Information provision device, information provision method, and information provision program

Abstract

An information provision device (10) is provided with a discussion unit (33) and a judgment unit (34). The discussion unit (33) holds a discussion on a legal case using a plurality of lawyer accounts that make different claims and that are set in a language model which generates answers to input prompts, the language model being generated by learning data stored in a database (20a) relating to legal cases. The judgment unit (34) judges the merits of the claims on the basis of the results of the discussion by the discussion unit (33), using a judge account set in the language model.

Description

Information providing device, information providing method, and information providing program

[0001] The disclosed embodiments relate to an information providing device, an information providing method, and an information providing program.

[0002] 2. Description of the Related Art Conventionally, there is known a system that uses a language model to generate an answer to a question input by a user (see, for example, Patent Document 1).

[0003] Special Publication No. 2022-503838

[0004] While conventional technology generates answers based on news articles, it does not take into consideration the use of language models specialized for specific fields.

[0005] The present invention has been made in view of the above, and has as its object to effectively utilize a language model specialized for a specific field.

[0006] An information providing device according to one aspect of the embodiment includes a language model generated by learning data stored in a database relating to legal cases, a discussion section in which legal cases are discussed using a plurality of lawyer accounts each making different claims set in the language model that generates answers to input prompts, and a judging section that judges the merits of the claims based on the results of the discussion by the discussion section using a judge account set in the language model.

[0007] According to one aspect of the embodiment, it is possible to effectively utilize a language model specialized for a specific field.

[0008] FIG. 1 is a diagram illustrating an overview of an information providing device according to a first embodiment. FIG. 2 is a functional block diagram illustrating an example of the configuration of an information providing device according to the first embodiment. FIG. 3 is a flowchart illustrating a processing procedure executed by the information providing device according to the embodiment. FIG. 4 is a flowchart illustrating a processing procedure executed by the information providing device according to the embodiment. FIG. 5 is a diagram schematically illustrating an example of a computer hardware configuration functioning as an information providing device. FIG. 6 is a diagram illustrating a processing flow of an information providing device according to a second embodiment. FIG. 7 is a diagram illustrating a configuration of an information providing system including an information providing device according to a third embodiment. FIG. 8 is a diagram illustrating a discussion. FIG. 9 is a diagram illustrating an overview of an information providing device according to a fourth embodiment. FIG. 10 is a diagram illustrating an overview of an information providing device according to a fifth embodiment. FIG. 11 is a diagram illustrating a discussion. FIG. 12 is a flowchart illustrating a processing procedure for a discussion. FIG. 13 is a diagram illustrating an overview of an information providing device according to a sixth embodiment. FIG. 14 is a diagram illustrating an overview of a discussion according to the sixth embodiment. FIG. 15 is a diagram illustrating an example of a prompt according to the sixth embodiment. FIG. 16 is a flowchart illustrating a processing procedure executed by an information providing device according to the sixth embodiment.

[0009] The present invention will be described below through embodiments, but the following embodiments do not limit the scope of the invention as claimed. Furthermore, not all of the combinations of features described in the embodiments are necessarily essential to the solution of the invention.

[0010] (1. First Embodiment) The processing flow of an information providing device according to a first embodiment will be described with reference to Fig. 1. Fig. 1 is a diagram illustrating an overview of an information providing device according to a first embodiment. Fig. 1 shows the configuration of an information providing system 1 including an information providing device 10 according to the first embodiment.

[0011] 1, the information provision system 1 includes an information provision device 10 and a legal case database 20a. The legal case database 20a is a database that stores data related to various laws. For example, the legal case database 20a stores data related to various laws related to civil cases and criminal cases, as well as data related to past court decisions. The legal case database 20a corresponds to an example of a database according to an embodiment.

[0012] The information providing device 10 is a device to which various devices, including the legal case database 20a, are plugged in. In the information providing system 1, which includes various devices such as various servers, vehicles, and smartphones, the information providing device 10 realizes the highest level of artificial intelligence by being plugged in with each device.

[0013] For example, in the information provision system 1, the information provision device 10 and each device including the legal case database 20a are built with semiconductor chips manufactured by the same manufacturer. Each of these semiconductor chips may perform machine learning or deep learning.

[0014] Furthermore, the semiconductor chips used in the information providing device 10 and the legal case database 20a are chips of a size appropriate for their respective housings. For example, if the chip size of the information providing device 10 is assumed to be XL, then L size chips are used for servers and vehicles, and S or M size chips are used for user terminals. Furthermore, for terminals with housings even smaller than the user terminals, SS size chips are used and incorporated into a system on a chip (SoC) to form a single chip. Note that the XL size has 200 tiles of semiconductor chips, L size 50 tiles, M size 20 tiles, S size 10 tiles, and SS size 2 tiles, but the number of tiles for each size is merely an example and is not limited to the above. If accessories such as a camera or microphone are to be attached to the user terminal, a dedicated chip for the accessories, separate from the semiconductor chip, may be installed in the empty space near the semiconductor chip in the SoC.

[0015] In this way, the information provision system 1 can be kept secure by using semiconductor chips from the same manufacturer for various devices connected to the information provision device 10. In other words, the information provision system 1 can highly accurately avoid hacking, virus infection, and deep fakes, and can ensure privacy between the information provision device 10 and data source providers including the legal case database 20a.

[0016] In this embodiment, the information providing device 10 learns a model based on a data source acquired from the legal case database 20a. The model is a language model. An example of a known language model is ChatGPT by OpenAI (see https: / / openai.com / blog / chatgpt). The language model may be a language model using a neural network, such as a Generative Adversarial Network (GAN) or a Variational Autoencoder (VAE).

[0017] In this embodiment, the language model generates a text response to a query text (hereinafter, a prompt) entered by a user. The language model is trained in advance using a data source. The language model may be trained using known machine learning techniques.

[0018] This allows the language model to generate answers based on the data source. In this embodiment, the information providing device 10 plugs in to the legal case database 20a (step S1) and trains the language model based on the data stored in the legal case database 20a (step S2).

[0019] In this way, the information providing device 10 trains the language model with various data stored in the legal case database 20a. As a result, the language model generates a response to an input prompt based on the learned legal cases. In other words, the information providing device 10 can generate a language model specialized for the legal field.

[0020] In this embodiment, a virtual law firm is designed using a language model specialized for the legal field. Specifically, in this embodiment, the accounts (roles) of "Lawyers A to C," "Prosecutor," and "Judge System" are set in advance by system prompts. "Lawyers A to C" are accounts that make different arguments. "Prosecutor" is an account that determines the topic of discussion between Lawyers A to C. Note that the "Prosecutor" may also play the role of plaintiff and make the plaintiff's arguments. "Judge System" is an account that judges the merits of the arguments of Lawyers A to C.

[0021] In this embodiment, the arguments made by each of the lawyers A to C accounts are input as prompts into the language model, and the system prompts are set so that the other lawyer accounts can debate the input prompts.

[0022] This setting allows "Lawyers A to C" to spontaneously hold a discussion using the language model. For example, as shown in FIG. 1, the "Prosecutor" sets the topic of the discussion (Step S3). The topic here corresponds to the content of a consultation with the corporate lawyer. The topic may be related to a lawsuit currently being handled by the company, or may be a fictitious topic.

