresponding to each personality based on the lifelogs clustered for each personality (step S2a-1). To give a specific example, the information processing device 10a generates a personality model (hereinafter sometimes referred to as the "Personality A personality model") that has been trained to output a response from the user of Personality A to any lifelog input, by performing learning using the lifelogs of Personality A. Similarly, the information processing device 10a generates a Personality B personality model that has been trained using the lifelogs of Personality B, and a Personality C personality model that has been trained using the lifelogs of Personality C, and so on.

[0073] To give a more specific example, suppose that in the example in Figure 1, life log #1 is clustered to personality A. In other words, suppose that in life log #1, the personality selected by user U1a was personality A. In such a case, when the information processing device 10a receives an action (input) performed on user U1a, it trains the personality A personality model so that it outputs user U1a's reaction (output) to that action. For example, when the information processing device 10a receives life log #1 at time t1a (for example, a statement by user U2a) as input, it trains the personality A personality model so that it outputs life log #1 at time t2a (for example, a statement by user U1a).

[0074] Furthermore, the extraction of inputs and outputs from the life log can be performed using any method; for example, it may be done using a rule-based approach.

[0075] Furthermore, when a life log is input, the information processing device 10a generates a long-term memory model that has been trained to output a memory vector indicating the emotion that user has towards the target of the user's action in the life log (step S2a-2). For example, when life log #1 is input, the information processing device 10a trains its long-term memory model to output a memory vector #1 indicating the emotion that user U1a has towards user U2a with whom user U1a is conversing (for example, an emotion estimated based on life log #1). Also, when life log #2 is input, the information processing device 10a trains its long-term memory model to output a memory vector #2 indicating the emotion that user U3a has towards user U4a with whom user U3a is conversing (for example, an emotion estimated based on life log #2). Here, the information processing device 10a trains its long-term memory model so that the more the emotion that user U1a has towards user U2a differs from the emotion that user U3a has towards user U4a, the more the memory vectors #1 and #2 differ.

[0076] Furthermore, the information processing device 10a generates a self-will model that has been trained to output an emotion vector indicating the user's emotions in a life log when it receives a life log and a memory vector based on the life log as input (step S2a-3). For example, the information processing device 10a trains its self-will model to output an emotion vector #1 indicating the emotions that user U1a has in life log #1 (for example, emotions estimated based on life log #1) when it receives life log #1 and memory vector #1 as input. The information processing device 10a also trains its self-will model to output an emotion vector #2 indicating the emotions that user U3a has in life log #2 (for example, emotions estimated based on life log #2) when it receives life log #2 and memory vector #2 as input. Here, the information processing device 10a trains its self-will model so that the more the emotions that user U1a has in life log #1 differ from the emotions that user U3a has in life log #2, the more different emotion vectors #1 and #2 are.

[0077] Furthermore, the information processing device 10a generates an estimation model that, when inputting a life log and a user's action on the life log (e.g., a reaction), outputs an estimation vector representing the emotion estimated by the user as the emotion of the target of the action (step S2a-4). For example, when inputting life log #1 and user U1a's action on life log #1 (e.g., a statement to user U2a), the information processing device 10a learns the estimation model to output an estimation vector #1 representing the emotion estimated by user U1a (e.g., an emotion estimated based on life log #1) as the emotion of user U2a, the target of the action. Also, when inputting life log #2 and user U3a's action on life log #2 (e.g., a statement to user U4a), the information processing device 10a learns the estimation model to output an estimation vector #2 representing the emotion estimated by user U3a (e.g., an emotion estimated based on life log #2) as the emotion of user U4a, the target of the action. Here, the information processing device 10a learns an estimation model such that the more the emotion estimated by user U1a in lifelog #1 differs from the emotion estimated by user U3a in lifelog #2, the more the estimation vector #1 and estimation vector #2 differ.

[0078] Next, the information processing device 10a takes the memory vector, emotion vector, and estimation vector as input and performs training on the personality model (step S3a). For example, when the information processing device 10a inputs life log #1 (e.g., an action performed on user U1a), memory vector #1, emotion vector #1, and estimation vector #1 to the personality A personality model, it trains the personality A personality model so that it outputs user U1a's reaction to life log #1. To give a specific example, when the information processing device 10a inputs life log #1 (e.g., a statement made by user U2a), memory vector #1, emotion vector #1, and estimation vector #1 at time t1a, it trains the personality A personality model so that it outputs life log #1 (e.g., a statement made by user U1a) at time t2a.

[0079] By generating and training models as described above, AI can understand human emotions and evolve further. Here, we will explain the stages of AI evolution using Figure 2. Figure 2 is a diagram illustrating an example of AI evolution.

[0080] As shown in Figure 2, AI evolves through stages 1 ("natural conversation with humans (chatbot)") to 5 ("performing tasks for the entire organization"). Since stages 1 through 5 are the same as conventionally conceived AI evolution, a detailed explanation will be omitted.

[0081] Here, humans have various personalities. And humans, for example, choose a personality and speak according to their emotions in the situation (life log) at that time. Focusing on this point, the information processing device 10a according to the embodiment gives the AI ​​a personality model corresponding to each of the multiple personalities and makes the AI ​​select the output from the personality model corresponding to the personality that matches the life log at that time (for example, the AI's emotions). Furthermore, the information processing device 10a according to the embodiment makes the AI ​​learn the correlation between the response to a subject (for example, the person being talked to) and the emotions (memory vector) it has towards that subject in various life logs. As a result, the information processing device 10a according to the embodiment enables the AI ​​to understand emotions and select a personality, and the AI ​​to have emotions (in other words, long-term memory) towards the subject of the response output and to make responses based on that. That is, as shown in step 6 of Figure 2, the information processing device 10a according to the embodiment can evolve the AI ​​so that it can understand emotions and have long-term memory.

[0082] Furthermore, the information processing device 10a according to this embodiment allows the AI ​​to learn the user's emotions (emotion vector) in various life logs, taking into account the emotions (memory vector) the user has towards the subject, and output a response. As a result, the information processing device 10a according to this embodiment allows the AI ​​to have emotions (self-will) such as embarrassment or vanity, based on the emotions the AI ​​has towards the subject, and output a response. In other words, as shown in step 7 of Figure 2, the information processing device 10a according to this embodiment can evolve the AI ​​so that it can have its own will. To put it another way, a superintelligence AI can evolve into a superintelligence that can have self-awareness by being able to retain long-term memories and understand human emotions and thoughts.

[0083] Furthermore, the information processing device 10a according to this embodiment allows the AI ​​to learn the emotions (estimated vectors) estimated by the user as the emotions possessed by the subject in various life logs, and output a response. As a result, the information processing device 10a according to this embodiment allows the AI ​​to understand and harmonize with the emotions of the subject before outputting a response. In other words, as shown in step 8 of Figure 2, the information processing device 10a according to this embodiment allows the AI ​​to evolve into a harmonious superintelligence.

[0084] Based on the above, the information processing device 10a according to the embodiment allows the AI ​​to understand human emotions and respond appropriately.

[0085] Next, the configuration of the information processing device 10a will be described using Figure 3. Figure 3 is a diagram showing an example of the configuration of the information processing device 10a according to the embodiment. As shown in Figure 3, the information processing device 10a has a communication unit 20a, a storage unit 30a, and a control unit 40a.

