Information processing device, information processing method, and program

Abstract

This information processing device comprises: a first acquiring unit that acquires a question to be answered regarding an accident; a first output unit that outputs the question acquired by the first acquiring unit; a first answer accepting unit that accepts an answer to the question output by the first output unit; a second acquiring unit that acquires a question asking for the rationale behind the answer accepted by the first answer accepting unit; a second output unit that outputs the question acquired by the second acquiring unit; and a second answer accepting unit that accepts an answer to the question output by the second output unit.

Description

Information Processing Apparatus, Information Processing Method, and Program

[0001] The present disclosure relates to an information processing apparatus, an information processing method, and a program.

[0002] Conventionally, a safety education system for improving the ability related to danger prediction has been known. For example, in Patent Document 1, a video is projected, a scene in which a dangerous location is hidden in the video is displayed as a still image on the screen, and a location that the responder determines to be dangerous with respect to this still image is pointed out on the screen, and a safety education system for determining the correctness of the pointing is disclosed.

[0003] Japanese Patent Application Laid-Open No. 2007-57578

[0004] However, in the prior art, confirmation and pointing have been performed only on the still images at specific times specified in advance by the administrator and only for the dangerous locations displayed. Therefore, the members passively perform the training regarding danger prediction as they get used to it, and the effect of the training has been limited.

[0005] Therefore, one of the purposes of the present disclosure is to provide a technology that enables a user to actively participate in training related to danger prediction and enhance the user's sensitivity to danger.

[0006] An information processing device according to one aspect of this disclosure is an information processing device for performing training on the prediction of accidents that may occur in a given environment, comprising: a first acquisition unit that acquires questions to be answered regarding an accident; a first output unit that outputs the questions acquired by the first acquisition unit; a first answer receiving unit that receives answers to the questions output by the first output unit; a second acquisition unit that, if the answer received by the first answer receiving unit is related to an accident but does not contain information regarding unsafe actions that cause an accident, acquires questions asking for the basis of the answer received by the first answer receiving unit; a second output unit that outputs the questions acquired by the second acquisition unit; and a second output unit that receives answers to the questions output by the second output unit. The system comprises a second response receiving unit that receives responses, a third acquisition unit that acquires questions to guide the response to unsafe behavior if the response received by the second response receiving unit does not contain information about unsafe behavior, a third output unit that outputs the questions acquired by the third acquisition unit, and a third response receiving unit that accepts answers to the questions output by the third output unit. The third acquisition unit acquires questions to guide the response to unsafe behavior until the response received by the third response receiving unit contains information about unsafe behavior, the third output unit outputs the questions acquired by the third acquisition unit, and the third response receiving unit accepts answers to the questions output by the third output unit, and this process is repeated.

[0007] According to one aspect of this disclosure, it is possible to provide technology that enables users to actively participate in training on hazard prediction, enhance their sensitivity to danger, and develop the ability to recognize unsafe behaviors on their own.

[0008] This figure shows an example of the system configuration of the information processing system according to this embodiment. This figure shows an example of the hardware configuration of the information processing device, terminal device, and external device according to this embodiment. This figure shows an example of the functional block configuration of the information processing device according to this embodiment. This figure shows an example of the risk prediction DB according to this embodiment. This figure shows an example of the risk prediction DB according to this embodiment. This figure shows an example of the user information DB according to this embodiment. This figure shows an example of the conversation history DB according to this embodiment. This figure shows an example of the functional block configuration of the terminal device according to this embodiment. This figure shows an example of the screen displayed on the terminal device according to this embodiment. This figure shows another example of the screen displayed on the terminal device according to this embodiment. This figure shows another example of the screen displayed on the terminal device according to this embodiment. This flowchart shows an example of a processing procedure executed by the information processing device according to this embodiment. This flowchart shows another example of a processing procedure executed by the information processing device according to this embodiment.

[0009] Embodiments of this disclosure will be described with reference to the attached drawings. In each drawing, components with the same reference numerals have the same or similar configuration. In this embodiment, the information processing system 1 provides, as an example, training on hazard prediction to users related to the retail industry, but is not limited to this. The information processing system 1 can provide training on hazard prediction to users related to sites such as factories, nursing care, construction, logistics, and agriculture, users related to learning places such as schools and kindergartens, or users related to places of movement such as trains, cars, and bicycles. In addition, it can provide training on hazard prediction to all users in their daily lives, such as at work sites, in transportation infrastructure, medical facilities, public facilities, education, and homes. Users related to the retail industry are, as an example, employees engaged in the retail industry, but are not limited to these. Users include part-time workers, laborers, managers, customers, volunteers, users, children, etc. The specified environment may also include sites, learning places, places of movement, etc.

[0010] "Hazard prediction" refers to anticipating and predicting potential hazards in a workplace or work environment. "Hazard prediction training" is an activity aimed at preventing accidents and disasters by predicting and identifying potential hazards in a given environment. This training also includes activities that train users' thinking skills to enable them to predict potential hazards in their environment. By undergoing hazard prediction training before entering an environment and preparing for potential dangers, users can prevent accidents. Furthermore, through this training, users can think about potential hazards in their environment and, for example, predict them. Hereafter, hazard prediction training will also be referred to as KYT.

[0011] <Information Processing System 1> The information processing system 1 in the disclosed technology will be described below. Figure 1 is a diagram showing an example of the configuration of the information processing system 1 according to this embodiment. The information processing system 1 shown in Figure 1 includes an information processing device 10, one or more terminal devices 20, and an external device 30. The information processing device 10, the terminal devices 20, and the external device 30 are connected to each other so as to be able to communicate with each other via a network N. Note that the external device 30 is not an essential component for configuring the information processing system 1.

[0012] Network N is a network for communication between the information processing device 10, the terminal device 20, and the external device 30. For example, Network N may be the Internet, an intranet, a LAN (Local Area Network), a mobile communication network, a dedicated line, a packet communication network, a telephone line, an internal corporate network, other communication lines, or a combination thereof. Furthermore, Network N may be wired or wireless.

[0013] Information processing system 1 can converse with the user and provide hazard prediction education, such as KYT (Hazard Prediction Training), through an AI (Artificial Intelligence) agent utilizing a large-scale language model. The AI ​​agent can converse with the user in an environment built within information processing system 1. The AI ​​agent can output information to the user that it has autonomously generated by utilizing a large-scale language model based on information acquired from the user by the information processing device 10.

[0014] <Information Processing Device 10> The information processing device 10 is composed of a server or a personal computer, etc. The information processing device 10 is an information processing device 10 that is responsible for some of the information processing functions provided by the information processing system 1, such as conducting training related to the prediction of accidents that may occur in a predetermined environment, including KYT (Kiken Yochi Training). The information processing device 10 may be composed using a virtual server or a cloud server, etc. The information processing device 10 may also be called a computer. The information processing device 10 may also send instructions to an external device 30 and obtain output from the external device 30.

[0015] In this embodiment, the user can improve their ability to predict hazards by communicating with the AI ​​agent and, for example, undergoing training on predicting accidents that may occur in an environment selected by the user. Information on the mutual communication between the AI ​​agent and the user may be displayed via the display of the terminal device 20. For example, the AI ​​agent's image and the content of the conversation, including text information, may be displayed via the display. The mutual communication between the AI ​​agent and the user may also be conducted by voice via a speaker and microphone.

[0016] An AI agent is a contactless, virtual person equipped with AI. An AI agent may also be a system or software program for autonomously performing specific tasks. Furthermore, an AI agent may utilize generative AI such as a Large Language Model (LLM) or a Large Action Model (LAM) to support the user's KYT (Kiken Yochi Training). The AI ​​agent can output information to respond to the user, for example, as text data or audio data. When outputting text data or audio data from the terminal device 20, a video or image that appears to show the AI ​​agent actually speaking can also be output.

[0017] Hereinafter, "user conversation information" refers to information that the user outputs to the information processing device 10 via the terminal device 20. "Response information" refers to information that the information processing device 10 outputs to the user's terminal device 20. "Conversation information" is a general term for information exchanged between the user and the information processing system 1, including "user conversation information" and "response information". Conversation information is information related to communication and includes chat, voice, images, videos, etc.

[0018] <Terminal Device 20> The terminal device 20 is an information processing device that, for example, allows the user to output user conversation information to the information processing device 10 by operating the terminal device 20. The user can also view the response information output to the user by the information processing device 10 via the terminal device 20. The terminal device 20 is a computer such as a mobile phone (including a smartphone), tablet, or personal computer. The terminal device 20 may also be called a computer. The terminal device 20 may also be a robot, etc. By operating the terminal device 20, the user can converse with an AI agent while looking at illustrations and conduct KYT (Kiken Yochi Training) about unsafe behavior (hereinafter also referred to as unsafe behavior). By conducting KYT, the user can train their thinking skills, prevent unsafe behavior, and receive appropriate information to prevent accidents. The user is not limited to one person; multiple people may conduct KYT. For example, multiple people may discuss and identify potential hazards, and a representative may operate the terminal device 20 to receive the outputted response from the information processing device 10. Alternatively, each person may operate the terminal device 20, and the terminal device 20 may extract and output responses that represent the group of people. The information processing device 10 may receive the outputted responses.