[0023] "Lawyers A to C" debate along the theme set by the "Prosecutor" (Step S4). For example, "Lawyers A to C" each make different claims and debate the logic supporting those claims. For example, "Lawyers A to C" debate each other by developing logic supporting their own claims based on relevant laws and past precedents.

[0024] Thereafter, the "judging system" makes a judgment based on the content of the debate (step S5). For example, when a predetermined termination condition is met, such as when the debate reaches a deadlock, the "judging system" judges which judgment system's argument is superior. For example, the "judging system" checks the relevant laws and past precedents to judge the merits of each lawyer's argument.

[0025] The "judgment system" then judges the arguments and logic of each lawyer based on past precedents, etc., and selects the best argument. In this way, the information providing device 10 trains a language model based on the data acquired from the legal case database 20a.

[0026] The information providing device 10 then uses a language model to hold a debate among the pre-set accounts and derives a conclusion from the debate content. For example, the information providing device 10 can hold such debates more times than in debates between actual lawyers. For example, it is possible to hold multiple debates on the same topic, and have the claims of the winner of one debate and the claims of the winner of the second debate debate against each other.

[0027] In other words, the information providing device 10 can refine each argument by repeatedly holding such discussions, which enables the information providing device 10 to make more meaningful arguments than an actual lawyer.

[0028] Therefore, the information providing device 10 according to the embodiment can effectively utilize a language model specialized for a specific field.

[0029] The configuration of the information providing device 10 will be described with reference to Fig. 2. Fig. 2 is a functional block diagram showing an example of the configuration of the information providing device according to the first embodiment.

[0030] As shown in FIG. 2 , the information providing device 10 includes a communication unit 11 , a storage unit 12 , and a control unit 13 .

[0031] The communication unit 11 transmits and receives information to and from the legal case database 20a and the like via the network.

[0032] The storage unit 12 is realized by, for example, a semiconductor memory element such as a random access memory (RAM) or a flash memory, or a storage device such as a hard disk drive (HDD), a solid state drive (SSD), or an optical disk. The storage unit 12 stores various programs and various data. The storage unit 12 stores language model information 121.

[0033] The language model information 121 is information about a language model. For example, the language model information 121 is information about parameters for constructing a language model. Such parameters include, for example, weights and biases of a neural network.

[0034] The control unit 13 is a controller and includes, for example, a microcomputer having a CPU (Central Processing Unit), ROM (Read Only Memory), RAM, input / output ports, and various other circuits. The control unit 13 may also be configured with hardware such as integrated circuits, including an ASIC (Application Specific Integrated Circuit) and an FPGA (Field Programmable Gate Array). The control unit 13 includes an acquisition unit 31, a generation unit 32, a discussion unit 33, and a judging unit 34.

[0035] The acquisition unit 31 acquires data from a data source provider. For example, the acquisition unit 31 acquires data related to legal cases from the legal case database 20a (see FIG. 1). For example, the acquisition unit 31 acquires data every time the legal case database 20a is updated.

[0036] The generation unit 32 generates a language model. Specifically, the generation unit 32 learns the language model based on a data source related to legal cases. The generation unit 32 also updates the parameters of the model through learning. The generation unit 32 may regenerate the model at each learning timing, or may reflect in the model a data source that is different from the previous learning.

[0037] The discussion unit 33 uses a language model generated by learning data stored in a database related to legal cases, which generates answers to input prompts, to hold discussions between multiple lawyer accounts each with different arguments.

[0038] First, the discussion unit 33 generates multiple lawyer accounts, prosecutor accounts, and judge system accounts using system prompts from the language model. For example, each lawyer account is set to make a different argument and to hold a discussion to support its own argument.

[0039] The prosecutor account is set to set the topic of the debate. The prosecutor account may also be set to debate with each lawyer account as the plaintiff. The judge system account is set to judge the results of the debate between each lawyer account, or the results of the debate between each lawyer account and the prosecutor account.

[0040] The discussion unit 33 sets these accounts and then sets the discussion theme. For example, the discussion theme may be one arbitrarily set by the user. In this case, for example, data when the user requests a lawyer may be input into the language model as the theme.

[0041] The discussion unit 33 may also use the consultation content set by the user as the topic of discussion. For example, the consultation content may be a consultation about whether or not a client company should file a lawsuit, or a consultation about the content of a contract that is advantageous to the client company.

[0042] The judge unit 34 uses a language model to judge the arguments based on the results of the discussion by the discussion unit 33. The judge unit 34 corresponds to a judge system account. The judge unit 34 judges the arguments of each lawyer account when the discussion between the lawyer accounts reaches a conclusion.

[0043] For example, the judging unit 34 inputs each claim made by each lawyer account as a prompt into the language model, and judges which of these claims is superior.

[0044] More specifically, the judging unit 34 judges the best argument by comparing it with relevant laws and past precedents using a language model. Furthermore, if the topic of the discussion is whether or not to file a lawsuit, the judging unit 34 determines the conclusion based on the results of the discussion by each lawyer's account.

[0045] Furthermore, when the topic of the discussion is the content of the contract, the judging unit 34 determines the draft of the contract based on the results of the discussion by each lawyer account. The results by the judging unit 34 are then output to the user terminal of the user (not shown).

[0046] Next, a processing procedure executed by the information providing device 10 according to the embodiment will be described with reference to Fig. 3 and Fig. 4. Fig. 3 and Fig. 4 are flowcharts showing a processing procedure executed by the information providing device 10 according to the embodiment.

[0047] First, the procedure for generating a language model will be described with reference to Fig. 3. As shown in Fig. 3, the information providing device 10 first acquires data related to legal cases from the legal case database 20a (step S101). Next, the information providing device 10 generates a language model based on the data acquired from the legal case database 20a (step S102), and then ends the process.

[0048] Next, the discussion process using the language model will be described with reference to Fig. 4. As shown in Fig. 4, the information providing device 10 first determines a topic for the discussion (step S111). Then, the information providing device 10 starts a discussion between the lawyer accounts (step S112).

[0049] Next, the information providing device 10 determines whether the discussion between each lawyer account has ended (step S113), and if it determines that the discussion has not ended (step S113; No), it repeatedly executes the processing of step S113 until the discussion has ended.

[0050] Furthermore, when the information providing device 10 determines that the discussion between the lawyer accounts has ended (step S113; Yes), it judges each argument based on the discussion results (step S114) and ends the process.

[0051] According to this embodiment, by setting up a virtual law firm using a language model trained based on data stored in the legal case database 20a, it is possible to effectively utilize language models specialized for specific fields.

[0052] 5 is a diagram schematically illustrating an example of a computer hardware configuration that functions as an information providing device. A program installed on the computer 1200 can cause the computer 1200 to function as one or more "parts" of the device according to the present embodiment, or can cause the computer 1200 to perform operations associated with the device according to the present embodiment or one or more "parts" thereof, and / or can cause the computer 1200 to perform a process according to the present embodiment or steps of the process. Such a program can be executed by the CPU 1212 to cause the computer 1200 to perform specific operations associated with some or all of the blocks in the flowcharts and block diagrams described herein.