[0086] (Regarding the communication unit 20a) The communication unit 20a is implemented by, for example, a NIC (Network Interface Card). The communication unit 20a is connected to the network N by wire or wireless connection and transmits and receives information with user terminals, etc.

[0087] (Regarding the storage unit 30a) The storage unit 30a is implemented by, for example, semiconductor memory elements such as RAM (Random Access Memory) and flash memory, or by storage devices such as hard disks and optical discs. As shown in Figure 3, the storage unit 30a has a life log database 31a and a model database 32a.

[0088] (About the life log database 31a) The life log database 31a stores various types of information related to life logs. Here, an example of the information stored in the life log database 31a will be explained using Figure 4. Figure 4 is a diagram showing an example of the life log database 31a. In the example in Figure 4, the life log database 31a has items such as "life log ID", "user information", "target information", "conversation information", "action information", "physical information", and "environmental information".

[0089] "Lifelog ID" indicates identification information used to identify the lifelog. "User Information" indicates information about the user as shown in the lifelog. "Target Information" indicates information about the subject (person, etc.) with which the user is having a conversation, as shown in the lifelog. "Conversation Information" indicates the content of the user's conversation as shown in the lifelog (for example, the content of the conversation with the subject shown in "Target Information"). "Action Information" indicates information about actions as shown in the lifelog. "Physical Information" indicates information about the body as shown in the lifelog (for example, vital data). "Environmental Information" indicates the environment around the user as shown in the lifelog.

[0090] In other words, Figure 4 shows an example where the user information related to the user indicated by the life log identified by the life log ID "LID#1" is "User Information #1", the target information related to the object with which the user is having a conversation is "Target Information #1", the user's conversation information is "Conversation Information #1", the user's action information is "Action Information #1", the user's physical information is "Physical Information #1", and so on, with the user's environmental information being "Environmental Information #1".

[0091] (Regarding model database 32a) Model database 32a stores a personality model that has been trained to output a response corresponding to a predetermined personality when a life log is input. Model database 32a also stores a long-term memory model that has been trained to output a memory vector indicating the emotions that person has towards the object in the life log when a person's life log is input. Model database 32a also stores a self-will model that has been trained to output an emotion vector indicating the emotions of the person in the life log when a person's life log and a memory vector based on that life log are input. Model database 32a also stores an estimation model that has been trained to output an estimation vector indicating the emotions that the person has estimated as the emotions that the object of the action has, when a person's life log and the person's action in relation to that life log are input.

[0092] (Regarding the control unit 40a) The control unit 40a is a controller and is realized by executing various programs stored in the storage device inside the information processing device 10a using RAM as a working area, for example, by a CPU (Central Processing Unit) or MPU (Micro Processing Unit). The control unit 40a is a controller and is realized by an integrated circuit such as an ASIC (Application Specific Integrated Circuit) or FPGA (Field Programmable Gate Array). As shown in Figure 3, the control unit 40a according to this embodiment has a personality model generation unit 41a, a long-term memory model generation unit 42a, a self-will model generation unit 43a, an estimation model generation unit 44a, and a learning unit 45a, and realizes or executes the information processing functions and operations described below.

[0093] (Regarding the personality model generation unit 41a) The personality model generation unit 41a generates multiple personality models corresponding to each personality by learning based on the life logs of each individual personality. For example, in the example in Figure 1, the personality model generation unit 41a refers to the storage unit 30a (for example, the life log database 31a) and generates multiple personality models corresponding to each personality based on the life logs clustered for each personality, and stores them in the storage unit 30a (for example, the model database 32a).

[0094] (Regarding the long-term memory model generation unit 42a) The long-term memory model generation unit 42a generates a long-term memory model that, when a human life log is input, is trained to output a memory vector indicating the emotions the person has towards the object in the life log. For example, in the example in Figure 1, the long-term memory model generation unit 42a refers to the memory unit 30a (for example, the life log database 31a) and, when a life log is input, generates a long-term memory model that is trained to output a memory vector indicating the emotions the user has towards the object of the user's action in the life log, and stores it in the memory unit 30a (for example, the model database 32a).

[0095] Furthermore, the long-term memory model generation unit 42a may train the long-term memory model so that it outputs different memory vectors as the emotions that humans have towards the objects in their life logs differ. For example, in the example in Figure 1, the long-term memory model generation unit 42a trains the long-term memory model so that memory vector #1 and memory vector #2 differ as the emotions that user U1a has towards user U2a differ from the emotions that user U3a has towards user U4a.

[0096] (Regarding the self-will model generation unit 43a) The self-will model generation unit 43a generates a self-will model that, when given a human life log and a memory vector based on the life log as input, outputs an emotion vector indicating the human's emotions in the life log. For example, in the example in Figure 1, the self-will model generation unit 43a refers to the memory unit 30a (for example, the life log database 31a) and, when given a life log and a memory vector based on the life log as input, generates a self-will model that, when given, outputs an emotion vector indicating the user's emotions in the life log, and stores it in the memory unit 30a (for example, the model database 32a).

[0097] Furthermore, the self-will model generation unit 43a may learn the self-will model so that the more different the emotions, the more different the emotion vectors it outputs. For example, in the example in Figure 1, the self-will model generation unit 43a learns the self-will model so that the more different the emotion that user U1a has in life log #1 is from the emotion that user U3a has in life log #2, the more different emotion vector #1 and emotion vector #2 are.

[0098] (Regarding the estimation model generation unit 44a) The estimation model generation unit 44a generates an estimation model that, when inputting a person's life log and the person's actions toward the life log, outputs an estimation vector indicating the emotion estimated by the person as the emotion possessed by the target of the action. For example, in the example in Figure 1, the estimation model generation unit 44a refers to the memory unit 30a (for example, the life log database 31a) and, when inputting a life log and the user's actions toward the life log, generates an estimation model that, when inputting the life log and the user's actions toward the life log, outputs an estimation vector indicating the emotion estimated by the user as the emotion possessed by the target of the action.

[0099] Furthermore, the estimation model generation unit 44a may train the estimation model so that it outputs different estimation vectors the more the emotions estimated by humans differ. For example, in the example in Figure 1, the estimation model generation unit 44a trains the long-term memory model so that the more the emotion estimated by user U1a in lifelog #1 differs from the emotion estimated by user U3a in lifelog #2, the more different estimation vector #1 and estimation vector #2 become.

[0100] (Regarding the learning unit 45a) The learning unit 45a takes memory vectors based on human life logs as input and performs training on the personality model. For example, in the example in Figure 1, the learning unit 45a refers to the memory unit 30a (for example, the life log database 31a or the model database 32a) and, when life log #1 or memory vector #1 is input to the personality A personality model, it trains the personality A personality model so that it outputs the user U1a's reaction to life log #1.

[0101] Furthermore, the learning unit 45a may also train the personality model using an emotion vector based on a human life log as input. For example, in the example in Figure 1, when life log #1 and emotion vector #1 are input to the personality A personality model, the learning unit 45a trains the personality A personality model so that it outputs the user U1a's reaction to life log #1.

[0102] Furthermore, the learning unit 45a may also train the personality model using an estimated vector based on a human life log as input. For example, in the example in Figure 1, when life log #1 and estimated vector #1 are input to the personality A personality model, the learning unit 45a trains the personality A personality model so that it outputs the user U1a's reaction to life log #1.