[0019] Furthermore, the terminal device 20 may have an application program (app) installed for using various functions provided by the information processing system 1. The app may be web browsing software. The app may cause the terminal device 20 to execute at least a part of the processing disclosed in the embodiments shown below, among the various functions provided by the information processing system 1. When the app is executed, the terminal device 20 may access the information processing device 10 to send and receive information used to execute the app.

[0020] <External Device 30> The external device 30 is, for example, a computer that operates a large-scale language model and processes strings consisting of input natural language. The external device 30 stores a large amount of data used in the large-scale language model and uses the data stored in the database to perform processing associated with the large-scale language model.

[0021] A large-scale language model is a language model specialized for natural language processing and may be a type of so-called generative AI. When a prompt is input to a large-scale language model, it generates and outputs text based on that prompt. A large-scale language model is a general-purpose language model that can be adapted to natural language processing tasks such as information extraction, text summarization, text generation, and question-and-answer sessions. Furthermore, a large-scale language model can handle various data formats, including not only text data but also audio data and video data.

[0022] Furthermore, a prompt may also be an input sentence to a large-scale language model. Prompts input to a large-scale language model may include, for example, prompts to convey preconditions to the large-scale language model, prompts to generate questions related to risk prediction that include user conversation information, etc. By inputting prompts to the large-scale language model, the large-scale language model can generate appropriate response information in response to the conversation with the user.

[0023] The external device 30 may constitute a generative AI using a large-scale language model. For example, the external device 30 can send and receive information with the information processing device 10 via the generative AI's API (Application Programming Interface).

[0024] The external device 30 includes, for example, services that can be provided on the cloud. The external device 30 may be, for example, a generative AI server that provides cloud-based services using a large-scale language model such as OpenAI's ChatGPT. However, the generative AI server is not limited to this, and may be a generative AI server that performs natural language processing using a large-scale language model that provides similar functionality.

[0025] The external device 30 may, for example, cause a large-scale language model to generate response information including question information based on user conversation information and output it to the information processing device 10. Alternatively, the external device 30 may use hazard prediction examples stored in the hazard prediction DB 100a as training data to train the large-scale language model on patterns and relationships and generate response information. The information processing device 10 may also cooperate with other external storage devices to provide KYT to the user. For example, by cooperating with a storage device managed by a certain company, the information processing device 10 can train a large-scale language model on company-specific examples and provide KYT related to those company-specific examples.

[0026] Furthermore, the external device 30 can have the large-scale language model search the hazard prediction DB 100a and obtain information. The information processing device 10 sends appropriate instructions to the external device 30, and the external device 30 obtains the information output from the information processing device 10. For example, when the information processing device 10 receives work details from a user, it sends an instruction to the external device 30 to generate questions about the work details along with the work details. The external device 30 has the large-scale language model refer to the hazard prediction DB 100a and generate questions about the work details based on the information it has referred to. The external device 30 then sends the questions generated by the large-scale language model to the information processing device 10 as response information. The information processing device 10 can send the generated questions to the terminal device 20 and have the questions displayed on the terminal device 20.

[0027] <Hardware Configuration> Figure 2 shows an example of the hardware configuration of the information processing device 10, terminal device 20, and external device 30 according to this embodiment. The information processing device 10, terminal device 20, and external device 30 each have a processor 11 such as a CPU (Central Processing Unit) or GPU (Graphics Processing Unit), a storage device 12 such as memory (for example, RAM (Random Access Memory) or ROM (Read Only Memory)), HDD (Hard Disk Drive) and / or SSD (Solid State Drive), a communication interface 13 such as a network interface (Network Interface) for wired or wireless communication, an input device 14 for receiving input operations, and an output device 15 for outputting information. The input device 14 is, for example, a keyboard, touch panel, mouse, camera, and / or microphone. The output device 15 is, for example, a display, touch panel, and / or speaker. The input device 14 receives various types of information from the user. The output device 15 may also include a display unit that displays various types of information to the user.

[0028] The hardware configuration described above is merely an example. The information processing device 10, terminal device 20, and external device 30 within the information processing system 1 may omit some of the hardware shown in Figure 2, or may include hardware not shown in Figure 2. Furthermore, the hardware shown in Figure 2 may be composed of one or more devices. For example, an example of an input device 14, such as a keyboard, touch panel, mouse, camera, and / or microphone, may be composed of other devices, and an example of an output device 15, such as a display, touch panel, and / or speaker, may be composed of other devices. Also, if the information processing device 10 is composed of multiple devices, each device may include at least some of this hardware, and the same applies to the terminal device 20 and external device 30.

[0029] <Functional Block Configuration> (Information Processing Device 10) Figure 3 is a diagram showing an example of the functional block configuration of the information processing device 10 according to this embodiment. The information processing device 10 includes a storage unit 100, a first acquisition unit 101, a first output unit 102, a first response receiving unit 103, a second acquisition unit 104, a second output unit 105, and a second response receiving unit 106. The information processing device 10 may further include a third acquisition unit 107, a third output unit 108, a third response receiving unit 109, a fourth acquisition unit 110, a fourth output unit 111, a fourth response receiving unit 112, a fifth acquisition unit 113, a fifth output unit 114, a fifth response receiving unit 115, and a prompt output unit 116.

[0030] The memory unit 100 can be implemented using the storage device 12 provided by the information processing device 10. Furthermore, the first acquisition unit 101, the first output unit 102, the first response receiving unit 103, the second acquisition unit 104, the second output unit 105, the second response receiving unit 106, the third acquisition unit 107, the third output unit 108, the third response receiving unit 109, the fourth acquisition unit 110, the fourth output unit 111, the fourth response receiving unit 112, the fifth acquisition unit 113, the fifth output unit 114, the fifth response receiving unit 115, and the prompt output unit 116 can be implemented by the processor 11 of the information processing device 10 executing a program stored in the storage device 12.

[0031] Furthermore, the program may be stored in a storage medium. The storage medium on which the program is stored may be a computer-readable non-transitor-readable medium. The non-transitor-readable medium is not particularly limited, but may be, for example, a USB (Universal Serial Bus) memory or a CD-ROM (Compact Disc Read-Only Memory).

[0032] The storage unit 100 stores data necessary for the information processing device 10 to perform information processing. The data includes a hazard prediction DB (Data Base) 100a, a user information DB 100b, and a conversation history DB 100c. It is possible to add data items to each DB as needed.

[0033] Figure 4 shows an example of the hazard prediction DB 100a according to this embodiment. The hazard prediction DB 100a stores various information related to hazard prediction cases. The hazard prediction DB 100a may store the following in association: hazard prediction case ID, hazard prediction case, industry, type of accident, cause of occurrence, unsafe behavior, countermeasures, and hazard prediction content.

[0034] The Hazard Prediction Case ID is information that identifies a Hazard Prediction Case. A Hazard Prediction Case is, for example, information such as a report summarizing the details of an accident. The industry, type of accident, contributing factors, unsafe behavior, countermeasures, and hazard prediction content are, for example, information extracted from a Hazard Prediction Case.

[0035] Figure 5 shows an example of a hazard prediction example 300 according to this embodiment. In the example shown in Figure 5, the hazard prediction example 300 includes, for example, a title area 301, a detail area 302, and a hazard prediction content area 303, based on an arbitrary format. Other areas may also be included. The hazard prediction example 300 may also include cases related to so-called "near misses," which are incidents where a dangerous situation occurred but fortunately did not result in an accident.

[0036] The title area 301 contains the title of the hazard prediction example. The title area 301 may also contain the name of the work, such as "transporting goods with a trolley" or "changing light bulbs using a stepladder," to make the work easier to visualize. Preferably, the title includes the key points of the accident, such as "injuring the lower back and shoulder while transporting goods with a trolley" or "falling from a stepladder while cleaning the outside of a building window." Preferably, the title is such that the content of the work can be easily visualized. The title area 301 may also include a subtitle, which may contain a summary of the hazard prediction example.

[0037] The detail area 302 contains detailed information about the accident for each predetermined item. The predetermined items in the detail area 302 include, for example, industry, type of accident, cause, circumstances of the accident, unsafe behavior, and countermeasures. Furthermore, the detail area 302 may also include items such as the causative agent and location information of the accident.

[0038] The hazard prediction content area 303 displays, for example, an illustration representing an overview of the hazard prediction content. The illustration may include, for example, an illustration depicting the occurrence situation. Note that the hazard prediction content is not limited to still images including illustrations and photographs, but may also include, for example, a video depicting the occurrence situation or audio.

[0039] The details of item 302 will now be explained. The industry is the industry to which the hazard prediction example is relevant. Examples of industries include agriculture, forestry, fisheries, construction, manufacturing, information and communications, transportation, postal services, wholesale, and retail. The industry may also include work in daily life, such as at work sites, transportation infrastructure, medical facilities, public facilities, education, and homes.