[0053] The computer 1200 according to this embodiment includes a CPU 1212, a RAM 1214, and a graphics controller 1216, which are interconnected by a host controller 1210. The computer 1200 also includes input / output units such as a communication interface 1222, a storage device 1224, a DVD drive, and an IC card drive, which are connected to the host controller 1210 via an input / output controller 1220. The DVD drive may be a DVD-ROM drive, a DVD-RAM drive, or the like. The storage device 1224 may be a hard disk drive, a solid state drive, or the like. The computer 1200 also includes input / output units such as a ROM 1230 and a keyboard, which are connected to the input / output controller 1220 via an input / output chip 1240.

[0054] The CPU 1212 operates according to programs stored in the ROM 1230 and the RAM 1214, thereby controlling each unit. The graphics controller 1216 acquires image data generated by the CPU 1212 into a frame buffer or the like provided in the RAM 1214 or into the graphics controller 1216 itself, and causes the image data to be displayed on the display device 1218.

[0055] The communication interface 1222 communicates with other electronic devices via a network. The storage device 1224 stores programs and data used by the CPU 1212 in the computer 1200. The DVD drive reads programs or data from a DVD-ROM or the like and provides them to the storage device 1224. The IC card drive reads programs and data from an IC card and / or writes programs and data to an IC card.

[0056] The ROM 1230 stores therein a boot program or the like that is executed by the computer 1200 upon activation, and / or programs that depend on the hardware of the computer 1200. The input / output chip 1240 may also connect various input / output units to the input / output controller 1220 via a USB port, a parallel port, a serial port, a keyboard port, a mouse port, etc.

[0057] The programs are provided by a computer-readable storage medium such as a DVD-ROM or an IC card. The programs are read from the computer-readable storage medium, installed in the storage device 1224, RAM 1214, or ROM 1230, which are also examples of computer-readable storage media, and executed by the CPU 1212. Information processing described in these programs is read by the computer 1200, and brings about cooperation between the programs and the various types of hardware resources described above. An apparatus or a method may be configured by implementing operations or processing of information in accordance with the use of the computer 1200.

[0058] For example, when communication is performed between computer 1200 and an external device, CPU 1212 may execute a communication program loaded into RAM 1214 and instruct communication interface 1222 to perform communication processing based on the processing described in the communication program. Under the control of CPU 1212, communication interface 1222 reads transmission data stored in a transmission buffer area provided in RAM 1214, storage device 1224, a DVD-ROM, or a recording medium such as an IC card, and transmits the read transmission data to a network, or writes received data received from the network to a reception buffer area or the like provided on the recording medium.

[0059] Furthermore, the CPU 1212 may cause all or a necessary portion of a file or database stored in an external recording medium such as the storage device 1224, a DVD drive (DVD-ROM), an IC card, etc. to be read into the RAM 1214, and may perform various types of processing on the data on the RAM 1214. The CPU 1212 may then write back the processed data to the external recording medium.

[0060] Various types of information, such as various types of programs, data, tables, and databases, may be stored on the recording medium and may undergo information processing. The CPU 1212 may perform various types of processing on data read from the RAM 1214, including various types of operations, information processing, conditional judgment, conditional branching, unconditional branching, information search / replacement, etc., as described throughout this disclosure and specified by the instruction sequences of the programs, and write the results back to the RAM 1214. The CPU 1212 may also search for information in a file, database, etc. on the recording medium. For example, if multiple entries, each having an attribute value of a first attribute associated with an attribute value of a second attribute, are stored on the recording medium, the CPU 1212 may search for an entry whose attribute value of the first attribute matches a specified condition from among the multiple entries, read the attribute value of the second attribute stored in the entry, and thereby obtain the attribute value of the second attribute associated with the first attribute that satisfies a predetermined condition.

[0061] The above-described programs or software modules may be stored in a computer-readable storage medium on or near the computer 1200. A recording medium such as a hard disk or RAM provided in a server system connected to a dedicated communication network or the Internet can also be used as a computer-readable storage medium, thereby providing the programs to the computer 1200 via the network.

[0062] The blocks in the flowcharts and block diagrams in the present embodiments may represent stages of a process in which an operation is performed or "parts" of a device responsible for performing the operation. Particular stages and "parts" may be implemented by dedicated circuitry, programmable circuitry provided with computer-readable instructions stored on a computer-readable storage medium, and / or a processor provided with computer-readable instructions stored on a computer-readable storage medium. The dedicated circuitry may include digital and / or analog hardware circuits, and may include integrated circuits (ICs) and / or discrete circuits. The programmable circuitry may include reconfigurable hardware circuits, such as field programmable gate arrays (FPGAs) and programmable logic arrays (PLAs), including AND, OR, XOR, NAND, NOR, and other logical operations, flip-flops, registers, and memory elements.

[0063] A computer-readable storage medium may include any tangible device capable of storing instructions that are executed by an appropriate device, such that a computer-readable storage medium having instructions stored thereon comprises an article of manufacture, including instructions that can be executed to create means for performing the operations specified in the flowcharts or block diagrams. Examples of computer-readable storage media may include electronic, magnetic, optical, electromagnetic, and semiconductor storage media. More specific examples of computer-readable storage media may include floppy disks, diskettes, hard disks, random access memories (RAMs), read-only memories (ROMs), erasable programmable read-only memories (EPROMs or flash memories), electrically erasable programmable read-only memories (EEPROMs), static random access memories (SRAMs), compact disc read-only memories (CD-ROMs), digital versatile discs (DVDs), Blu-ray discs, memory sticks, integrated circuit cards, and the like.

[0064] The computer readable instructions may include either assembler instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state-setting data, or source or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk®, JAVA®, C++, etc., and conventional procedural programming languages ​​such as the “C” programming language or similar programming languages.

[0065] Computer-readable instructions may be provided locally or over a local area network (LAN), a wide area network (WAN) such as the Internet, to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus, or programmable circuitry, such that the processor or programmable circuitry executes the computer-readable instructions to generate means for performing the operations specified in the flowcharts or block diagrams. Examples of processors include computer processors, processing units, microprocessors, digital signal processors, controllers, microcontrollers, etc.

[0066] (2. Second Embodiment) Next, a second embodiment will be described. Note that, below, explanations of parts that overlap with the first embodiment will be omitted. FIG. 6 explains the processing flow of the information providing device according to the second embodiment. As shown in FIG. 6, the information providing system 1 according to the second embodiment includes an information providing device 10 and an accounting matter database 20b. The accounting matter database 20b is a database that stores data related to various laws. For example, the accounting matter database 20b stores data related to various accounting laws and past legal precedents.

[0067] The information providing device 10 is a device to which various devices, including the accounting matter database 20b, are plugged in. The various devices include various types of servers, vehicles, smartphones, and the like.

[0068] In this embodiment, the information providing device 10 learns a model based on a data source acquired from the accounting matter database 20b. The information providing device 10 plugs in to the accounting matter database 20b (step S1-1) and learns a language model based on the data stored in the accounting matter database 20b (step S2-1).

[0069] In this way, the information providing device 10 trains the language model with various data stored in the accounting case database 20b. As a result, the language model generates a response to an input prompt based on the learned legal cases. In other words, the information providing device 10 can generate a language model specialized for the legal field.

[0070] A virtual accounting firm is designed using a language model specialized for the legal field. Specifically, the accounts (roles) of "Accountants A to C," "National Tax Agency," and "Judge System" are set up in advance using system prompts. "Accountants A to C" are accounts that make different claims. "National Tax Agency" is an account that participates in discussions between Accountants A to C from the perspective of the National Tax Agency. As will be described later, "National Tax Agency" may also decide the topic of discussions between Accountants A to C. "Judge System" is an account that judges the merits of the claims of Accountants A to C.