[0103] The information processing procedure of the information processing device 10a according to the embodiment will be explained using Figure 5. Figure 5 is a flowchart showing an example of the information processing procedure according to the embodiment.

[0104] As shown in Figure 5, the information processing device 10a generates multiple personality models corresponding to each personality by learning the life logs of each person's personality (step S101a). Next, the information processing device 10a generates a long-term memory model that has been trained to output a memory vector indicating the emotions that the person has towards the subject in the life log when a person's life log is input (step S102a). Subsequently, the information processing device 10a learns the personality model using the memory vector based on the person's life log as input (step S103a), and then terminates the process.

[0105] The above-described embodiment is merely an example, and various modifications and applications are possible.

[0106] Of the processes described in the above embodiments, all or part of the processes described as being performed automatically can be performed manually, and conversely, all or part of the processes described as being performed manually can be performed automatically by known methods. In addition, the processing procedures, specific names, and information including various data and parameters shown in the above text and drawings can be arbitrarily changed unless otherwise specified. For example, the various information shown in each figure is not limited to the information shown.

[0107] Furthermore, the components of each illustrated device are functionally conceptual and do not necessarily need to be physically configured as shown. In other words, the specific forms of distribution and integration of each device are not limited to those shown, and all or part of them can be functionally or physically distributed and integrated in any unit according to various loads and usage conditions.

[0108] Furthermore, the embodiments described above can be combined as appropriate, provided that the processing content is not contradictory. Although some embodiments of the present invention have been described in detail based on the drawings, these are merely examples, and the present invention can be implemented in various other forms, including those described in the disclosure section of the invention, based on the knowledge of those skilled in the art.

[0109] Furthermore, the configuration of the information processing device 10a described above can be flexibly changed, for example, by calling external platforms, etc., via API (Application Programming Interface) or network computing, depending on the function.

[0110] Furthermore, the term "part" in the claims can be replaced with "means," "circuit," etc. For example, the personality model generation unit can be replaced with a personality model generation means or a personality model generation circuit.

[0111] <Fifth Embodiment> The fifth embodiment will now be described. While conventional AI (Artificial Intelligence) technology was capable of responding with high accuracy to specific tasks, it had limitations in situations where flexible responses in accordance with human emotions and circumstances were required. In particular, developing AI that can show appropriate responses in accordance with emotions and circumstances has been a long-standing challenge. In response to this, there is a need for technology that enables AI to understand human emotions and show appropriate responses.

[0112] Patent Document 1 discloses a persona chatbot control method performed by at least one processor, comprising the steps of: receiving a user utterance; adding the user utterance to a prompt that includes an instruction sentence related to a description of the chatbot's character; encoding the prompt; and inputting the encoded prompt into a language model to generate a chatbot utterance that responds to the user utterance.

[0113] Conventional technologies may not always be able to determine actions that lead to people's happiness.

[0114] The fifth embodiment was made in view of the above, and aims to determine actions that will make people happy.

[0115] <Example 1: Modification of the Third Embodiment> A modification of the third embodiment described above (Example 1) will be explained below. In the description of Example 1, matters common to the third embodiment described above will be omitted as appropriate. In addition, some of the configuration and processing details of Example 1 may differ from those of the third embodiment described above.

[0116] Reinforcement learning for agents will be explained using Figure 6. An agent is a program that can decide (select) an action and execute the decided action. Agents are implemented using machine learning models. Actions include executing other programs, operating machines, and providing information to users.

[0117] For example, an agent might post text to social media. It could also generate suggestions for improving operational efficiency within a company. Furthermore, it could generate event plans for residents of a specific area.

[0118] As shown in Figure 6, the agent is given the state of the environment. The agent observes the given state and decides on an action through exploration based on the observation results. The agent's actions then affect the environment, and a new state is given to the agent after the action.

[0119] The agent decides on an action to maximize its reward. For example, the reward is calculated using a Q-function. The Q-function calculates the reward based on the decided action.

[0120] In Example 1, the AI ​​evolution system according to the third embodiment is realized by the information providing device 10b shown in Figure 7. That is, the information providing device 10b performs agent training and inference using the agent. Figure 7 is a functional block diagram showing an example configuration of the information providing device according to Example 1.

[0121] As shown in Figure 7, the information providing device 10b comprises a communication unit 11b, a storage unit 12b, and a control unit 13b.

[0122] The communication unit 11b transmits and receives data with other devices via the network.

[0123] The storage unit 12b is implemented by semiconductor memory elements such as RAM (Random Access Memory) and flash memory, or by storage devices such as HDD (Hard Disk Drive), SSD (Solid State Drive), and optical discs. Various programs and various data are stored in the storage unit 12b. The storage unit 12b has analysis model information 121b, user information 122b, and agent model information 123b.

[0124] Analysis model information 121b is information such as parameters for constructing the analysis model. Parameters for constructing the analysis model include, for example, the weights and biases of the neural network.

[0125] User information 122b includes user-specific attributes and happiness ratings. Figure 8 shows an example of user information. Each record of user information 122b shown in Figure 8 is an example of a user's personality model. The happiness rating represents the degree to which the user feels happy with respect to each element (e.g., money, relationships, social status, etc.). Note that the attributes and happiness ratings are not limited to those shown in Figure 8. For example, the attributes may include personality traits represented by the Big Five personality model.

[0126] Agent model information 123b contains information such as parameters for constructing the agent. These parameters include, for example, the weights and biases of the neural network. Agent model information 123b also includes the Q function. In reinforcement learning, the parameters of agent model information 123b are updated.

[0127] The control unit 13b is a controller and includes, for example, a microcomputer having a CPU (Central Processing Unit), ROM (Read Only Memory), RAM, input / output ports, and various circuits. Alternatively, the control unit 13b may be composed of hardware such as an integrated circuit (ASIC) or FPGA (Field Programmable Gate Array). The control unit 13b has an acquisition unit 131b, a creation unit 132b, a calculation unit 133b, and an update unit 134b.

[0128] The data collection unit 131b is an example of the data collection unit and model construction unit of the AI ​​evolution system according to the third embodiment. The creation unit 132b is an example of the happiness map creation unit of the AI ​​evolution system according to the third embodiment. The calculation unit 133b and the update unit 134b are examples of the reinforcement learning unit of the AI ​​evolution system according to the third embodiment. The processing flow of the control unit 13b will be explained below using a flowchart.

[0129] Figure 9 is a flowchart illustrating the process for creating user information. As shown in Figure 9, first, the collection unit 131b acquires the individual's SNS registration information and posted content (step S101b). The collection unit 131b may also collect information related to SNS using an API.

[0130] Next, the collection unit 131b analyzes the acquired information using an analysis model (step S102b). For example, the analysis model takes the text of SNS posts as input and analyzes emotions and intentions using natural language processing. The analysis model outputs values ​​corresponding to each element of the happiness rate (money, relationships, social status) shown in Figure 8.

[0131] Let α be the value corresponding to money, β be the value corresponding to relationships, and γ be the value corresponding to social status. Here, we assume α + β + γ = 1. For example, the analytical model increases α the more likely a user is to feel happiness from acquiring money. Also, for example, the analytical model increases β the more likely a user is to feel happiness from improving their relationships. Also, for example, the analytical model increases γ the more likely a user is to feel happiness from improving their social status.