[0040] An accident type includes phenomena related to the causative object that caused the injury or illness. For example, an accident type is described as the circumstances under which the accident occurred, such as getting a hand caught in machinery while repairing it, or getting burned while performing gas welding. Accident types also include "falls," "tumbles," "collisions," "falling objects," "collapse," "being hit," "being caught in," "cuts / abrasions," "stepping through," "drowning," "contact with high / low temperature objects," "contact with hazardous substances," "electric shock," "explosion," "rupture," and "fire." Accident types may also include "back pain," "twisting the waist," and "dust getting in the eyes." If multiple types are in conflict, the primary type may be selected when considering measures to prevent accidents.

[0041] The cause of occurrence is the factor that led to the accident when the accident occurred. The cause of occurrence may be a factor related to things such as aging, fatigue, usage limit, incomplete protection / safety devices, etc., or a factor related to people such as forgetfulness, speculation, etc., or a factor related to management such as failure to take measures against unexpected dangers, using defective machines and tools, etc. Also, the cause of occurrence may be a factor related to actions such as "losing balance", "slipping", "tripping", etc.

[0042] The situation of occurrence includes the content that describes the situation when the accident occurred, such as the date, time, weather, surrounding environmental conditions, the content of the work the victim was doing, etc. The content of the work the victim was doing can be classified into high-altitude work, hot work, lifting work, assembly work, welding work, electrical wiring work, tank entry work, heat stroke, heavy object transportation, lathe work, forklift transportation, knife work, grinder work, cleaning work, step ladder work, trolley transportation, etc., and may be tagged.

[0043] Unsafe behavior is an action that may impede the safety of the person himself or related persons. Unsafe behavior may include actions that cause accidents, dangerous actions, and actions that lead to accidents. Unsafe behavior may include unconscious, inattentive, and other conscious actions. As an example of unsafe behavior, carrying objects without looking ahead, working with a stretched back, working without hanging tools on a strap, and moving while holding tools in hand without putting them in a waist pouch may be included.

[0044] Also, the content described in the items of unsafe behavior may be described by associating the factors that induce unsafe behavior with unsafe behavior. The factors that induce unsafe behavior include factors of workers, factors of work, factors of work environment, factors of safety management, factors of organization, etc. Also, in the items of unsafe behavior, the "should-be posture" may be described in association with unsafe behavior, and unsafe behavior may be a deviation from the "should-be posture" of work behavior.

[0045] Measures include the measures established from the perspective of how to prevent the accident in advance for the accident that occurred. Specific actions may be described in the measures.

[0046] FIG. 6 is a diagram showing an example of the user information DB 100b according to the present embodiment. The user information DB 100b stores various types of information related to the user. The user information DB 100b may store the user ID, user information, and response information in association with each other.

[0047] The user ID is information for identifying the user. The user information stores, for example, the user name (nickname) used in KYT, profile information, personal information such as work history, the position of the terminal device 20 used by the user (location information), the user's experience value, the user's level, and the like.

[0048] The response information stores, for example, the response information answered by the user in KYT, the difficulty level of the risk prediction cases learned in KYT, and the like. In addition, the content of the user's answer to the question may be stored in the response information. Further, the response information may store the degree of understanding and achievement calculated based on analyzing the user's answer content and whether the user was able to answer correctly.

[0049] Specifically, the response information may store, for each user, evaluation information after training (for example, five-level evaluation), the number of training sessions for each predetermined item such as the type of accident, the degree of progress without relying on questions or hints, scores for unsafe behaviors, the number of types of countermeasures answered, the progress of the entire KYT, the cumulative number of training sessions of KYT, the cumulative training time, the satisfaction level in questions of KYT, and the like. Note that the user's experience value, level, degree of understanding, and / or degree of achievement may be calculated based on the above response information.

[0050] In addition, the response information may store the score calculated based on the unsafe behavior answered by the user. For example, the score may be calculated based on a predetermined standard such that the act of stepping on the top plate of a stepladder has a risk level of 100%, the act of straddling a stepladder has a risk level of 90%, and the act of leaning out from a stepladder has a risk level of 80%. Further, the score may be calculated based on the number of types of countermeasures for avoiding the considered risks, or the score may be calculated based on the degree of reach within the set conversation time.

[0051] Furthermore, KYT (Know Yourself) may have a difficulty level assigned to it. The difficulty level may be, for example, an indicator showing the difficulty of the KYT, with a higher value indicating a more difficult KYT. For example, an advanced difficulty KYT may have more countermeasures to answer, more questions, fewer hints, or a longer KYT time than an intermediate difficulty KYT, but it is not limited to these.

[0052] Figure 7 shows an example of the conversation history DB 100c according to this embodiment. The conversation history DB 100c stores the conversation history acquired in KYT based on conversation information with the user. The conversation history DB 100c includes a user ID, a history ID that identifies the conversation history, acquired user conversation information, output response information, and a history date and time indicating the date and time when the user conversation information was acquired or the response information was output. Return to Figure 3 and continue the explanation.

[0053] The first acquisition unit 101 acquires questions that elicit responses regarding accidents that may occur in a given environment. The first acquisition unit 101 may, for example, cooperate with an external device 30 and acquire questions that elicit responses regarding accidents from the external device 30. The questions that elicit responses regarding accidents may be questions in which the answer includes the type of accident, or questions in which the answer includes the cause of occurrence, the circumstances of occurrence, unsafe behavior, countermeasures, etc.

[0054] Furthermore, the first acquisition unit 101 may acquire various information from the user before acquiring questions to prompt the user to answer about an accident. For example, it may acquire user information including the user ID and the type of industry to which the user belongs, information related to the work the user will perform, and information related to the scope of KYT requested by the user. Here, as an example, we will explain the provision of hazard prediction training to a user related to the retail industry, and the user related to the retail industry will be described as an employee engaged in the retail industry, but this is not limited to this.

[0055] The first acquisition unit 101 generates a prompt that incorporates information received from the user (for example, information about the scope of KYT to be performed, work content, etc., obtained from the user) into pre-set generation request information (hereinafter also referred to as a prompt). The prompt output unit 116 then outputs information including the prompt incorporating the work content and a prompt for generating questions to ask about accidents to an external device 30 having a large-scale language model.

[0056] The external device 30 may use the prompt output by the prompt output unit 116 to generate a question that prompts the large-scale language model to answer about an accident. For example, the external device 30 may cause the large-scale language model to search the hazard prediction DB 100a. The large-scale language model may generate a question related to the content output by the prompt output unit 116 based on the information it has retrieved. The external device 30 then outputs the question generated by the large-scale language model as response information to the information processing device 10.

[0057] The first acquisition unit 101 may acquire response information output from the external device 30. The first acquisition unit 101 may acquire a question, such as, "What dangers are there when changing a light bulb using a stepladder?", which is included in the acquired response information and is used to generate an accident response question for the large-scale language model.

[0058] Furthermore, the large-scale language model may be pre-trained on the contents of the hazard prediction DB 100a and generate appropriate questions within the scope of the contents of the hazard prediction DB 100a. For example, the large-scale language model may receive work details from the user and identify hazard prediction examples related to the work details from the hazard prediction DB 100a. The large-scale language model may then generate questions based on the identified hazard prediction examples. In addition, when the user provides an answer to a question, the large-scale language model may access the hazard prediction DB 100a, refer to the contents of the identified hazard prediction examples, and determine the content of the answer based on predetermined criteria.

[0059] The first output unit 102 outputs the question acquired by the first acquisition unit 101. The question acquired by the first acquisition unit 101 is output to the user from an output device 15 such as a display, touch panel and / or speaker. The first answer receiving unit 103 receives the answer to the question output by the first output unit 102. For example, the answer to the question output by the first output unit 102 is received from the user via an input device 14 such as a keyboard, touch panel, mouse, camera and / or microphone. For example, the first answer receiving unit 103 receives the answer "falling from the stepladder" output from the terminal device 20 to the question "What dangers are there when changing a light bulb using a stepladder?" output by the first output unit 102 to the terminal device 20 as user conversation information.

[0060] The second acquisition unit 104 acquires a question asking for the basis of the answer received by the first response receiving unit 103 if the answer received by the first response receiving unit 103 is related to an accident but does not include information about unsafe behaviors that cause accidents. The information related to an accident may include information that includes the type of accident.

[0061] The prompt output unit 116 may, for example, output the answer to the question received by the first answer receiving unit 103 to a large-scale language model and have the large-scale language model determine whether or not the information is related to an accident but does not contain information about unsafe behavior that causes the accident. For example, if the answer to the received question contains only content such as "losing balance," the large-scale language model may determine that the answer is a factor in the accident and does not contain information about unsafe behavior that causes the accident.