[0071] The assertions made by each of the accountant accounts A to C are input as prompts into the language model, and the other accountant accounts are set as system prompts to discuss the input prompts.

[0072] This setting allows "Accountants A to C" to spontaneously hold discussions using the language model. For example, as shown in FIG. 6, the "National Tax Agency" sets the topic of the discussion (step S3-1). The topic here corresponds to the content of a consultation that a company wishes to have with its accountant. The topic may be related to an accounting matter that the company is currently facing, or it may be a fictitious topic.

[0073] "Accountants A to C" hold a debate based on the theme set by the "National Tax Agency" (Step S4-1). For example, "Accountants A to C" each make different claims and debate the logic that supports those claims. For example, "Accountants A to C" debate each other by developing logic that supports their own claims based on relevant laws and past precedents.

[0074] The "judging system" then makes a judgment based on the content of the discussion (step S5-1). For example, when a predetermined termination condition is met, such as when the discussion reaches a deadlock, the "judging system" judges which accountant's argument is superior. For example, the "judging system" compares relevant laws and past legal precedents to judge the merits of each accountant's argument.

[0075] The "judging system" then judges the arguments and logic of each accountant based on past precedents, etc., and selects the best argument. In this way, the information providing device 10 learns a language model based on the data acquired from the accounting case database 20b.

[0076] The information providing device 10 then uses a language model to hold a debate among the pre-set accounts and derives a conclusion from the debate content. For example, the information providing device 10 can hold such debates more times than in debates between actual accountants. For example, it is possible to hold multiple debates on the same topic, and have the claims of the winner of one debate and the claims of the winner of the second debate debate against each other.

[0077] More specifically, the information providing device 10 uses a language model that is generated by learning data stored in a database related to accounting cases and generates answers to input prompts, and holds discussions between multiple accountant accounts, each with different arguments.

[0078] First, the information providing device 10 generates multiple accountant accounts, a tax bureau account, and a judge system account using system prompts from a language model. For example, each accountant account is set to make a different claim and to engage in discussions to support its own claim.

[0079] The National Tax Agency account is set to set the discussion topic. The National Tax Agency account can also be set to participate in the discussions of each accountant account from the perspective of the National Tax Agency. The judge system account is set to judge the results of the discussions by each accountant account, or the results of the discussions by each accountant account and the National Tax Agency account.

[0080] The information providing device 10 sets these accounts and then sets a discussion topic. For example, the discussion topic may be arbitrarily set by the user. In this case, for example, data that the user requests from an accountant may be input into the language model as a topic.

[0081] The information providing device 10 may also set the consultation content set by the user as the topic of discussion.

[0082] The information providing device 10 judges the assertions based on the discussion results using a language model. When the discussion between the accountant accounts reaches a climax, the information providing device 10 judges the assertions of each accountant account.

[0083] For example, the information providing device 10 inputs each claim made by each accountant account as a prompt into the language model, and judges which of these claims is superior.

[0084] More specifically, the information providing device 10 judges the best argument by comparing it with relevant laws and past precedents using a language model. The judgment result by the information providing device 10 is then output to a user terminal (not shown) of the user.

[0085] In other words, the information providing device 10 can refine each argument through repeated discussions, thereby enabling the information providing device 10 to make more meaningful arguments than an actual accountant.

[0086] Therefore, the information providing device 10 according to the second embodiment can effectively utilize a language model specialized for a specific field.

[0087] (3. Third Embodiment) Next, a third embodiment will be described. Fig. 7 is a diagram illustrating an overview of an information providing device 10 according to the third embodiment. Fig. 7 shows the configuration of an information providing system 1 including the information providing device 10 according to the third embodiment.

[0088] 7, the information provision system 1 includes an information provision device 10 and a medical case database 20c. The medical case database 20c is a database that stores various medical data. For example, the medical case database 20c stores medical papers and knowledge, various medical records, various data related to pharmaceuticals, various data related to insurance, various data related to medical equipment, and the like. The information provision device 10 plugs in to the medical case database 20c (step S1-2) and learns a language model based on the data stored in the medical case database 20c (step S2-2).

[0089] In this way, the information providing device 10 trains the language model with various data stored in the medical case database 20c. As a result, the language model generates a response to an input prompt based on the medical data that it has learned. In other words, the information providing device 10 can generate a language model specialized for the medical field.

[0090] In this embodiment, a language model specialized for the medical field is used to provide advice on medical treatment for patients. Specifically, in this embodiment, accounts (roles) of "Doctor," "Patient," "Family," "Experts A to C," and "Judge System" are set in advance using system prompts from the language model.

[0091] In this embodiment, accounts include bot accounts and human accounts. A bot account is a language model that can set system prompts. A bot account outputs answers using a language model. On the other hand, a human account is a human being. In other words, a human account outputs answers according to human input.

[0092] "Doctor," "Experts A to C," "Judge System," "Patient," and "Family" in Figure 1 are high-level accounts. High-level accounts include at least either bot accounts or human accounts. White circles in Figure 7 represent bot accounts. Black circles in Figure 1 represent human accounts.

[0093] Hereinafter, bot accounts or human accounts included in higher-level accounts may be referred to as lower-level accounts. Also, when it is clear from the preceding and following explanations or drawings, or when there is no need to distinguish between higher-level and lower-level accounts, the terms "higher level" and "lower level" will be omitted and the term "account" will be used.

[0094] The following explanation assumes that all accounts are bot accounts. The configuration shown in Figure 7 can be realized by replacing bot accounts with human accounts as appropriate, or by adding human accounts as subordinate accounts to higher-level accounts.

[0095] "Doctor" is a higher-level account corresponding to, for example, the patient's doctor. While Figure 7 shows a case where "Doctor" is one account, in reality, multiple accounts are set. Note that the multiple accounts may be, for example, doctors with different specialties (internal medicine, surgery, etc.).

[0096] "Patient" is an account corresponding to a patient. "Family" is an account corresponding to a patient's family. "Experts A to C" are expert accounts involved in the treatment of a patient.

[0097] Examples of expert accounts include medical researchers, hospitals, nurses, pharmacists, pharmaceutical manufacturers, and medical device manufacturers. The "judge system" is an account that judges the results of discussions between each account. For example, the judge system judges the merits of each account's arguments and proposes the content to an actual doctor (e.g., the patient's primary care physician).

[0098] In this embodiment, the assertions made by each of the multiple doctor accounts are input as prompts into the language model, and the system prompts are set so that other doctor accounts can discuss the input prompts.

[0099] With this setting, each doctor account can spontaneously hold a discussion using the language model. As shown in Fig. 7, the information providing device 10 uses the language model to hold a discussion regarding the treatment plan for the patient by each account (step S3-2).