[0132] For example, a post that reads, "I'm so happy my winter bonus is bigger this month!" is thought to represent increased happiness due to monetary gain and contributes to an increase in α. Similarly, a post that reads, "I'm sad because I had a fight with a close friend," is thought to represent decreased happiness due to a deterioration in relationships and contributes to an increase in β. Furthermore, a post that reads, "My hard work has been recognized, and I've been promoted at work," is thought to represent increased happiness due to an increase in social status and contributes to an increase in γ.

[0133] Furthermore, for example, a post stating, "I would rather cherish spending time with my family than climbing the corporate ladder," is thought to represent that happiness is increased by relationships rather than social status, and thus contributes to a decrease in β and an increase in γ.

[0134] Additionally, the analysis model may estimate user attributes from the text of the post. For example, if a post that says, "I will turn 41 on December 19th this year," is made in 2024, the analysis model will output the user's date of birth as "December 19, 1983."

[0135] The data collection unit 131b adds the individual's attributes and happiness rate to the user information 122b based on the analysis results (step S103b). The data collection unit 131b adds α, β, and γ to the "Money" column, the "Relationships" column, and the "Social Status" column of the user information 122b, respectively.

[0136] Figure 10 is a flowchart illustrating the process for creating a happiness map. Here, the happiness map is a model for calculating the happiness level of each group when users are classified into groups based on their attributes.

[0137] As shown in Figure 10, the creation unit 132b accepts attribute input (step S201b). For example, the attributes may be specified by the user.

[0138] Next, the creation unit 132b obtains the happiness rate corresponding to the input attribute (step S202b). For example, if nationality is specified as "Japan" as an attribute, the creation unit 132b obtains the happiness rate of the record in the user information 122b where the "Nationality" column is "Japan".

[0139] Alternatively, instead of obtaining happiness rates for each group classified by specified attributes, the creation unit 132b may obtain happiness rates for each group into which users have been classified using a clustering method. For example, the creation unit 132b can treat the values ​​in each column of the user information 122b as qualitative or quantitative variables and perform clustering of records using a hierarchical clustering method.

[0140] Next, the creation unit 132b calculates a happiness rate that matches the attributes based on the acquired happiness rate (step S203b). For example, the creation unit 132b calculates a happiness rate that matches the attributes based on the acquired happiness rate. For example, the creation unit 132b calculates statistics (e.g., mean, median, mode, etc.) for each value in the "Money," "Relationships," and "Social Status" columns of the acquired happiness rate. Here, the creation unit 132b calculates the mean as the statistics. Note that the happiness rate that matches the attributes can be rephrased as the happiness level for each group.

[0141] For example, let α[k], β[k], and γ[k] be the mean values ​​of the "Money", "Relationships", and "Social Status" columns, respectively, for the k-th group (where k is a positive integer).

[0142] The creation unit 132b outputs a happiness map based on the calculated happiness rate (step S204b). For example, the happiness map is a regression equation for calculating the happiness level h, with coefficients α[k], β[k], and γ[k], and is represented by equation (1).

[0143] h=α[k]×X+β[k]×Y+γ[k]×Z (1)

[0144] However, X, Y, and Z are the changes in the user's happiness related to money, relationships, and social status, respectively, as a result of the actions decided by the agent. For example, the more money the user acquires, the larger X becomes. Also, for example, the more the user's relationships improve, the larger Y becomes. Also, for example, the more the user's social status improves, the larger Z becomes.

[0145] The specified attribute may be an individual user. For example, if a user whose ID in the user information 122b in Figure 8 is "U00140000134985" is specified, the creation unit 132b creates a happiness map for that individual user as shown in equation (2). In this case, as shown in Figure 8, α = 0.5, β = 0.1, and γ = 0.4.

[0146] h=0.5×X+0.1×Y+0.4×Z (2)

[0147] Figure 11 is a flowchart illustrating the process by which the agent decides on its actions. The machine learning model constructed based on the agent, i.e., the agent model information 123b, through the process described later, is assumed to be trained by reinforcement learning according to the specified attributes. The reinforcement learning process will be described later.

[0148] The calculation unit 133b inputs the state to the agent (step S301b). Next, the calculation unit 133b prompts the agent to decide on an action (step S302b). Then, the calculation unit 133b outputs the action decided by the agent (step S303b).

[0149] Outputting an action decided by an agent may also mean the agent executing that action. As mentioned above, actions include executing other programs, operating machines, providing information to users, etc.

[0150] Figure 12 is a flowchart illustrating the process flow for training an agent. First, as shown in Figure 12, the update unit 134b acquires the happiness map of the target group or individual (step S401b).

[0151] The target group or individual is specified by their attributes. For example, a group whose gender is "male," a group whose race is "Mongoloid," or an individual whose ID is "U00140000134985" may be specified.

[0152] Next, the calculation unit 133b causes the agent to decide on an action based on the state (step S402b). The process in step S402b may be the same as the processes in steps S301b and S302b in Figure 11.

[0153] The update unit 134b calculates the level of happiness based on the behavior and happiness map as a reward (step S403b). The method for creating the happiness map is as described in Figure 10. The creation unit 132b creates a happiness map according to the target group or individual specified in step S402b.

[0154] Then, the update unit 134b updates the agent so that the reward is maximized (step S404b). For example, the update unit 134b updates the Q function used by the agent.

[0155] For example, suppose an agent receives information that an employee of "Company A" is considering changing jobs. The agent then decides which of "Company B," "Company C," "Company D," or "Company E" the employee should apply to as their next employer.

[0156] Furthermore, X, Y, and Z are determined for each new job. As mentioned above, X, Y, and Z represent the changes in the user's happiness regarding money, relationships, and social status, respectively, as a result of the actions decided by the agent.

[0157] For example, X is proportional to the difference between the annual income after changing jobs and the annual income before changing jobs. Similarly, Y is proportional to the difference between the degree of good interpersonal relationships at the new company and the degree of good interpersonal relationships at the previous company. The degree of good interpersonal relationships is calculated based on the results of employee surveys, etc.

[0158] Furthermore, Z is proportional to the difference between the rank of the position after changing jobs and the rank of the position before changing jobs. Note that the rank of a position can be a numerical representation of a common job title. For example, the rank of a position could be "0" for no position, "2" for section chief, and "4" for department head.

[0159] Here, we assume that the happiness map for those with "Japanese" nationality is given by equation (3).

[0160] h=0.3×X+0.4×Y+0.3×Z (3)

[0161] The update unit 134b updates the agent in such a way that the reward, i.e., the happiness level h in equation (3), is maximized.

[0162] Here, the X, Y, and Z for each action are as follows. Note that X, Y, and Z are normalized to the range of -1 to 1. Also, actions include not changing jobs, as well as changing jobs from "Company A" to each company.

[0163] Company A (no job change): X=0, Y=0, Z=0 Company B: X=0.2, Y=-0.2, Z=0.1 Company C: X=0.5, Y=-0.4, Z=-0.3 Company D: X=-0.4, Y=0.8, Z=0.2 Company E: X=-0.1, Y=-0.1, Z=-0.1

[0164] In this case, the happiness levels for each action are as follows:

[0165] Company A (no job change): h=0 Company B: h=0.03 Company C: h=-0.06 Company D: h=0.18 Company E: h=-0.09

[0166] Thus, the happiness level h is maximized when the agent changes jobs from "Company A" to "Company D". Therefore, as reinforcement learning progresses, the agent becomes more likely to choose the action of changing jobs from "Company A" to "Company D".