[0062] If the large-scale language model determines that the information concerns an accident but does not contain information about unsafe behaviors that cause the accident, the second acquisition unit 104 may generate a prompt that incorporates the answer to the question received by the first answer receiving unit 103 into a pre-set prompt. The prompt output unit 116 may then output information to the large-scale language model that includes a prompt incorporating the answer to the question received by the first answer receiving unit 103, and a prompt for generating a question asking for the basis of the answer to the question received by the first answer receiving unit 103.

[0063] Furthermore, the external device 30 may use the prompt output by the prompt output unit 116 to cause the large-scale language model to generate a question asking for the basis of the answer to the question received by the first answer receiving unit 103. For example, the external device 30 may cause the large-scale language model to search the risk prediction DB 100a. Based on the information retrieved, the large-scale language model may generate a question regarding the content output from the prompt output unit 116. The external device 30 may then output the question generated by the large-scale language model as response information to the information processing device 10.

[0064] The second acquisition unit 104 may acquire response information output from the external device 30. The second acquisition unit 104 may acquire a question that asks for the basis of the answer to the question received by the first answer receiving unit 103, which was generated by the large-scale language model, as part of the acquired response information, for example, "Why did you fall off the stepladder?"

[0065] The second output unit 105 outputs the question acquired by the second acquisition unit 104. The question acquired by the second acquisition unit 104 is output to the user from an output device 15 such as a display, touch panel, and / or speaker. The second answer receiving unit 106 receives the answer to the question output by the second output unit 105. For example, the answer to the question output by the second output unit 105 is received from the user via an input device 14 such as a keyboard, touch panel, mouse, camera, and / or microphone. For example, the second answer receiving unit 106 receives the answer "Because you lose your balance" output from the terminal device 20 in response to the question "Why do people fall off stepladders?" output by the second output unit 105 to the terminal device 20, as user conversation information.

[0066] The third acquisition unit 107 acquires a question that prompts the user to answer about unsafe behavior if the answer received by the second answer receiving unit 106 does not contain information about unsafe behavior. The question that prompts the user to answer about unsafe behavior may also be a question that involves thinking about unsafe behavior. For example, the prompt output unit 116 may output the answer to the question received by the second answer receiving unit 106 to a large-scale language model and have the large-scale language model determine whether or not it contains information about unsafe behavior.

[0067] If the large-scale language model determines that it does not contain information about safe behavior, the third acquisition unit 107 may generate a prompt that incorporates the answer to the question received by the second response receiving unit 106 into a pre-set prompt. The prompt output unit 116 may then output information to the large-scale language model that includes a prompt incorporating the answer to the question received by the second response receiving unit 106, and a prompt for generating a question that leads to an answer about unsafe behavior.

[0068] Furthermore, the external device 30 may use the prompt output by the prompt output unit 116 to generate questions that guide the large-scale language model to respond with unsafe behavior. For example, the external device 30 may cause the large-scale language model to search the risk prediction DB 100a. Based on the information retrieved, the large-scale language model may generate questions related to the content output from the prompt output unit 116. The external device 30 may then output the questions generated by the large-scale language model as response information to the information processing device 10.

[0069] The third acquisition unit 107 may acquire response information output from the external device 30. The third acquisition unit 107 may acquire a question included in the acquired response information that guides the large-scale language model to respond with an unsafe behavior, such as, for example, "Why is the balance disrupted?"

[0070] The third output unit 108 outputs the question acquired by the third acquisition unit 107. The question acquired by the third acquisition unit 107 is output to the user from an output device 15 such as a display, touch panel, and / or speaker. The third answer receiving unit 109 receives the answer to the question output by the third output unit 108. For example, the answer to the question output by the third output unit 108 is received from the user via an input device 14 such as a keyboard, touch panel, mouse, camera, and / or microphone. For example, the third answer receiving unit 109 receives the answer "Because you reach diagonally upwards to change the light bulb" output from the terminal device 20 to the question "Why do you lose your balance?" output by the third output unit 108 to the terminal device 20 as user conversation information.

[0071] Furthermore, the prompt output unit 116 acquires questions that guide the user to respond with information about unsafe behavior until the response received by the third response receiving unit 109 contains information about unsafe behavior. For example, the prompt output unit 116 may output the response received by the third response receiving unit 109 to a large-scale language model and have the large-scale language model determine whether or not the received response contains information about unsafe behavior.

[0072] If the large-scale language model determines that the received response does not contain information about unsafe behavior, the third acquisition unit 107 may generate a prompt that incorporates the answer to the question received by the third response receiving unit 109 into a pre-set prompt. The prompt output unit 116 may then output information to the large-scale language model that includes a prompt incorporating the answer to the question received by the third response receiving unit 109, and a prompt for generating a question that leads to the response of unsafe behavior. For example, if the answer to the received question contains content such as "reach your hand diagonally upwards," the large-scale language model may determine that the answer contains information about unsafe behavior.

[0073] The external device 30 may use the prompt output by the third acquisition unit 107 to generate a question that guides the large-scale language model to respond with an unsafe behavior. If the large-scale language model determines that it has generated a question that guides it to respond with an unsafe behavior multiple times, it may generate a question with a reduced difficulty level, for example, depending on the number of times it has generated a question that guides it to respond with an unsafe behavior. For example, it may generate a question that includes content related to behaviors that lead to unsafe behavior, or it may generate a question that includes unsafe behaviors in similar cases. The external device 30 may then output the question generated by the large-scale language model to the information processing device 10 as response information.

[0074] The third acquisition unit 107 may acquire response information output from the external device 30. The third acquisition unit 107 may acquire a question that leads to the large-scale language model to respond with an unsafe behavior, which is included in the acquired response information, for example, "What position should you be in to change a light bulb?"

[0075] The process repeats as follows: "The third output unit 108 outputs the question acquired by the third acquisition unit 107," and "The third answer receiving unit 109 receives the answer to the question output by the third output unit 108." For example, the third answer receiving unit 109 receives the answer "Reach diagonally upwards" output from the terminal device 20 to the question "What position would you be in to change a light bulb?" which was output by the third output unit 108 to the terminal device 20, as user conversation information.

[0076] Through the above processing, the information processing device 10 asks the user questions that prompt them to think about accidents, enabling the user to actively engage in KYT (Kiken Yochi Training). Furthermore, by asking the user questions that prompt them to think about unsafe behaviors, the user can consider what kinds of behaviors constitute unsafe behaviors. As a result, through conversations including question-and-answer sessions within KYT, the user can intuitively grasp what is dangerous and what conditions lead to dangerous situations, thereby enhancing their ability to sense the risk of accidents (hereinafter also referred to as risk sensitivity).

[0077] Furthermore, by asking the user for the reasons behind their answers, the information processing device 10 allows the user to consider all possibilities, examine their answers from a different perspective, and think about the root cause, thereby increasing their risk awareness.

[0078] Furthermore, the second acquisition unit 104 may acquire questions that inquire about the basis for the answers to the questions received by the first answer receiving unit 103, such as, "Why did you think that way?" or "Why would that lead to an accident?". By being questioned about the reasons for their answers and their attitude towards risk assessment through the questions acquired by the second acquisition unit 104, users can be made aware that they are taking unsafe actions that deviate from the "ideal state" due to familiarity or overconfidence, such as "I thought this much would be okay," "It was too much trouble," "Everyone else is doing it," or "It was unavoidable in order to speed up the work."

[0079] The fourth acquisition unit 110 acquires questions that guide the respondent to provide countermeasures to prevent unsafe behavior if the response to the question asking for justification received by the second response receiving unit 106 includes information about unsafe behavior.

[0080] Furthermore, if the answer to the question prompting the user to respond about unsafe behavior, which was received by the third response receiving unit 109, includes information about unsafe behavior, the fourth acquisition unit 110 may acquire a question prompting the user to respond about measures to prevent unsafe behavior.

[0081] The prompt output unit 116 may, for example, output the answer to the question received by the second answer receiving unit 106 to a large-scale language model and have it determine whether or not it contains information indicating unsafe behavior. If the large-scale language model determines that it contains information related to unsafe behavior, the fourth acquisition unit 110 may generate a prompt that incorporates the answer to the question received by the second answer receiving unit 106 into a pre-set prompt. The prompt output unit 116 then outputs information to the large-scale language model that includes the prompt incorporating the answer to the question and a prompt for generating a question in which the answer is a measure to prevent unsafe behavior. For example, if the answer to the received question contains content such as "reach your hand diagonally upwards," the large-scale language model may determine that the answer contains information related to unsafe behavior.

[0082] Furthermore, the external device 30 may use the prompt output by the prompt output unit 116 to generate questions that guide the large-scale language model to respond with countermeasures to prevent unsafe behavior. For example, the external device 30 may cause the large-scale language model to search the risk prediction DB 100a. Based on the information retrieved, the large-scale language model may generate questions related to the content output from the prompt output unit 116. The external device 30 then outputs the questions generated by the large-scale language model as response information to the information processing device 10.

[0083] The fourth acquisition unit 110 may acquire response information output from the external device 30. The fourth acquisition unit 110 may acquire a question included in the acquired response information that guides the user to answer with countermeasures to prevent unsafe behavior generated by a large-scale language model, such as, "Let's decide on countermeasures to avoid danger. What countermeasures can be considered to avoid danger?"