[0100] For example, a discussion begins by first entering various data such as the patient's age and gender, as well as the patient's medical record, as prompts. More specifically, doctor accounts are further divided into multiple accounts, such as Doctor 1, Doctor 2, and Doctor 3, with Doctor 1 distinguished by personality traits, such as conservative, Doctor 2 proactive, and Doctor 3 data-oriented. Each account uses a language model to debate. Doctor 1 further subdivides into subaccounts, such as Doctor 1-1, 1-2, and 1-3, while Doctor 2 further subdivides into Doctor 2-1, Doctor 2-2, and Doctor 2-3, and reaches a conclusion as Doctor 1 or Doctor 2. The virtual experts engage in billions of debates per second. This enables verification and problem solving from all angles, instantly proposing highly accurate diagnoses and treatments based on past medical data. Alternatively, Doctor 1 may perform multimodal data analysis and propose the results of an integrated analysis of various types of medical data, such as genetic data, blood data, MRIs, and X-ray images. Doctor 2 may use a real-time learning model to obtain new data from medical facilities around the world in real time and propose constantly incorporating the latest medical information and research findings into treatment. Doctor 3 may use advanced simulation capabilities to virtually simulate the progression of disease and the effectiveness of treatments, proposing the most appropriate treatment and its outcomes in advance. The doctor account may then determine a treatment plan based on a comprehensive assessment of the outputs from Doctors 1, 2, and 3. Other ideas that the doctor account may adopt include promoting personalized medicine by proposing individually optimized treatment plans based on each patient's genetic information, lifestyle, and other background information. Alternatively, by linking with 3D bioprinting, the doctor may identify the optimal organ or tissue structure for each patient and propose using 3D bioprinting technology to create individually customized artificial organs or tissues.

[0101] For example, a family account may use multiple accounts set up within the family to reach a conclusion on a medical policy for the family. Each expert account may hold discussions with multiple accounts in each field and reach a conclusion on a medical policy for each field. The doctor account may then hold a discussion on a treatment policy for the patient based on the discussions in each field. The final discussion may be held by all accounts, including each account.

[0102] For example, the doctor accounts discuss the treatment method, cost, risk, cure rate, and the method and timing of informing the patient and their family as a treatment policy for the patient. The "judgment system" then makes a judgment based on the content of the discussion (step S4-2). For example, the "judgment system" is made up of multiple accounts. The multiple accounts that make up the judgment system discuss the content of the discussions between the doctor accounts and other accounts and judge the final medical policy. For example, when a predetermined termination condition is met, such as when the discussion reaches a deadlock, the judgment system determines the most appropriate medical policy by comparing the arguments of each account with the data stored in the medical case database 20c.

[0103] The "judgment system" then proposes the decided medical policy to, for example, the patient's doctor. In other words, by using a language model, the information provision system 1 can simulate a huge discussion regarding medical policies between doctors and others. In addition, because the language model is trained in advance with a huge amount of data related to medical cases, it can propose medical policies based on various past findings.

[0104] In this way, the information providing device 10 according to the embodiment generates a language model based on various knowledge registered in the medical case database 20c, conducts a discussion using the language model, and then makes a proposal regarding a medical policy.

[0105] Therefore, the information providing device 10 according to the embodiment can effectively utilize a language model specialized for a specific field.

[0106] The information providing device 10 conducts a discussion using a language model by the following steps 1 to 5.

[0107] Step 1: Setting up multi-functional virtual experts. There are hundreds of sets of virtual experts in different medical fields. For example, specific characters such as "Neuro-Expert A," "Neuro-Expert B," and "Neuro-Expert C" are set up for neurology, and "Cardiologist A," "Cardiologist B," and "Cardiologist C" are set up for cardiology.

[0108] Step 2: Accelerating information exchange These experts exchange information based on patient data and debate billions of times per second, enabling them to instantly make decisions about the patient's condition, symptoms, and treatment options.

[0109] Step 3: Data Handling The virtual experts share and discuss various types of information, such as MRI data, genetic data, and blood data, allowing them to quickly arrive at optimal solutions from multiple perspectives.

[0110] Step 4: Self-learning and updating If new knowledge, information, or treatment methods are discovered during the discussion, they are immediately reflected in the database and used in future discussions, allowing for diagnosis and treatment based on the latest medical information.

[0111] Step 5: Forming a consensus. Discussion leads to the formation of a final "consensus" through the clash of different opinions and perspectives. This allows information and knowledge from multiple sources to be comprehensively incorporated when making a diagnosis or treatment proposal for the patient. There may be multiple treatment proposals.

[0112] The above has been explained about the doctor account, but each of the patient account, family account, and expert A to C accounts also has its own character set, and each account is further divided into various characters, and by each holding discussions, a consensus is formed as a patient, a family, and as experts A to C.

[0113] The information providing device 10 judges the results of the discussion by the information providing device 10 using a judge account set in the language model and proposes a treatment plan. For example, the information providing device 10 makes a judgment when a predetermined termination condition is met, such as when the discussion by the information providing device 10 reaches a deadlock.

[0114] For example, the information providing device 10 refers to various data registered in the medical case database 20c, judges the medical policy that is estimated to be the most appropriate among the medical policies discussed by each doctor account, and proposes that medical policy. Alternatively, a medical policy may be proposed by taking into account various consensuses, such as the patient's consensus (wanting to avoid hospitalization as much as possible), the family's consensus (wanting to leave scars as little as possible), and the Expert A's consensus (how much insurance will pay).

[0115] [Discussion when a human account is included] Up until now, discussion has been explained assuming that the lower-level accounts are bot accounts. Here, we will explain discussion when the lower-level accounts include human accounts. In this case, the processing content of steps 1 and 2 above by the information providing device 10 changes.

[0116] First, in step 1, the information providing device 10 changes the configuration of at least some of the higher-level accounts so that they include human accounts as lower-level accounts. For example, the information providing device 10 replaces bot accounts with human accounts or adds new human accounts. A user, i.e., a real person, is associated with each human account.

[0117] Next, in step 2, the information providing device 10 presents the text output as an answer by each account to the user associated with the human account. Then, the information providing device 10 accepts text input from the user. The answer text output by each account and the action of outputting text as an answer are called "statements." "Statements" include text input by the user.

[0118] Bot accounts treat utterances from human accounts in the same way as bot accounts. For example, bot accounts input utterances from other bot accounts and human accounts as prompts into their language models and then output answers.

[0119] The processing of the information providing device 10 will be described using Figure 8. Figure 8 is a diagram for explaining a discussion. As shown in Figure 8, the information providing device 10 conducts a discussion as the phases progress. Here, it is assumed that Doctor 1, which is a higher-level account, is made up of three lower-level accounts: Doctor 1-1, Doctor 1-2, and Doctor 1-3.

[0120] Doctor 1-1 is a bot account for a surgeon. For example, the language model for Doctor 1-1 outputs text that prioritizes surgical treatment methods.

[0121] Doctor 1-2 is a bot account of an internist. For example, the language model of Doctor 1-2 outputs text that prioritizes medical treatment methods.

[0122] The substance of a bot account is a language model, and discussions are realized by the information providing device 10 repeatedly inputting prompts to the language model. The prompts are, for example, statements made by the language model in the previous phase. The type of data to be emphasized in a statement may be set by a system prompt. As a result, each bot account tends to respond differently.

[0123] Doctors 1-3 are human accounts. A user is associated with each of the doctors 1-3. The personalities of the doctors 1-3 (what they prioritize) can be anything.

[0124] 8, in phase 1, doctor 1-1 says, "I think it would be better to perform an abdominal surgery as soon as possible." At this point, the information providing device 10 progresses the phase.