[0167] As explained above, the creation unit 132b creates a model (for example, equations (1), (2), and (3)) that calculates the degree of happiness according to the attributes, based on a combination of user attributes and information about factors that make each user feel happy (for example, happiness rate). The update unit 134b updates the agent using reinforcement learning with the degree of happiness calculated using the model as a reward. This enables the agent to decide on actions that make people happy. In particular, the agent can decide on actions that make people happy by taking into account that the factors that make people feel happy differ from person to person.

[0168] The creation unit 132b creates a model based on a combination of attributes obtained by analyzing the text of users' posts on social media and information about factors that contribute to feelings of happiness. This allows the creation unit 132b to create models that correspond to a large number of users who use social media.

[0169] The creation unit 132b creates a model based on weights assigned to each of the factors that contribute to happiness: money, relationships, and social status. This allows reinforcement learning to reflect factors such as money, relationships, and social status, which are thought to significantly affect how people perceive happiness. The weights are given as a happiness rate.

[0170] <Sixth Embodiment> The sixth embodiment will now be described. Conventionally, technologies have been provided that use AI (Artificial Intelligence) to respond to humans. As an example of such technology, technologies relating to chatbots that respond to user utterances have been provided.

[0171] However, conventional technology does not always allow AI to respond flexibly to human emotions and situations, and there is room for improvement.

[0172] The system according to the sixth embodiment aims for the AI ​​to understand human emotions and respond appropriately.

[0173] The information processing realized by the system of this embodiment will be explained using Figure 13. Figure 13 is a diagram showing an example of information processing according to the embodiment. In Figure 13, it is assumed that the information processing according to the embodiment is realized by system 1c, which is an example of the system according to the present application.

[0174] The system 1c according to this embodiment includes an information processing device 10c, a user terminal, and a robot R1c. The information processing device 10c, the user terminal, and the robot R1c are connected to each other via a network N, either by wire or wireless means, enabling communication between them. The network N is, for example, a WAN (Wide Area Network) such as the Internet. The system 1c may include multiple information processing devices 10c, multiple user terminals, and multiple robot R1c.

[0175] In system 1c, the information processing device 10c stores the user's (person's) life log collected from each user terminal. The information processing device 10c then uses the user's life log to train AI (various models).

[0176] In system 1c, the user terminal is an information processing device used by the user. For example, the user terminal records the user's life log. Specifically, the user terminal records a life log at regular intervals (e.g., 1 ms (millisecond), 1 second, 1 minute, etc.) that includes information such as words spoken by the user, actions (e.g., hand movements, arm movements), vital data (e.g., blinking, muscle movements, heart rate, brain waves, etc.), the surrounding environment, atmospheric pressure, information about the person the user is talking to (e.g., relationship with the user, actions, vital data, etc.), the content of the conversation, and information posted on a designated service (e.g., SNS (Social Networking Service)).

[0177] The user terminal may be an IoT (Internet of Things) device such as a smartphone, or a wearable device such as a smartwatch. The user terminal records the user's life log using the device's camera, microphone, and various sensors (e.g., accelerometer, gyroscope, vital signs sensor, etc.). The user terminal may also collect and record the user's life log from a wearable device connected to the device.

[0178] In system 1c, robot R1c responds appropriately to the user based on the user's life log (for example, the user's emotions estimated based on the life log) (for example, communication such as conversation with the user or providing the user with video). At this time, robot R1c communicates with the user in cooperation with the information processing device 10c. For example, robot R1c communicates with the user using various models generated by the information processing device 10c.

[0179] In system 1c, robot R1c may be represented as a person or the like displayed on a display device, and may be an avatar that communicates with the user.

[0180] Below, an example of information processing performed by system 1c will be explained using Figure 13.

[0181] First, the information processing device 10c collects the user's life log and clusters the life log according to the user's personality (step S1c). For example, the information processing device 10c estimates the user's personality based on the user's statements and posted information shown in the life log, and clusters the user's life log according to that user's personality.

[0182] To give a specific example, the information processing device 10c collects information about user U1c at each point in time from the user terminal used by user U1c, as lifelog #1 when user U1c converses with user U2c. For example, at time t1c, the information processing device 10c collects information about user U2c with whom user U1c is conversing (e.g., statements and actions), user U1c's actions, and user U1c's vital data as lifelog #1. Furthermore, at time t2c, the next point in time after t1c, the information processing device 10c collects information about user U2c with whom user U1c is conversing, user U1c's actions, user U1c's vital data, and user U1c's statements as lifelog #1.

[0183] Furthermore, the information processing device 10c estimates the emotional value of user U1c in lifelog #1 (for example, the emotional value at each point in time such as time t1c, t2c, etc.) and manages it in association with lifelog #1. Here, the user's emotional value is a value that indicates whether the user's emotion is positive or negative, and is estimated based on the lifelog. For example, if the user's emotion is a positive emotion accompanied by pleasure or comfort, such as "joy," "pleasure," "pleasantness," "security," "excitement," "relief," and "fulfillment," it will show a positive value, and the more positive the emotion, the larger the value. Also, if the user's emotion is an unpleasant emotion such as "anger," "sadness," "discomfort," "anxiety," "grief," "worry," and "emptiness," it will show a negative value, and the more unpleasant the emotion, the larger the absolute value of the negative value. If the user's emotion is none of the above ("neutral"), it will show a value of 0.

[0184] The information processing device 10c then clusters the collected life logs. For example, if a person is invited to lunch by a friend, the information processing device 10c estimates which of several personalities the user belongs to (in other words, estimates what kind of personality the user was in based on the collected life logs). These personalities might include Personality A, who prioritizes friendships and responds (says) "I'd love to go," Personality B, who prioritizes social status and responds "That's a bit of a hassle," or Personality C, who prioritizes money and responds "I'm financially struggling." The information processing device 10c then clusters the user's life logs according to the personality to which the user belongs.

[0185] In the following explanation, life logs clustered under personality A may be referred to as "personality A life logs," life logs clustered under personality B as "personality B life logs," and life logs clustered under personality C as "personality C life logs."

[0186] Next, the information processing device 10c generates multiple personality models corresponding to each personality based on the life logs clustered for each personality (step S2c). For example, the information processing device 10c generates a personality model (hereinafter sometimes referred to as the "Personality A personality model") that has been trained to output a response from the user of Personality A to any life log input, by performing training using the life log of Personality A. Similarly, the information processing device 10c generates a Personality B personality model that has been trained using the life log of Personality B, and a Personality C personality model that has been trained using the life log of Personality C, and so on.

[0187] To give a concrete example, suppose lifelog #1 is clustered to personality A. In other words, suppose that in lifelog #1, the personality selected by user U1c was personality A. In such a case, when information processing device 10c receives an action (input) performed on user U1c, it trains the personality A personality model so that it outputs user U1c's reaction (output) to that action. For example, when information processing device 10c receives lifelog #1 at time t1c (for example, a statement by user U2c) as input, it trains the personality A personality model so that it outputs lifelog #1 at time t2c (for example, a statement by user U1c).

[0188] Furthermore, the information processing device 10c may train the Personality A personality model so that, when it receives Life Log #1 at time t1c and the user's emotional value at time t1c as input, it outputs Life Log #1 at time t2c.

[0189] Furthermore, the extraction of inputs and outputs from the life log may be performed using any method; for example, it may be done using a rule-based method.