[0084] The fourth output unit 111 outputs the question acquired by the fourth acquisition unit 110. The question acquired by the fourth acquisition unit 110 is output to the user from an output device 15 such as a display, touch panel and / or speaker. The fourth answer receiving unit 112 receives the answer to the question output by the fourth output unit 111. For example, the answer to the question output by the fourth output unit 111 is received from the user via an input device 14 such as a keyboard, touch panel, mouse, camera and / or microphone. The fourth answer receiving unit 112 may receive an answer from the user such as, "Place the stepladder directly under the light bulb, and work using three-point support by using only one side of the stepladder without straddling it."

[0085] Through the above processing, the information processing device 10 asks the user about accident countermeasures, enabling the user to consider what measures can be taken to prevent accidents and thereby increasing their risk awareness. Furthermore, the user can realize that if they become aware of unsafe behavior, they can take effective countermeasures.

[0086] If the answer to the question prompting the person to answer about the accident, which was received by the first answer receiving unit 103, does not contain any information about the accident, the fifth acquisition unit 113 acquires a hint for the question acquired by the first acquisition unit 101.

[0087] The prompt output unit 116 may, for example, output the answer to the question received by the first answer receiving unit 103 to the large-scale language model and have it determine whether or not it contains information related to an accident. If the large-scale language model determines that it does not contain information related to an accident, the fifth acquisition unit 113 generates a prompt that incorporates the answer to the question received by the first answer receiving unit 103 into a pre-set prompt. The prompt output unit 116 then outputs information to the large-scale language model that includes the prompt incorporating the answer to the question and a prompt for generating hints for the question.

[0088] Similarly, the prompt output unit 116 may output the answer to the question received by the second answer receiving unit 106 to a large-scale language model to determine whether or not it contains information indicating unsafe behavior. If the large-scale language model determines that it does not contain information indicating unsafe behavior, the fifth acquisition unit 113 generates a prompt that incorporates the answer to the question received by the second answer receiving unit 106 into a pre-set prompt. The prompt output unit 116 then outputs the generated prompt to the external device 30. The fifth acquisition unit 113 may use the generated prompt to obtain hints for the question generated by the large-scale language model in the external device 30.

[0089] The external device 30 may use the prompt output by the prompt output unit 116 to cause the large-scale language model to generate hints for the question. For example, the external device 30 may cause the large-scale language model to search the risk prediction DB 100a. Based on the information retrieved, the large-scale language model may generate a question related to the content output from the prompt output unit 116. The external device 30 then outputs the question generated by the large-scale language model as response information to the information processing device 10.

[0090] The fifth acquisition unit 113 may acquire response information output from the external device 30. If the fifth acquisition unit 113 determines that the response information does not contain any information about an accident, it may acquire hints included in the acquired response information for questions generated by the large-scale language model, such as, "Let's consider the type of accident. Examples of accident types include 'falling,' 'stumbling,' and 'collision.' What type of accident applies in this case?"

[0091] Furthermore, if the fifth data acquisition unit 113 determines that no information indicating unsafe behavior is included, it may acquire a hint such as, "Try to express risk factors as much as possible using combinations of unsafe behavior (actions) and unsafe conditions." As a result, the user can consider answers that clearly identify "unsafe behavior" and "unsafe conditions."

[0092] Furthermore, if the user conversation information from the user includes content that is not related to the illustration (for example, if the user answers a question about a ladder illustration and provides information about safety measures such as a safety harness), the fifth acquisition unit 113 may acquire hints such as "Let's think about it from the perspective of standing on the ladder" or "Let's think about it within the scope of what is relevant to the illustration."

[0093] Furthermore, if user conversation information cannot be obtained from the user after a predetermined time (for example, 15 seconds), a hint such as "Have you ever had a scary experience on a stepladder?" may be obtained. The information processing device 10 may also output a conversation history, including past user conversation information, which is the content of past conversations with the user, to a large-scale language model. The fifth acquisition unit 113 may then obtain a hint for a question generated by the large-scale language model based on the conversation history.

[0094] The fifth output unit 114 outputs the hint acquired by the fifth acquisition unit 113. The hint acquired by the fifth acquisition unit 113 is output to the user from an output device 15 such as a display, touch panel, and / or speaker. The fifth response receiving unit 115 receives the response to the hint output by the fifth output unit 114. For example, the response to the hint output by the fifth output unit 114 is received from the user via an input device 14 such as a keyboard, touch panel, mouse, camera, and / or microphone. Note that the response to the hint output by the fifth output unit 114 is not limited to the fifth response receiving unit 115; for example, the first response receiving unit 103 or the second response receiving unit 106 may also receive it.

[0095] Furthermore, the fifth acquisition unit may continue to acquire hints for questions acquired by the first acquisition unit until the fifth answer receiving unit 115 receives hints for questions that include information about an accident, the fifth output unit may output the hints acquired by the fifth acquisition unit, and the fifth answer receiving unit may continue to receive the hints output by the fifth output unit.

[0096] Furthermore, if the hint response received by the fifth response receiving unit 115 includes information related to an accident, the second acquisition unit 104 may acquire a question asking for the basis of the answer to the question received by the first response receiving unit 103. The second output unit 105 may output the question asking for the basis acquired by the second acquisition unit 104, and the second response receiving unit 106 may receive the answer to the question asking for the basis output by the second output unit 105.

[0097] Through the above processing, the information processing device 10 provides the user with hints for questions, allowing the user to enhance their risk awareness by considering answers based on the provided hints. Furthermore, the information processing device 10 generates and outputs response information to provide the user with appropriate hints based on the acquired user conversation information, thereby making the user feel that they are actively participating in KYT (Kiken Yochi Training).

[0098] Furthermore, the questions acquired by at least one of the first acquisition unit 101, the second acquisition unit 104, and the third acquisition unit 107 may include at least one of a video and a still image related to the accident. For example, if at least one of the first acquisition unit 101, the second acquisition unit 104, and the third acquisition unit 107 acquires information that includes performing work related to a stepladder in the acquired user conversation information, it may acquire illustrations from the hazard prediction DB 100a as hazard prediction content related to work related to a stepladder, such as "a situation where a light bulb is being changed on the ceiling using a stepladder" or "a situation where a POP is being attached to a wall using a stepladder." At least one of the first output unit 102, the second output unit 105, and the third output unit 108 may output the illustrations together with the questions. Note that the hazard prediction content is not limited to being acquired from the hazard prediction DB 100a, but may be acquired by any method. For example, hazard prediction content including illustrations may be acquired by searching the web using a large-scale language model, or the large-scale language model may generate the hazard prediction content.

[0099] Furthermore, if the question includes hazard prediction content that includes illustrations, the question may also include information related to the hazard prediction content. For example, the first acquisition unit 101 generates a prompt that incorporates the content of the acquired illustration into a pre-set prompt. The prompt output unit 116 then outputs the generated prompt to the external device 30. The first acquisition unit 101 may acquire a question from the external device 30, generated by a large-scale language model, such as, "Consider the dangers hidden in the illustration. You are changing a light bulb on the ceiling using a stepladder, as shown in the illustration. What are the dangers?"

[0100] Through the above processing, the information processing device 10 can not only output text and audio information to the user, but also output visual information such as illustrations representing work scenes and environments where accidents may occur, which are hazard prediction content. As a result, it can help anticipate dangerous situations and support effective KYT (Hazard Prediction Training).

[0101] Furthermore, if the response received by the first response receiving unit 103 is related to an accident but does not include information about unsafe behaviors that cause accidents, the second acquisition unit 104 may acquire questions that lead to the response regarding the factors that cause accidents. The factors that cause accidents may include the factors that led to the accident if an accident occurs.

[0102] The second acquisition unit 104 may acquire questions that guide the user to answer about the factors causing the accident, such as, "What do you think are the factors causing the accident?" or "Why do accidents like this occur?", which are included in the acquired response information.

[0103] Through the above processing, the information processing device 10 asks questions that guide the user to answer about the factors that cause accidents, thereby enabling the user to improve their risk awareness by considering what factors might trigger an accident.

[0104] (Terminal device 20) Figure 8 is a diagram showing an example of the functional block configuration of the terminal device 20 according to this embodiment. The terminal device 20 includes a storage unit 200, a UI (User Interface) unit 201, and a control unit 202. The storage unit 200 can be realized using a storage device 12 provided by the terminal device 20. The UI unit 201 and the control unit 202 can be realized by the processor 11 of the terminal device 20 executing a program stored in the storage device 12. The program can be stored in a storage medium. The storage medium storing the program may be a computer-readable non-temporary storage medium. The non-temporary storage medium is not particularly limited, but may be, for example, a USB (Universal Serial Bus) memory or a CD-ROM.

[0105] The storage unit 200 stores the data necessary for the control unit 202 to perform information processing.