[0125] In Phase 2, Doctor 1-2 inputs Doctor 1-1's statement in Phase 1 into the language model as a prompt and, based on the results, says, "Considering the patient's physical strength, it might be best to start with medication and see how it goes."

[0126] Next, in phase 3, the human account makes a statement such as, "The results of the blood test haven't come out yet." The statement of the human account is content input by the user. For example, the information providing device 10 may display the statements of each account on the user's terminal using a chat UI and accept user input from an input field of the chat UI.

[0127] It should be noted that each account, including bot accounts and human accounts, does not need to make a statement in every phase. The information providing device 10 only needs to allow at least one account to make a statement in each phase.

[0128] Furthermore, if the accounts participating in the discussion include human accounts, it is difficult to allow users to make comments every time in each of the potentially billions of phases. For this reason, the information providing device 10 may accept user comments only in a specific phase, such as once in 10,000 phases.

[0129] Furthermore, the information providing device 10 may accept a user's comment when a certain condition is met. For example, the condition may be that a bot account has repeatedly made the same comment a certain number of times. By having the bot account use the user's comment as a prompt, it is expected that a stalled discussion will begin to move toward a conclusion.

[0130] In this way, when the answer satisfies the condition, the information providing device 10 presents the answer to the user, and then collects the text input in response to the presented answer.

[0131] Additionally, users may have knowledge that was not used to train the bot account's language model. In such cases, including human accounts may lead to more useful conclusions in the discussion than if only bot accounts were included.

[0132] (4. Fourth Embodiment) Next, a fourth embodiment will be described. Fig. 9 is a diagram illustrating an overview of an information providing device according to the fourth embodiment. Fig. 9 shows the configuration of an information providing system 1 including an information providing device 10 according to the fourth embodiment.

[0133] 9, the information providing system 1 includes an information providing device 10 and a policy information database 20d. The policy information database 20d is a database that stores various data related to various policies of the country, local governments, etc.

[0134] The information providing device 10 is a device to which various devices, including the policy information database 20d, are plugged in. The various devices include various types of servers, vehicles, smartphones, and the like.

[0135] In this embodiment, the information providing device 10 learns a model based on a data source acquired from the policy information database 20d. In this embodiment, the information providing device 10 plugs in to the policy information database 20d (step S1-3) and learns a language model based on the data stored in the policy information database 20d (step S2-3).

[0136] In this way, the information providing device 10 trains the language model on various data stored in the policy information database 20d. As a result, the language model generates a response to an input prompt based on the learned policy-related data. In other words, the information providing device 10 can generate a language model specialized for a policy field.

[0137] In this embodiment, a language model specialized in policy fields is used to provide advice on national or local government policies. Specifically, in this embodiment, accounts (roles) of "Chairpersons A to C," "Experts A to C," and "Judge System" are set in advance using system prompts from the language model.

[0138] "Chairpersons A to C" are accounts corresponding to the heads of the nation or local governments. That is, Chairpersons A to C are accounts corresponding to the prime minister, prefectural governor, city mayor, town mayor, village mayor, etc. In this embodiment, Chairperson A holds discussions using multiple accounts such as Chairperson A-1, Chairperson A-2, and Chairperson A-3, and Chairperson B holds discussions using multiple accounts such as Chairperson B-1, Chairperson B-2, and Chairperson B-3, and derives policy guidelines as Chairperson A and Chairperson B.

[0139] "Experts A to C" are expert accounts in various fields involved in the management of the nation or local governments.

[0140] Specific examples of expert accounts include the fields of finance, security, government, energy, and healthcare. The "Judge System" is an account that judges the results of discussions between each account. For example, the Judge System judges the merits of each account's arguments and proposes their content. Expert Account A holds discussions with multiple accounts, such as Expert Account A-1, Expert Account A-2, and Expert Account A-3, and Expert Account B holds discussions with multiple accounts, such as Expert Account B-1, Expert Account B-2, and Expert Account B-3, and arrives at a policy conclusion for each field.

[0141] In this embodiment, the assertions made by each of the multiple chairperson accounts are input as prompts into the language model, and the system prompts are set so that other chairperson accounts can discuss the input prompts.

[0142] With this setting, each chairperson account can use the language model to hold a spontaneous discussion. As shown in Fig. 9, the information providing device 10 uses the language model to hold a political discussion through each account (step S3-3).

[0143] For example, Chair Accounts A, B, C represent different political ideologies, such as conservative, reformist, right-wing, and left-wing, and each uses language models to debate. The virtual experts debate billions of times per second, enabling verification and problem-solving from all angles and allowing policy proposals to be made instantly.

[0144] Similarly, each expert account holds discussions with multiple accounts in each field and draws conclusions regarding the policy direction for that field. Then, chairperson accounts A to C hold discussions regarding the policy direction based on the discussions in each field. Note that the final discussion may be held by all accounts, including each account.

[0145] For example, the chairperson account holds a discussion on policy, and the "judging system" makes a judgment based on the content of the discussion (step S4-3). For example, the "judging system" is made up of multiple accounts. The multiple accounts that make up the judging system discuss the content of the discussion held by the chairperson account, etc., and judge the final policy. For example, when a predetermined termination condition is met, such as when the discussion reaches a deadlock, the judging system determines the most appropriate policy direction by comparing the arguments made by each account with the data stored in the policy information database 20d.

[0146] The "judge system" then proposes the determined policy guidelines to the actual chairperson. In other words, the information provision system 1 uses a language model to enable a large amount of policy-related discussion to be conducted in a simulated manner.

[0147] In this way, the information providing device 10 according to the embodiment generates a language model based on various knowledge registered in the policy information database 20d, conducts a discussion using the language model, and then makes a proposal regarding a policy guideline.

[0148] Therefore, the information providing device 10 according to the embodiment can effectively utilize a language model specialized for a specific field.

[0149] The above discussion is carried out through the following steps 1 to 5.

[0150] Step 1: Setting up multi-functional virtual experts. There are hundreds of sets of virtual experts in different policy fields. For example, specific characters such as "Political Science Expert A," "Political Science Expert B," and "Political Science Expert C" are set up for political science, and "Economics Expert A," "Economics Expert B," and "Economics Expert C" are set up for economics.

[0151] Step 2: Accelerating information exchange These experts exchange information based on each data and debate billions of times per second, making it possible to instantly make decisions on optimal policies and economic measures.

[0152] Step 3: Handling Data The virtual experts discuss various types of information, such as stock prices and birth rates, and quickly arrive at optimal solutions from multiple perspectives.

[0153] Step 4: Self-learning and updating: If new knowledge or information is discovered during the discussion, it is immediately reflected in the database and used in future discussions, allowing for diagnosis and treatment based on the latest policy information.

[0154] Step 5: Building consensus. Discussion leads to the formation of a final "consensus" through the clash of different opinions and perspectives. This allows information and knowledge from multiple sources to be comprehensively incorporated when policy proposals are made.

[0155] The above has been explained about the chairman account, but each of the Expert A to C accounts also has its own character set, and each account is further divided into various characters, and by each holding a discussion, a consensus is formed as Experts A to C.

[0156] (5. Fifth Embodiment) Next, a fifth embodiment will be described. Fig. 10 is a diagram illustrating an overview of an information providing device according to an embodiment. Fig. 10 shows the configuration of an information providing system 1 including an information providing device 10 according to an embodiment.