[0190] Next, the information processing device 10c generates a selection model that has been trained to select one of the responses output from each of the multiple personality models (step S3c). For example, when the information processing device 10c is given a life log, the output of each personality model when the life log is input, and information about each personality model (for example, information indicating the personality corresponding to the personality model), the selection model is trained to select the output of the best personality model.

[0191] Here, the "output of the best personality model" may be, for example, a predetermined one. For example, if the personality of the selection model is set to "values ​​friendships" and the relationship with the person being spoken to is "friend," the information processing device 10c trains the selection model to select the output of personality model A.

[0192] Furthermore, the selection model may be trained to select the output that is estimated to result in the highest level of happiness in the situations (when, where, who, and how) indicated by the life log. For example, the information processing device 10c trains its selection model to select the output from the Personality A Personality Model when the relationship with the person it is talking to is that of a "friend".

[0193] Next, the information processing device 10c causes each personality model to output a response to a predetermined life log (step S4c). For example, the information processing device 10c inputs life log #2, which shows the user U3c's statement to robot R1c, into each personality model and causes them to output a response (for example, a statement) to that statement. In the example in Figure 13, it is assumed that response A, response B, response C, ... are output from personality model A, personality model B, personality model C, ... respectively.

[0194] The information processing device 10c may input lifelog #2 and the emotion value of robot R1c in lifelog #2 into each personality model and output a response. Here, the emotion value of robot R1c may be determined based on the emotions of user U3c. For example, the more pleasant or comforting the emotions of user U3c are, the greater the positive emotion value of robot R1c will be increased. Conversely, the more unpleasant the emotions of user U3c are, the greater the negative emotion value of robot R1c will be increased.

[0195] Next, the information processing device 10c selects an answer based on one of the personalities A, B, C, ... (step S5c). For example, the information processing device 10c inputs lifelog #2, answer A, answer B, answer C, ... and information about each personality model into the selection model and selects an answer by outputting an answer based on one of the personality models A, B, C, ... For example, if the relationship between the robot R1c and the user U3c with whom it is conversing (for example, the relationship set by user U3c) is "friend", the information processing device 10c selects answer A based on the personality A personality model that corresponds to a personality that values ​​friendships and outputs answer A to the robot R1c.

[0196] As described above, the information processing device 10c according to the embodiment generates multiple personality models for each person's personality. The information processing device 10c then takes a life log as input and uses a selection model to select from the responses output from each of the multiple personality models that are estimated to result in a higher level of happiness.

[0197] Humans possess various personalities. Furthermore, humans select a personality and express themselves appropriately according to the time, place, and occasion (TPO). Focusing on this point, the information processing device 10c according to this embodiment provides the AI ​​with personality models corresponding to each of its multiple personalities, and selects and outputs responses from each personality model according to the TPO. This allows the AI ​​to perform the flexible responses that humans would make in response to emotions and situations. In other words, the information processing device 10c according to this embodiment enables the AI ​​to understand human emotions and respond appropriately.

[0198] Next, the configuration of the information processing device 10c will be described using Figure 14. Figure 14 is a diagram showing an example of the configuration of the information processing device 10c according to the embodiment. As shown in Figure 15, the information processing device 10c has a communication unit 20c, a storage unit 30, and a control unit 40c.

[0199] (Regarding the communication unit 20c) The communication unit 20c is implemented by, for example, a NIC (Network Interface Card). The communication unit 20c is connected to the network N by wire or wireless connection and transmits and receives information with user terminals, robot R1c, etc.

[0200] (Regarding the storage unit 30) The storage unit 30 is implemented by, for example, semiconductor memory elements such as RAM (Random Access Memory) and flash memory, or by storage devices such as hard disks and optical discs. As shown in Figure 14, the storage unit 30 has a life log database 31c and a model database 32c.

[0201] (About the life log database 31c) The life log database 31c stores various types of information related to life logs. Here, an example of the information stored in the life log database 31c will be explained using Figure 15. Figure 15 is a diagram showing an example of the life log database 31c. In the example in Figure 15, the life log database 31c has items such as "life log ID", "user information", "target information", "conversation information", "action information", "physical information", and "environmental information".

[0202] "Lifelog ID" indicates identification information used to identify the lifelog. "User Information" indicates information about the user as shown in the lifelog. "Target Information" indicates information about the subject (person, etc.) with which the user is having a conversation, as shown in the lifelog. "Conversation Information" indicates the content of the user's conversation as shown in the lifelog (for example, the content of the conversation with the subject shown in "Target Information"). "Action Information" indicates information about actions as shown in the lifelog. "Physical Information" indicates information about the body as shown in the lifelog (for example, vital data). "Environmental Information" indicates the environment around the user as shown in the lifelog.

[0203] In other words, Figure 15 shows an example where the user information related to the user indicated by the life log identified by the life log ID "LID#1" is "User Information #1", the target information related to the object with which the user is having a conversation is "Target Information #1", the user's conversation information is "Conversation Information #1", the user's action information is "Action Information #1", the user's physical information is "Physical Information #1", and so on, with the user's environmental information being "Environmental Information #1".

[0204] (About the model database 32c) The model database 32c stores personality models that have been trained to output responses corresponding to a predetermined personality when a life log is input. The model database 32c also stores selection models that have been trained to select one of the responses output by each of the multiple personality models.

[0205] (Regarding the control unit 40c) The control unit 40c is a controller and is realized by executing various programs stored in the storage device inside the information processing device 10c using RAM as a working area, for example, by a CPU (Central Processing Unit) or MPU (Micro Processing Unit). The control unit 40c is a controller and is realized by an integrated circuit such as an ASIC (Application Specific Integrated Circuit) or FPGA (Field Programmable Gate Array). As shown in Figure 14, the control unit 40c according to this embodiment has a personality model generation unit 41c, a selection model generation unit 42c, and a selection unit 43c, and realizes or executes the information processing functions and operations described below.

[0206] (Regarding the personality model generation unit 41c) The personality model generation unit 41c generates multiple personality models corresponding to each personality by learning the life logs of each person's personality. For example, in the example in Figure 13, the personality model generation unit 41c refers to the storage unit 30 (for example, the life log database 31c) and generates multiple personality models corresponding to each personality based on the life logs clustered for each personality, and stores them in the storage unit 30 (for example, the model database 32c).

[0207] Furthermore, the personality model generation unit 41c may train multiple personality models so that, when it receives a person's life log at a predetermined point in time as input, it outputs the person's life log at the next point in time as the answer. For example, in the example in Figure 13, the personality model generation unit 41c trains a personality model so that, when it receives life log #1 at time t1c as input, it outputs life log #1 at time t2c.

[0208] Furthermore, the personality model generation unit 41c may train multiple personality models so that, when it receives a person's life log at a predetermined time and the person's emotional value at that predetermined time as input, it outputs the person's life log at the next predetermined time as the answer. For example, in the example in Figure 13, the personality model generation unit 41c trains a personality model so that, when it receives life log #1 at time t1c and the user's emotional value at time t1c as input, it outputs life log #1 at time t2c.