[0106] The UI unit 201 has the function of receiving various inputs from the user and displaying various screens on the display. The UI unit 201 may also display a chat screen on the display (display unit) of the terminal device 20 in accordance with the instructions of the information processing device 10.

[0107] The control unit 202 works in conjunction with the information processing device 10 to provide various functions necessary for executing information processing.

[0108] Regarding the functional block configuration described above, the terminal device 20 may be configured to provide all or part of the following components included in the information processing device 10: the storage unit 100, the first acquisition unit 101, the first output unit 102, the first response receiving unit 103, the second acquisition unit 104, the second output unit 105, the second response receiving unit 106, the third acquisition unit 107, the third output unit 108, the third response receiving unit 109, the fourth acquisition unit 110, the fourth output unit 111, the fourth response receiving unit 112, the fifth acquisition unit 113, the fifth output unit 114, and the prompt output unit 116. In other words, the various processes according to this embodiment may be executed by the processor of the information processing device 10, by the processor of the terminal device 20, or by the processors of the information processing device 10 and the terminal device 20 working together.

[0109] In this disclosure, "part" does not merely mean a physical means, but also includes cases where the functions of that "part" are realized by software. Furthermore, even if the functions of one "part" or device are realized by two or more physical means, devices, or software, the functions of two or more "parts" or devices may be realized by one physical means, device, or software.

[0110] <Screen Display Example> Figure 9 shows an example of a screen displayed on the terminal device 20 according to this embodiment. The difficulty level selection screen A100 is an example of a screen that accepts the user's selection of the difficulty level of KYT. The user can select one difficulty level from those displayed on the difficulty level selection screen A100 via an input device 14 such as a keyboard, touch panel, mouse, camera and / or microphone. Alternatively, after or before the user selects a difficulty level, a screen may be displayed that accepts the user's information about the task to be performed in the future (for example, 10 minutes later). The user can set the task content and difficulty level before performing hazard prediction training. Alternatively, a screen may be displayed that accepts the user before the user starts KYT. For example, the user may be recognized by facial recognition using a camera, or the user may be recognized by having them enter their name, login ID, etc. Also, before starting KYT, the time for performing KYT (for example, 5 minutes) can be set, and if the time for performing KYT is set in advance, the system may be controlled to ensure that the KYT is completed within the set time. Furthermore, the settings that can be configured before conducting KYT (Kiken Yochi Training) are not limited to user authentication, work content, and difficulty level settings.

[0111] Figure 10 shows another example of a screen displayed on the terminal device 20 according to this embodiment. The hazard prediction content display screen A200 is an example of a screen in which an illustration is displayed as hazard prediction content. The user can check the illustration B100 displayed on the hazard prediction content display screen A200 and perform KYT (Hazard Prediction Training). The AI ​​agent may spontaneously output response information using a large-scale language model based on user information, work content and / or the illustration displayed as hazard prediction content. The AI ​​agent may, for example, engage in casual conversation with the user based on the user's profile information stored in the user information DB. Once the casual conversation is finished, KYT can proceed.

[0112] The conversation information display screen A300 is an example of a screen that displays conversation information between an AI agent and a user. The conversation information display screen A300, which is a screen of the terminal device 20, can display the AI ​​agent. The user can converse with the AI ​​agent controlled in an environment constructed in the information processing system 1. The screen of the terminal device 20 may display an AI agent character generated by 3D (three-dimensional) modeling. The AI ​​agent may also control its mouth, facial expressions, or gestures in accordance with the conversation with the user. The AI ​​agent may perform actions such as nodding, nodding in agreement, or smiling while listening in response to the conversation with the user.

[0113] The user can view the conversation information displayed on the conversation information display screen A300 and receive hazard prediction training. The user can output answers B300 to questions B200, which are output from the AI ​​agent and sent to the user via output devices 15 such as a display, touch panel, and / or speaker, via voice input via input devices 14 such as a keyboard, touch panel, mouse, camera, and / or microphone, or via chat. The system may be controlled to praise the user before ending the KYT session.

[0114] Furthermore, the AI ​​agent can remember conversations unrelated to KYT (Kiken Yochi Training) and tailor the conversation to the user. The AI ​​agent may also be controlled to steer the conversation back to KYT if the user talks about something unrelated during the KYT session. Additionally, the AI ​​agent may offer greetings appropriate to the time of day.

[0115] The AI ​​agent's appearance may be configured to correspond to the KYT provided by the Information Processing System 1. For example, the AI ​​agent may be given work clothes and have a friendly appearance. The AI ​​agent may also be configured to provide verbal guidance and instructions from the system according to each user's KYT progress and level. For example, it may say to the user, "Let's think about the dangers together." Furthermore, if the user answers about the type of accident, such as "It might fall," the AI ​​agent may correct the wording to say, "It will fall, right?"

[0116] Figure 11 shows another example of a screen displayed on the terminal device 20 according to this embodiment. The results display screen A400 is an example of a screen that displays information that is fed back to the user after they have received hazard prediction training. By checking the KYT results displayed on the results display screen A400, the user can check their level of understanding and achievement of the KYT they have performed. By visualizing the KYT results with numbers and graphs, the user can feel that their awareness of safety has improved. In addition, the user's KYT level may be increased and displayed based on the cumulative number of sessions and cumulative time. This allows the user to grasp their own progress and continue KYT in an enjoyable way.

[0117] The results display screen A500 is an example of a screen that displays information provided as feedback to the user after completing hazard prediction training. The user can review the KYT (Hazard Prediction Training) by checking their answers within the KYT displayed on the results display screen A500. Furthermore, the results display screen A500 can display hazard prediction examples similar to the completed KYT from the hazard prediction database 100a.

[0118] <Operation> Next, the operation of the information processing device 10 according to this embodiment will be described. Figure 12 is a flowchart showing an example of a processing procedure performed by the information processing device 10 according to this embodiment. In this embodiment, it is assumed that various data are stored in the storage unit 100 before the processing in Figure 12 is started. As shown in Figure 12, the information processing device 10 may acquire various questions, and the large-scale language model may determine the content of the answers based on user conversation information including the answers from the user and generate response information. Note that the flowchart described in Figure 12 is an example of a processing procedure performed by the information processing device 10 and can be modified as appropriate.

[0119] In S101, the first acquisition unit 101 acquires questions that prompt responses regarding accidents that may occur in a predetermined environment. The first acquisition unit 101 may, for example, cooperate with an external device 30 and acquire questions that prompt responses regarding accidents from the external device 30. In addition, the first acquisition unit 101 may acquire various information from the user before acquiring questions that prompt responses regarding accidents that may occur in a predetermined environment. For example, it may acquire user information including the user ID and the industry to which the user belongs, information related to the work the user will perform, and information related to the scope of KYT requested by the user.

[0120] The questions acquired by the first acquisition unit 101 may include at least one of the video and / or still images related to the accident. For example, if the information acquired from the user includes performing work related to a stepladder, the first acquisition unit 101 may acquire an illustration representing "a situation where a light bulb is being changed using a stepladder" from the hazard prediction DB 100a as hazard prediction content related to work related to a stepladder.

[0121] In S102, the first output unit 102 outputs the question acquired by the first acquisition unit 101. The question acquired by the first acquisition unit 101 is output to the user from an output device 15 such as a display, touch panel and / or speaker. The first output unit 102 may also output an illustration along with the question. The first answer receiving unit 103 receives the answer to the question output by the first output unit 102. For example, the answer to the question output by the first output unit 102 is received from the user via an input device 14 such as a keyboard, touch panel, mouse, camera and / or microphone. For example, the first answer receiving unit 103 receives the answer "falling from the stepladder" output from the terminal device 20 to the question "What dangers are there when changing a light bulb using a stepladder?" output by the first output unit 102 to the terminal device 20 as user conversation information.

[0122] In S103, the prompt output unit 116 outputs the answer to the question received by the first answer receiving unit 103 to the large-scale language model. The large-scale language model determines whether the received answer is related to an accident but does not contain information about unsafe behaviors that could cause an accident.

[0123] If the large-scale language model determines that the received response is related to an accident but does not contain information about unsafe behaviors that could cause the accident (S103: YES), the process proceeds to S104. In S104, the second acquisition unit 104 acquires a question asking for the basis of the response received by the first response receiving unit 103.

[0124] In S105, the second output unit 105 outputs the question acquired by the second acquisition unit 104. The question acquired by the second acquisition unit 104 is output to the user from an output device 15 such as a display, touch panel, and / or speaker. The second answer receiving unit 106 receives the answer to the question output by the second output unit 105. For example, the answer to the question output by the second output unit 105 is received from the user via an input device 14 such as a keyboard, touch panel, mouse, camera, and / or microphone.

[0125] In S106, the prompt output unit 116 converts the answer to the question received by the second answer receiving unit 106 into a large-scale language model. The large-scale language model determines whether or not the answer received by the second answer receiving unit 106 contains information about unsafe behavior.

[0126] If the large-scale language model determines that the response received by the second response receiving unit 106 does not contain information about unsafe behavior (S106: NO), the process proceeds to S107. In S107, the third acquisition unit 107 acquires a question in which the response is unsafe behavior.