[0157] 10, the information providing system 1 includes an information providing device 10 and an investment information database 20e. The investment information database 20e is a database that stores various data related to investments.

[0158] In the information providing system 1, the investment information database 20e stores information published by various news media (Nikkei average, long-term and short-term interest rates, employment statistics, etc.) and information acquired independently by the information providing system 1.

[0159] The investment information database 20e independently acquires and stores information such as information on the state of crop yield and the number of cars parked in parking lots of large supermarkets and the like, obtained, for example, from photo analysis of short-range satellites, as well as POS data, individual warehouse information, credit card information, housing sales market data, etc. In this embodiment, the information providing device 10 plugs in to the investment information database 20e (step S1-4) and learns a language model based on the data stored in the investment information database 20e (step S2-4).

[0160] In this way, the information providing device 10 trains the language model on various data stored in the investment information database 20e. As a result, the language model generates a response to an input prompt based on the trained investment-related data. In other words, the information providing device 10 can generate a language model specialized for the investment field.

[0161] In this embodiment, a language model specialized for the investment field is used to provide investment advice. Specifically, in this embodiment, accounts (roles) of "Investors A to C," "Experts A to C," "Judge System," and "Risk Management" are set in advance using system prompts from the language model.

[0162] In this embodiment, accounts include bot accounts and human accounts. A bot account is a language model that can set system prompts. A bot account outputs answers using a language model. On the other hand, a human account is a human being. In other words, a human account outputs answers according to human input.

[0163] "Investors A to C," "Experts A to C," "Judge System," and "Risk Management" in Figure 10 are high-level accounts. High-level accounts include at least either bot accounts or human accounts. White circles in Figure 10 represent bot accounts. Black circles in Figure 10 represent human accounts.

[0164] Hereinafter, bot accounts or human accounts included in higher-level accounts may be referred to as lower-level accounts. Also, when it is clear from the preceding and following explanations or drawings, or when there is no need to distinguish between higher-level and lower-level accounts, the terms "higher level" and "lower level" will be omitted and the term "account" will be used.

[0165] The following explanation assumes that all accounts are bot accounts. The configuration shown in Figure 10 can be realized by replacing bot accounts with human accounts as appropriate, or by adding human accounts as subordinate accounts to higher-level accounts.

[0166] "Investors A to C" are higher-level accounts corresponding to, for example, institutional investors who manage so-called funds and individual investors. In this embodiment, Investor A is made up of Investor A-1, Investor A-2, and Investor A-3, and Investor B is made up of Investor B-1, Investor B-2, and Investor B-3, etc., and each of Investors A to C is made up of multiple lower-level accounts, and Investor A reaches a conclusion regarding his / her investment policy through discussions with Investors A-1, A-2, and A-3. For example, even if Investor A-1 places importance on the inflation rate, Investor A-2 places importance on employment statistics, and Investor A-3 places importance on the mining and manufacturing production index, each of them will hold discussions and reach a conclusion as Investor A. In this way, there are various types of indicators in the world: 1. Inflation rate: Indicates the rate of price increase. 2. Unemployment rate: Indicates the proportion of unemployed people in the labor force. 3. GDP growth rate: Indicates the growth of gross domestic product. 4. Consumer confidence index: Indicates consumers' confidence in the economy. 5. 5. Industry Production Index: Indicates production activity in the manufacturing, mining, and public utilities industries. 6. Employment Growth: Number of new employees. 7. Real Wages: Wage movements taking into account price fluctuations. 8. Home Sales Data: Sales status of new and existing homes. 9. Consumer Price Index (CPI): Indicates price fluctuations for general consumers. 10. Producer Price Index (PPI): Price fluctuations at the production stage. 11. Retail Sales: Sales status of the retail industry. 12. Business Inventories: The amount of goods and materials in stock held by companies. 13. International Trade Balance: The difference between exports and imports. 14. Manufacturing Orders Data: Order status of the manufacturing industry. 15. Non-Manufacturing Activity Index: Indicates economic activity in the service industry, etc. 16. Financial Market Trends: Stock prices, bond yields, currency prices, etc. 17. Bank Lending Trends: Status of bank lending activity. 18. 18. Core inflation rate: Inflation rate excluding food and energy. 19. Wage growth rate: Indicates the growth of average wages. 20. Global economic trends: Economic growth rates and policy trends in major countries. There are various indicators such as these, and the event is characterized by the fact that investors who place importance on each of these indicators are divided into groups to hold discussions. "Experts A to C" are expert accounts in various fields related to investment.

[0167] The various investment fields include commodities, corporate bonds, real estate, FX, and portfolios. The "Judge System" is an account that judges the results of discussions between each account. For example, the Judge System judges the merits of each account's argument and proposes its content.

[0168] In this embodiment, the assertions made by each of a plurality of accounts are input as prompts into the language model, and the system prompts are set so that other accounts can debate the input prompts.

[0169] Such a setting allows each account to spontaneously hold a discussion using the language model. The information providing device 10 uses the language model to hold a discussion about investment policies among the accounts (step S3-4).

[0170] For example, investor accounts A, B, and C have different investment philosophies, and each account uses a language model to debate with the other. The virtual experts debate billions of times per second, enabling verification and problem-solving from every angle, and allowing for instantaneous proposals for investment policies, etc.

[0171] Similarly, each expert account holds discussions with multiple accounts in each field and draws a conclusion on investment policy for each field.The investor account then holds discussions on investment policy based on the discussions in each field.The final discussion may include all accounts, including each account.

[0172] As a result of these discussions, the "judgment system" makes a judgment based on the content of the discussion (step S4-4). For example, the "judgment system" is made up of multiple accounts. The multiple accounts that make up the judgment system discuss the content of the discussions of the investor accounts, etc., and judge the final investment policy. For example, when a predetermined termination condition is met, such as when the discussion reaches a deadlock, the judgment system determines the most appropriate investment policy by comparing the claims of each account with the data stored in the investment information database 20e.

[0173] The "judgment system" then proposes the determined investment policy to actual investors. In other words, the information provision system 1 uses a language model to enable a large number of simulated discussions regarding investment policies.

[0174] Furthermore, "Risk Management" discusses the risks in the current portfolio based on the latest information registered in the investment information database 20e. "Risk Management" is made up of multiple accounts, and the multiple accounts that make up the risk management discuss investment risks. As a result, "Risk Management" advises discontinuing the investment if the investment risk exceeds a threshold (step S5-4).

[0175] In other words, the information providing device 10 can prevent or minimize investment losses and the like through the "risk management" function.

[0176] In this way, the information providing device 10 according to the embodiment generates a language model based on various knowledge registered in the investment information database 20e, conducts a discussion using the language model, and then makes a proposal regarding an investment policy.

[0177] Therefore, the information providing device 10 according to the embodiment can effectively utilize a language model specialized for a specific field.

[0178] The above discussion is realized by the following steps 1 to 5.

[0179] Step 1: Set up multi-functional virtual experts. There are hundreds of sets of virtual experts in different investment fields. For example, in the real estate field, specific characters such as "Real Estate Expert A," "Real Estate Expert B," and "Real Estate Expert C" are set up, and in FX, specific characters such as "FX Expert A," "FX Expert B," and "FX Expert C" are set up.