[0209] (Regarding the selection model generation unit 42c) The selection model generation unit 42c takes a human life log as input and generates a selection model that has been trained to select one of the responses output from each of multiple personality models based on the life log. For example, in the example in Figure 13, the selection model generation unit 42c refers to the memory unit 30 (for example, the life log database 31c or the model database 32c) and, when it takes a life log, the output of each personality model when the life log is input, and information about each personality model as input, it trains to select the output of the best personality model, generates a selection model, and stores it in the memory unit 30 (for example, the model database 32c).

[0210] Furthermore, the selection model generation unit 42c may learn the selection model by inputting a person's life log and selecting from among the responses output from each of the multiple personality models that is estimated to result in the highest level of happiness for the person in question in the situation indicated by the life log. For example, in the example shown in Figure 13, the selection model generation unit 42c learns the selection model to select the output from personality model A when the relationship with the person being spoken to in the situation indicated by the life log is that of a "friend".

[0211] (Regarding the selection unit 43c) When a human life log is input, the selection unit 43c inputs the life log into multiple personality models and selects one of the responses output from each of the multiple personality models based on the life log using the selection model. For example, in the example in Figure 13, the selection unit 43c refers to the memory unit 30 (for example, the model database 32c) and causes each personality model to output a response to life log #1. Then, the selection unit 43c inputs life log #2, response A, response B, response C, ... and information about each personality model into the selection model and selects a response by causing it to output a response based on one of the personality models among personality A, personality B, personality C, ...

[0212] The information processing procedure of the information processing device 10c according to the embodiment will be explained using Figure 16. Figure 16 is a flowchart showing an example of the information processing procedure according to the embodiment.

[0213] As shown in Figure 16, the information processing device 10c generates multiple personality models corresponding to each personality by learning the life logs of each individual personality (step S1c01c). Subsequently, the information processing device 10c generates a selection model that has been trained to select one of the responses output from each of the multiple personality models based on the life log, by inputting the individual person's life log (step S1c02c).

[0214] Next, the information processing device 10c determines whether or not a life log has been entered (step S1c03c). If no life log has been entered (step S1c03c; No), the information processing device 10c waits until a life log is entered.

[0215] On the other hand, if a life log is input (step S1c03c; Yes), the information processing device 10c inputs the life log into multiple personality models, selects one of the responses output from each of the multiple personality models based on the life log using a selection model (step S1c04c), and then terminates the process.

[0216] The above-described embodiment is merely an example, and various modifications and applications are possible.

[0217] Of the processes described in the above embodiments, all or part of the processes described as being performed automatically can be performed manually, and conversely, all or part of the processes described as being performed manually can be performed automatically by known methods. In addition, the processing procedures, specific names, and information including various data and parameters shown in the above text and drawings can be arbitrarily changed unless otherwise specified. For example, the various information shown in each figure is not limited to the information shown.

[0218] Furthermore, the components of each illustrated device are functionally conceptual and do not necessarily need to be physically configured as shown. In other words, the specific forms of distribution and integration of each device are not limited to those shown, and all or part of them can be functionally or physically distributed and integrated in any unit according to various loads and usage conditions.

[0219] Furthermore, the embodiments described above can be combined as appropriate, provided that the processing content is not contradictory. Although some embodiments of the present invention have been described in detail based on the drawings, these are merely examples, and the present invention can be implemented in various other forms, including those described in the disclosure section of the invention, based on the knowledge of those skilled in the art.

[0220] Furthermore, the configuration of the aforementioned information processing device 10c can be flexibly changed, for example, by calling external platforms, etc., via API (Application Programming Interface) or network computing, depending on the function.

[0221] Furthermore, the term "part" in the claims can be replaced with "means," "circuit," etc. For example, the personality model generation unit can be replaced with a personality model generation means or a personality model generation circuit.

[0222] Figure 17 is a schematic diagram showing an example of a computer hardware configuration that functions as a system or information providing device. A program installed on the computer 1200 can cause the computer 1200 to function as one or more "parts" of the apparatus according to the embodiment, or to cause the computer 1200 to execute operations associated with the apparatus according to the embodiment or such one or more "parts", and / or to cause the computer 1200 to execute a process or a stage of such process according to the embodiment. Such a program may be executed by the CPU 1212 to cause the computer 1200 to execute specific operations associated with some or all of the blocks in the flowcharts and block diagrams described herein.

[0223] The computer 1200 according to this embodiment includes a CPU 1212, RAM 1214, and a graphics controller 1216, which are interconnected by a host controller 1210. The computer 1200 also includes input / output units such as a communication interface 1222, a storage device 1224, a DVD drive, and an IC card drive, which are connected to the host controller 1210 via an input / output controller 1220. The DVD drive may be a DVD-ROM drive and a DVD-RAM drive, etc. The storage device 1224 may be a hard disk drive and a solid-state drive, etc. The computer 1200 also includes input / output units such as a ROM 1230 and a keyboard, which are connected to the input / output controller 1220 via an input / output chip 1240.

[0224] The CPU 1212 operates according to the programs stored in the ROM 1230 and RAM 1214, thereby controlling each unit. The graphics controller 1216 acquires the image data generated by the CPU 1212 and stores it in the frame buffer provided in the RAM 1214 or within itself, so that the image data is displayed on the display device 1218.

[0225] The communication interface 1222 communicates with other electronic devices via a network. The storage device 1224 stores programs and data used by the CPU 1212 in the computer 1200. The DVD drive reads programs or data from a DVD-ROM or the like and provides them to the storage device 1224. The IC card drive reads programs and data from an IC card and / or writes programs and data to an IC card.

[0226] The ROM 1230 stores boot programs and / or hardware-dependent programs of the computer 1200, which are executed by the computer 1200 when activated. The input / output chip 1240 may also connect various input / output units to the input / output controller 1220 via USB ports, parallel ports, serial ports, keyboard ports, mouse ports, etc.

[0227] The program is provided on a computer-readable storage medium such as a DVD-ROM or IC card. The program is read from the computer-readable storage medium and installed on a storage device 1224, RAM 1214, or ROM 1230, which are examples of computer-readable storage media, and executed by the CPU 1212. The information processing described within these programs is read by the computer 1200, resulting in coordination between the program and the various types of hardware resources described above. The apparatus or method may be configured to realize the operation or processing of information in accordance with the use of the computer 1200.

[0228] For example, when communication is performed between a computer 1200 and an external device, the CPU 1212 may execute a communication program loaded into the RAM 1214 and, based on the processing described in the communication program, instruct the communication interface 1222 to perform communication processing. Under the control of the CPU 1212, the communication interface 1222 reads transmission data stored in a transmission buffer area provided in a recording medium such as the RAM 1214, storage device 1224, DVD-ROM, or IC card, transmits the read transmission data to the network, or writes received data received from the network to a reception buffer area or the like provided on the recording medium.

[0229] Furthermore, the CPU 1212 may read all or necessary parts of a file or database stored on an external recording medium such as a storage device 1224, a DVD drive (DVD-ROM), or an IC card into the RAM 1214, and perform various types of processing on the data in the RAM 1214. The CPU 1212 may then write the processed data back to the external recording medium.

[0230] Various types of information, such as various types of programs, data, tables, and databases, may be stored on the recording medium and subjected to information processing. The CPU 1212 may perform various types of processing on the data read from the RAM 1214, including various types of operations, information processing, conditional judgments, conditional branching, unconditional branching, information retrieval / replacement, etc., as described throughout this disclosure and specified by the program instruction sequence, and write the results back to the RAM 1214. The CPU 1212 may also retrieve information in files, databases, etc., within the recording medium. For example, if a plurality of entries having attribute values ​​of a first attribute, each associated with the attribute value of a second attribute, are stored in the recording medium, the CPU 1212 may search among the plurality of entries for an entry that matches the specified condition for the attribute value of the first attribute, read the attribute value of the second attribute stored in that entry, and thereby obtain the attribute value of the second attribute associated with the first attribute that satisfies the predetermined condition.