[0127] In S108, the third output unit 108 outputs the question acquired by the third acquisition unit 107. The question acquired by the third acquisition unit 107 is output to the user from an output device 15 such as a display, touch panel, and / or speaker. The third answer receiving unit 109 receives the answer to the question output by the third output unit 108. For example, the answer to the question output by the third output unit 108 is received from the user via an input device 14 such as a keyboard, touch panel, mouse, camera, and / or microphone.

[0128] In S109, the prompt output unit 116 outputs the response received by the third response receiving unit 109 to the large-scale language model. The large-scale language model determines whether or not the response received by the third response receiving unit 109 contains information about unsafe behavior.

[0129] If the large-scale language model determines that the response received by the third response receiving unit 109 does not contain any information about unsafe behavior (S109: NO), the process proceeds to S107.

[0130] On the other hand, if the large-scale language model determines that the response received by the third response receiving unit 109 contains information about unsafe behavior (S109: YES), the process proceeds to S110. Similarly, if the large-scale language model determines that the response received by the second response receiving unit 106 contains information about unsafe behavior (S106: YES), the process also proceeds to S110. In S110, the fourth acquisition unit 110 acquires questions that lead the user to respond with measures to prevent unsafe behavior.

[0131] In S111, the fourth output unit 111 outputs the question acquired by the fourth acquisition unit 110. The question acquired by the fourth acquisition unit 110 is output to the user from an output device 15 such as a display, touch panel, and / or speaker. Then, the fourth answer receiving unit 112 terminates processing when it receives the answer to the question output by the fourth output unit 111. For example, the answer to the question output by the fourth output unit 111 is received from the user via an input device 14 such as a keyboard, touch panel, mouse, camera, and / or microphone. When terminating, the fourth output unit 111 may output a message such as, "Thank you for your hard work, this concludes the KYT."

[0132] Furthermore, in S103, if the large-scale language model determines that the received response is not related to an accident, or if it determines that the received response is related to an accident but does not contain information about unsafe behavior that could cause an accident (S103: NO), the process proceeds to S112. In S112, the fifth acquisition unit 113 acquires hints for the question acquired by the first acquisition unit 101.

[0133] In S113, the fifth output unit 114 outputs the hint acquired by the fifth acquisition unit 113. The hint acquired by the fifth acquisition unit 113 is output to the user from an output device 15 such as a display, touch panel, and / or speaker. The fifth response receiving unit 115 receives the response to the hint output by the fifth output unit 114. For example, the response to the hint output by the fifth output unit 114 is received from the user via an input device 14 such as a keyboard, touch panel, mouse, camera, and / or microphone.

[0134] In S114, the prompt output unit 116 converts the hint response received by the fifth response receiving unit 115 into a large-scale language model. The large-scale language model determines whether or not the response received by the fifth response receiving unit 115 contains information related to the accident.

[0135] If the large-scale language model determines that the response received by the fifth response receiving unit 115 does not contain information about the accident (S114: NO), the process proceeds to S112. In S112, the fifth acquisition unit 113 obtains hints for the question acquired by the first acquisition unit 101. On the other hand, if the large-scale language model determines that the response received by the fifth response receiving unit 115 does contain information about the accident (S114: YES), the process proceeds to S104.

[0136] Next, other operations of the information processing device 10 according to this embodiment will be described. Figure 13 is a flowchart showing another example of a processing procedure executed by the information processing device 10 according to this embodiment.

[0137] In S201, the third acquisition unit 107 acquires questions in which the answer is an unsafe behavior. The third acquisition unit 107 may, for example, cooperate with an external device 30 and acquire questions in which the answer is an unsafe behavior from the external device 30. In addition, the third acquisition unit 107 may acquire various information from the user before acquiring questions in which the answer is an unsafe behavior. The questions acquired by the third acquisition unit 107 may include at least one of video and still images related to the accident.

[0138] In S202, the third output unit 108 outputs the question acquired by the third acquisition unit 107. The question acquired by the third acquisition unit 107 is output to the user from an output device 15 such as a display, touch panel, and / or speaker. The third answer receiving unit 109 receives the answer to the question output by the third output unit 108. For example, the answer to the question output by the third output unit 108 is received from the user via an input device 14 such as a keyboard, touch panel, mouse, camera, and / or microphone.

[0139] In S203, the prompt output unit 116 outputs the response received by the third response receiving unit 109 to the large-scale language model. The large-scale language model determines whether or not the response received by the third response receiving unit 109 contains information about unsafe behavior.

[0140] If the large-scale language model determines that the response received by the third response receiving unit 109 does not contain any information about unsafe behavior (S203: NO), the process proceeds to S201.

[0141] On the other hand, if the large-scale language model determines that the response received by the third response receiving unit 109 contains information about unsafe behavior (S203: YES), the process proceeds to S204. In S204, the fourth acquisition unit 110 acquires a question that guides the user to respond with measures to avoid unsafe behavior.

[0142] In S205, the fourth output unit 111 outputs the question acquired by the fourth acquisition unit 110. The question acquired by the fourth acquisition unit 110 is output to the user from an output device 15 such as a display, touch panel, and / or speaker. The fourth answer receiving unit 112 then terminates processing when it receives the answer to the question output by the fourth output unit 111. For example, the answer to the question output by the fourth output unit 111 is received from the user via an input device 14 such as a keyboard, touch panel, mouse, camera, and / or microphone. When terminating, the fourth output unit 111 may output a message such as, "Thank you for your hard work, this concludes the KYT."

[0143] Through the above processing, the information processing device 10 can perform KYT (Kiken Yochi Training) in a way that suits the user's situation and needs, such as when the user wants to perform KYT in a short amount of time. Furthermore, by starting KYT, for example, by acquiring questions that guide the user to answer questions about unsafe behavior, the information processing device 10 can perform KYT that focuses on key points, thereby increasing the user's risk awareness.

[0144] The embodiments described above are provided to facilitate understanding of this disclosure and are not intended to limit it. The flowcharts, modes, elements, arrangements, conditions, etc., described in the embodiments are not limited to those exemplified and can be modified as appropriate. Furthermore, configurations shown in different embodiments can be partially substituted or combined.

[0145] For example, although the above describes an information processing system 1 using an external device 30, the external device 30 is not an essential component for configuring the information processing system 1. Instead of the external device 30, a learning model stored in the storage unit 100 of the information processing device 10 may be used. By pre-training the learning model with the contents of the hazard prediction DB 100a, the learning model can generate questions that fall within the scope of the contents of the hazard prediction DB 100a. In addition, the questions acquired from each acquisition unit, such as the first acquisition unit 101, may be questions that have been pre-stored in the storage unit 100.

[0146] Furthermore, in this specification (including the claims), terms such as "first," "second," etc., are used simply to distinguish between two or more configurations, and are not necessarily intended to impose technical meanings such as temporal aspects or order on the subject. Therefore, for example, when the first acquisition unit 101 and the second acquisition unit 104 acquire information, it does not necessarily mean that the first acquisition unit 101 must acquire information before the second acquisition unit 104.

[0147] Although various embodiments have been described above with reference to the drawings, it goes without saying that this disclosure is not limited to such examples. It is clear to those skilled in the art that various modifications or alterations can be conceived within the scope of the claims, and these will naturally also fall within the technical scope of this disclosure. Furthermore, the components of the above embodiments may be combined in any way without departing from the spirit of the invention.

[0148] This application is based on a Japanese patent application (Patent Application No. 2025-009957) filed on January 23, 2025, the contents of which are incorporated by reference within this application.

[0149] According to one aspect of this disclosure, it is possible to provide technology that enables users to actively participate in training on hazard prediction, enhance their sensitivity to danger, and develop the ability to recognize unsafe behaviors on their own.

[0150] 1... Information processing system, 10... Information processing device, 11... Processor, 12... Memory device, 13... Communication IF, 14... Input device, 15... Output device, 20... Terminal device, 30... External device, 100... Storage unit, 100a... Risk prediction DB, 100b... User information DB, 100c... Conversation history DB, 101... First acquisition unit, 102... First output unit, 103... First response reception unit, 104... Second acquisition 105...Second output unit, 106...Second response receiving unit, 107...Third acquisition unit, 108...Third output unit, 109...Third response receiving unit, 110...Fourth acquisition unit, 111...Fourth output unit, 112...Fourth response receiving unit, 113...Fifth acquisition unit, 114...Fifth output unit, 115...Fifth response receiving unit, 116...Prompt output unit, 200...Storage unit, 201...UI unit, 201...UI unit, 202...Control unit

Claims

1. An information processing device for performing training on the prediction of accidents that may occur in a given environment, comprising: a first acquisition unit for acquiring questions that prompt the user to answer about the accident; a first output unit for outputting the questions acquired by the first acquisition unit; a first answer receiving unit for receiving answers to the questions output by the first output unit; a second acquisition unit for acquiring questions that inquire about the basis for the answer received by the first answer receiving unit if the answer received by the first answer receiving unit is related to the accident but does not include information about unsafe behavior that causes the accident; a second output unit for outputting the questions acquired by the second acquisition unit; a second answer receiving unit for receiving answers to the questions output by the second output unit; a third acquisition unit for acquiring questions that lead the user to answer about unsafe behavior if the answer received by the second answer receiving unit does not include information about unsafe behavior; a third output unit for outputting the questions acquired by the third acquisition unit; and a third answer receiving unit for receiving answers to the questions output by the third output unit. An information processing apparatus wherein, until the response received by the third response receiving unit contains information about the unsafe behavior, the third acquisition unit acquires questions that lead to the response regarding the unsafe behavior, the third output unit outputs the questions acquired by the third acquisition unit, and the third response receiving unit repeatedly receives the response to the questions output by the third output unit.