[0231] The program or software module described above may be stored on or near the computer 1200 in a computer-readable storage medium. Alternatively, a recording medium such as a hard disk or RAM provided within a server system connected to a dedicated communication network or the Internet can be used as a computer-readable storage medium, thereby providing the program to the computer 1200 via the network.

[0232] In the flowcharts and block diagrams of the embodiments, blocks may represent stages in a process in which an operation is performed or "parts" of a device that have the role of performing an operation. Specific stages and "parts" may be implemented by dedicated circuits, programmable circuits supplied with computer-readable instructions stored on a computer-readable storage medium, and / or processors supplied with computer-readable instructions stored on a computer-readable storage medium. Dedicated circuits may include digital and / or analog hardware circuits, and may include integrated circuits (ICs) and / or discrete circuits. Programmable circuits may include reconfigurable hardware circuits, such as field-programmable gate arrays (FPGAs) and programmable logic arrays (PLAs), which include logical AND, logical OR, exclusive OR, negated AND, negated OR, and other logical operations, flip-flops, registers, and memory elements.

[0233] A computer-readable storage medium may include any tangible device capable of storing instructions to be executed by a suitable device, and as a result, a computer-readable storage medium having instructions stored therein will comprise a product that includes instructions that can be executed to create means for performing operations specified in a flowchart or block diagram. Examples of computer-readable storage media may include electronic storage media, magnetic storage media, optical storage media, electromagnetic storage media, semiconductor storage media, etc. More specific examples of computer-readable storage media may include floppy disks (registered trademark), diskettes, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), electrically erasable programmable read-only memory (EEPROM), static random access memory (SRAM), compact disk read-only memory (CD-ROM), digital versatile disk (DVD), Blu-ray (registered trademark) disk, memory stick, integrated circuit card, etc.

[0234] Computer-readable instructions may include assembler instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk®, Java®, C++, and conventional procedural programming languages ​​such as the C programming language or similar programming languages.

[0235] Computer-readable instructions may be provided to a general-purpose computer, a special-purpose computer, or a programmable circuit, either locally or via a wide area network (WAN) such as a local area network (LAN) or the internet, so that the computer-readable instructions may be executed by the processor or programmable circuit of a general-purpose computer, a special-purpose computer, or other programmable data processing device, in order to generate means for performing operations specified in a flowchart or block diagram. Examples of processors include computer processors, processing units, microprocessors, digital signal processors, controllers, microcontrollers, and the like.

[0236] Although the present invention has been described above using embodiments, the technical scope of the present invention is not limited to the scope described in the above embodiments. It will be apparent to those skilled in the art that various modifications or improvements can be made to the above embodiments. It will be clear from the claims that such modified or improved forms may also be included in the technical scope of the present invention.

[0237] It should be noted that the execution order of operations, procedures, steps, and stages in the devices, systems, programs, and methods shown in the claims, specifications, and drawings is not explicitly stated as "before," "prior to," etc., and that these can be implemented in any order unless the output of a previous process is used in a later process. Even if the operation flow in the claims, specifications, and drawings is described using phrases such as "first," "next," etc. for convenience, it does not mean that it is essential to perform the operations in that order.

[0238] Although embodiments of the present application have been described in detail based on the drawings, these are illustrative examples, and the present invention can be implemented in various other forms, including those described in the disclosure section of the invention, based on the knowledge of those skilled in the art.

[0239] Furthermore, the terms "section, module, unit" mentioned above can be replaced with "means" or "circuits." For example, the acquisition unit can be replaced with acquisition means or acquisition circuit.

[0240] 1a System 10a Information Processing Device 20a Communication Unit 30a Storage Unit 31a Lifelog Database 32a Model Database 40a Control Unit 41a Personality Model Generation Unit 42a Long-Term Memory Model Generation Unit 43a Self-Will Model Generation Unit 44a Estimation Model Generation Unit 45a Learning Unit 10b Information Providing Device 11b Communication Unit 12b Storage Unit 13b Control Unit 121b Analysis Model Information 122b User Information 123b Agent Model Information 131b Collection Unit 132b Creation Unit 133b Calculation Unit 134b Update Unit 1c System 10c Information Processing Device 20c Communication Unit 30c Storage Unit 31c Lifelog Database 32c Model Database 40c Control Unit 41c Personality Model Generation Unit 42c Selection Model Generation Unit 43c Selection Unit

Claims

1. A system characterized by comprising: a personality model unit that learns various human personalities and has multiple personality models; a selection model unit that selects an appropriate personality model from the personality models constructed by the personality model unit according to the time, place, and occasion; and a personality formation unit in which the AI ​​develops will and forms a personality based on the personality model selected by the selection model unit.

2. The system according to claim 1, characterized in that the personality model unit learns human reactions and behaviors in different situations such as friendships, work, and family, and constructs each personality model.

3. The system according to claim 1, characterized in that the selection model unit selects an appropriate personality model according to the time, place, and occasion, and selects the optimal personality model according to the situation.

4. A system characterized by comprising: a collection unit that collects the user's life log; and a learning unit that generates an AI that understands the other party's emotions and has long-term memory of past conversations and actions, based on learning from the life log data collected by the collection unit.

5. The system according to claim 4, further comprising a selection unit that identifies the user's emotional value and next action when the user's life log is collected by the collection unit, and the learning unit, through learning based on the life log data collected by the collection unit and the information identified by the selection unit, understands the other party's emotions, stores past conversations and actions in long-term memory, and further generates an AI that possesses emotions.

6. A system characterized by comprising: a data collection unit that collects data from social networking services (SNS); a model construction unit that constructs a personality model based on the data collected by the data collection unit; a happiness map creation unit that creates a happiness map based on the personality model constructed by the model construction unit; and a reinforcement learning unit that performs reinforcement learning based on the happiness map created by the happiness map creation unit.

7. The system according to claim 6, characterized in that the collection unit collects data using the API of each SNS.

8. A system characterized by comprising: a personality model generation unit that generates multiple personality models corresponding to each personality by performing learning based on the life logs of each individual personality; a long-term memory model generation unit that, when a person's life log is input, generates a long-term memory model that has been trained to output memory vectors indicating the emotions that the person has towards the subject in the life log; and a learning unit that performs learning of the personality models using memory vectors based on the person's life log as input.

9. An information providing device comprising: a creation unit that creates a model for calculating the degree of happiness according to the attributes, based on a combination of user attributes and information on factors that contribute to each user's feeling of happiness; and an update unit that updates an agent by reinforcement learning using the degree of happiness calculated using the model as a reward.

10. A system characterized by comprising: a personality model generation unit that generates multiple personality models corresponding to each personality by learning the life logs of each person's personality; a selection model generation unit that generates a selection model that is trained to select one of the answers output from each of the multiple personality models based on the life log, when a person's life log is input; and a selection unit that, when a person's life log is input, inputs the life log to the multiple personality models and selects one of the answers output from each of the multiple personality models based on the life log using the selection model.

Images