2. The information processing apparatus according to claim 1, comprising: a fourth acquisition unit that acquires a question that guides the user to provide a response regarding measures to prevent unsafe behavior if the response received by the second response receiving unit includes information regarding unsafe behavior; a fourth output unit that outputs the question acquired by the fourth acquisition unit; and a fourth response receiving unit that receives the response to the question output by the fourth output unit.

3. The information processing apparatus according to claim 1, comprising: a fourth acquisition unit that acquires questions that lead to a response indicating measures to prevent the unsafe behavior; a fourth output unit that outputs the questions acquired by the fourth acquisition unit; and a fourth answer receiving unit that receives the answer to the question output by the fourth output unit, wherein if the answer received by the third answer receiving unit includes information relating to the unsafe behavior, the fourth acquisition unit acquires questions that lead to a response indicating measures to prevent the unsafe behavior, the fourth output unit outputs the questions acquired by the fourth acquisition unit, and the fourth answer receiving unit receives the answer to the question output by the fourth output unit.

4. The information processing apparatus according to claim 1, comprising: a fifth acquisition unit that acquires a hint for a question acquired by the first acquisition unit if the answer received by the first answer receiving unit does not contain information relating to the accident; a fifth output unit that outputs the hint acquired by the fifth acquisition unit; and a fifth answer receiving unit that receives a response to the hint output by the fifth output unit, wherein the fifth acquisition unit acquires the hint for a question acquired by the first acquisition unit, the fifth output unit outputs the hint acquired by the fifth acquisition unit, and the fifth answer receiving unit receives a response to the hint output by the fifth output unit, and this process is repeated until the answer received by the fifth answer receiving unit contains information relating to the accident.

5. If the response received by the fifth response receiving unit includes information relating to the accident, the second acquisition unit acquires a question asking for the basis of the response received by the first response receiving unit, the second output unit outputs the question acquired by the second acquisition unit, and the second response receiving unit accepts the answer to the question output by the second output unit, as described in claim 4.

6. The information processing apparatus according to claim 1, wherein the questions acquired by at least one of the first acquisition unit, the second acquisition unit, and the third acquisition unit include at least one of a video and a still image relating to the accident.

7. An information processing apparatus according to claim 1, comprising: a prompt output unit that outputs information to a large language model including a prompt for generating at least one of the following: a question to elicit a response regarding the accident; a question to inquire about the basis for the response received by the first response receiving unit; and a question to guide the user to respond regarding the unsafe behavior, wherein if the large language model generates a question to elicit a response regarding the accident, the first acquisition unit acquires the question to elicit a response regarding the accident generated by the large language model; if the large language model generates a question to inquire about the basis for the response received by the first response receiving unit, the second acquisition unit acquires the question to inquire about the basis for the response received by the first response receiving unit generated by the large language model; and if the large language model generates a question to guide the user to respond regarding the unsafe behavior, the third acquisition unit acquires the question to guide the user to respond regarding the unsafe behavior generated by the large language model.

8. An information processing method for conducting training on the prediction of accidents that may occur in a given environment, comprising: a first acquisition step of acquiring a question to prompt the user to answer about the accident; a first output step of outputting the question acquired in the first acquisition step; a first answer receiving step of receiving the answer to the question output in the first output step; a second acquisition step of acquiring a question to inquire about the basis of the answer received in the first answer receiving step if the answer received in the first answer receiving step is related to the accident but does not include information about unsafe behavior that causes the accident; a second output step of outputting the question acquired in the second acquisition step; a second answer receiving step of receiving the answer to the question output in the second output step; a third acquisition step of acquiring a question to prompt the user to answer about unsafe behavior if the answer received in the second answer receiving step does not include information about unsafe behavior; a third output step of outputting the question acquired in the third acquisition step; and a third answer receiving step of receiving the answer to the question output in the third output step. An information processing method performed by an information processing device, wherein the third acquisition step acquires questions that lead to a response regarding the unsafe behavior until the response received by the third response receiving step includes information regarding the unsafe behavior, the third output step outputs the questions acquired by the third acquisition step, and the third response receiving step repeatedly accepts the response to the questions output by the third output step.

9. A program for training on predicting accidents that may occur in a given environment, comprising: a first acquisition step for acquiring a question to answer about the accident; a first output step for outputting the question acquired in the first acquisition step; a first answer receiving step for receiving the answer to the question output by the first output step; a second acquisition step for acquiring a question asking for the basis of the answer received by the first answer receiving step if the answer received by the first answer receiving step is related to the accident but does not include information about unsafe behavior that causes the accident; a second output step for outputting the question acquired in the second acquisition step; a second answer receiving step for receiving the answer to the question output by the second output step; a third acquisition step for acquiring a question to lead to answering about unsafe behavior if the answer received by the second answer receiving step does not include information about unsafe behavior; a third output step for outputting the question acquired in the third acquisition step; and a third answer receiving step for receiving the answer to the question output by the third output step. A program for a computer to run, wherein the third acquisition step acquires questions that lead to a response regarding the unsafe behavior until the response received by the third response receiving step includes information regarding the unsafe behavior, the third output step outputs the questions acquired by the third acquisition step, and the third response receiving step repeatedly accepts the response to the questions output by the third output step.

10. An information processing device for providing training to a user on predicting accidents that may occur in a predetermined environment, comprising: a first acquisition unit that acquires questions whose answers are unsafe actions that cause accidents in the predetermined environment; a first output unit that outputs the questions acquired by the first acquisition unit; and a first answer receiving unit that receives answers to the questions output by the first output unit, wherein the first acquisition unit acquires questions whose answers are unsafe actions, the first output unit outputs the questions acquired by the first acquisition unit, and the first answer receiving unit receives answers to the questions output by the first output unit, and this process is repeated until the answer to the question whose answer is an unsafe action received by the first answer receiving unit includes information about the unsafe action.

11. The information processing apparatus according to claim 10, comprising: a second acquisition unit that acquires a question whose answer is a measure to prevent the aforementioned unsafe behavior; a second output unit that outputs the question acquired by the second acquisition unit; and a second answer receiving unit that receives the answer to the question output by the second output unit, wherein if the answer to the question whose answer is the aforementioned unsafe behavior, received by the first answer receiving unit, includes information relating to the aforementioned unsafe behavior, the second acquisition unit acquires a question whose answer is a measure to prevent the aforementioned unsafe behavior, the second output unit outputs the question acquired by the second acquisition unit, and the second answer receiving unit receives the answer to the question output by the second output unit.

12. The information processing apparatus according to claim 10, wherein the questions acquired by the first acquisition unit include at least one of a video and a still image relating to the accident.

13. An information processing apparatus according to claim 10, comprising: a prompt output unit that outputs information to a large-scale language model including a prompt for generating at least one of the questions for which the answer is the unsafe behavior, wherein if the large-scale language model generates a question for which the answer is the unsafe behavior, the first acquisition unit acquires the question for which the answer is the unsafe behavior generated by the large-scale language model.

14. An information processing method for providing training to a user on predicting accidents that may occur in a predetermined environment, comprising: a first acquisition step of acquiring questions whose answers are unsafe actions that cause accidents in the predetermined environment; a first output step of outputting the questions acquired in the first acquisition step; and a first answer receiving step of receiving answers to the questions output by the first output step, wherein the first acquisition step acquires questions whose answers are unsafe actions, the first output step outputs the questions acquired in the first acquisition step, and the first answer receiving step receives answers to the questions output by the first output step, and this process is repeated until the answer to the question whose answer is an unsafe action received by the first answer receiving step includes information about the unsafe action.

15. A program for a computer to run, which provides training to a user on predicting accidents that may occur in a given environment, comprising: a first acquisition step of acquiring questions whose answers are unsafe actions that cause accidents in the given environment; a first output step of outputting the questions acquired in the first acquisition step; and a first answer receiving step of receiving answers to the questions output by the first output step, wherein the first acquisition step acquires questions whose answers are unsafe actions, the first output step outputs the questions acquired in the first acquisition step, and the first answer receiving step receives answers to the questions output by the first output step, and this process is repeated until the answer to the question whose answer is an unsafe action received by the first answer receiving step includes information about the unsafe action.

Images