[0180] Step 2: Accelerating information exchange These experts exchange information based on each data and debate billions of times per second, making it possible to instantly make decisions on optimal investment policies and other matters.

[0181] Step 3: Data Handling The virtual experts share various types of information, such as stock prices, while holding discussions, allowing them to quickly arrive at optimal solutions from multiple perspectives.

[0182] Step 4: Self-learning and updating If new knowledge or information is discovered during the discussion, it is immediately reflected in the database and used in future discussions. This allows us to provide optimal investment policies based on the latest investment information.

[0183] Step 5: Building consensus. Discussion leads to the formation of a final "consensus" through the clash of different opinions and perspectives. This allows information and knowledge from multiple sources to be comprehensively incorporated when proposing investment policies.

[0184] The above has been explained about investor accounts, but each of the Expert A to C accounts also has its own character set, and each account is further divided into various characters, and each of them holds discussions to form a consensus as Experts A to C.

[0185] The information providing device 10 judges the results of the discussion by the information providing device 10 using a judge account set in the language model and proposes an investment policy. For example, the information providing device 10 makes a judgment when a predetermined termination condition is met, such as when the discussion by the information providing device 10 reaches a deadlock.

[0186] For example, the information providing device 10 refers to various data registered in the investment information database 20e, judges the investment policy that is estimated to be the most appropriate among the investment policies discussed by each investor account, and proposes that investment policy. Alternatively, the information providing device 10 may propose an investment policy by taking into account various consensuses, such as the consensus of expert A, in addition to the consensus results of the investor accounts.

[0187] Furthermore, when the risk management account determines that the risk is high, the information providing device 10 advises the investor to withdraw from the investment, etc. This makes it possible to reduce the risk associated with the investment.

[0188] [Discussion when a human account is included] Up until now, discussion has been explained assuming that the lower-level accounts are bot accounts. Here, we will explain discussion when the lower-level accounts include human accounts. In this case, the processing content of steps 1 and 2 above by the information providing device 10 changes.

[0189] First, in step 1, the information providing device 10 changes the configuration of at least some of the higher-level accounts so that they include human accounts as lower-level accounts. For example, the information providing device 10 replaces bot accounts with human accounts or adds new human accounts. A user, i.e., a real person, is associated with each human account.

[0190] Next, in step 2, the information providing device 10 presents the text output as an answer by each account to the user associated with the human account. Then, the information providing device 10 accepts text input from the user. The answer text output by each account and the action of outputting text as an answer are called "statements." "Statements" include text input by the user.

[0191] Bot accounts treat utterances from human accounts in the same way as bot accounts. For example, bot accounts input utterances from other bot accounts and human accounts as prompts into their language models and then output answers.

[0192] The processing of the information providing device 10 will be described using Figure 11. Figure 11 is a diagram for explaining a discussion. As shown in Figure 11, the information providing device 10 conducts a discussion as the phases progress. Here, it is assumed that investor A, which is a higher-level account, is composed of three lower-level accounts: investor A-1, investor A-2, and investor A-3.

[0193] Investor A-1 is a bot account that places importance on home sales data. For example, the language model for investor A-1 places importance on home sales data as the basis and outputs text related to investments.

[0194] Investor A-2 is a bot account that places importance on employment statistics. For example, the language model of Investor A-2 places importance on employment statistics as a basis and outputs text related to investment.

[0195] The substance of a bot account is a language model, and discussions are realized by the information providing device 10 repeatedly inputting prompts to the language model. The prompts are, for example, statements made by the language model in the previous phase. The type of data to be emphasized in a statement may be set by a system prompt. As a result, each bot account tends to respond differently.

[0196] Investor A-3 is a human account. A user is associated with investor A-3. Investor A-3 may have any personality (what he or she values).

[0197] 11, in phase 1, investor A-1 says, "Houses are selling well in Tokyo, so let's invest in real estate in Tokyo." At this point, the information providing device 10 progresses the phase.

[0198] In Phase 2, Investor A-2 inputs the statement made by Investor A-1 in Phase 1 into the language model as a prompt and then says, "However, it seems that employment is decreasing in Tokyo, so there may be a risk."

[0199] Next, in phase 3, the human account utters, "I heard that the number of people living in Osaka is increasing." The utterance of the human account is content input by the user. For example, the information providing device 10 may display the utterances of each account on the user's terminal using a chat UI and accept user input from an input field of the chat UI.

[0200] It should be noted that each account, including bot accounts and human accounts, does not need to make a statement in every phase. The information providing device 10 only needs to allow at least one account to make a statement in each phase.

[0201] Furthermore, if the accounts participating in the discussion include human accounts, it is difficult to allow users to make comments every time in each of the potentially billions of phases. For this reason, the information providing device 10 may accept user comments only in a specific phase, such as once in 10,000 phases.

[0202] Furthermore, the information providing device 10 may accept a user's comment when a certain condition is met. For example, the condition may be that a bot account has repeatedly made the same comment a certain number of times. By having the bot account use the user's comment as a prompt, it is expected that a stalled discussion will begin to move toward a conclusion.

[0203] In this way, when the answer satisfies the condition, the information providing device 10 presents the answer to the user, and then collects the text input in response to the presented answer.

[0204] Additionally, users may have knowledge that was not used to train the bot account's language model. In such cases, including human accounts may lead to more useful conclusions in the discussion than if only bot accounts were included.

[0205] The processing procedure for a discussion will be described with reference to Fig. 12. Fig. 12 is a flowchart showing the processing procedure for a discussion. As shown in Fig. 12, when a phase starts, the information providing device 10 determines whether to allow a human account to make a comment (step S201). The conditions for allowing a human account to make a comment include whether a specified number of phases have passed, whether a bot account has repeated the same comment a certain number of times, etc.

[0206] If the information providing device 10 does not allow a human account to make a comment (step S202, No), the information providing device 10 proceeds to step S204. On the other hand, if the information providing device 10 allows a human account to make a comment (step S202, Yes), the information providing device 10 collects the comments of the human account (step S203). For example, the information providing device 10 collects text entered by a user in an input field of a chat UI as the comments of the human account. Furthermore, the information providing device 10 may present the comments of each account to the user immediately before step S203.

[0207] Next, the information providing device 10 collects comments from the bot account (step S204). If the discussion has ended (step S205, Yes), the information providing device 10 performs a discussion termination process (step S208). For example, the information providing device 10 may determine that the discussion has ended when a predetermined number of phases have elapsed or when a judge account performs a judgment. The discussion termination process involves transitioning to a discussion between higher-level accounts (e.g., step S3 in FIG. 1).

[0208] If the discussion has not ended (step S205, No), the information providing device 10 advances the phase (step S206), inputs a prompt into the language model of each bot account based on the collected comments (step S207), returns to step S201, and performs processing in the phase after the advance.

[0209] 6. Sixth Embodiment Next, a sixth embodiment will be described. Fig. 13 is a diagram illustrating an overview of an information providing device according to the sixth embodiment.

[0210] 13 , the information providing system 1 includes an information providing device 10 and a database 20. The database 20 is a database that stores various data. Note that the database 20 may be configured with multiple databases, or may be configured inside the information providing device 10.

[0211] The information providing device 10 is a device in which various devices, including a database 20, are plugged in. The various devices include various servers, vehicles, smartphones, and other devices. That is, data is transmitted in real time from various servers, vehicles, smartphones, and other devices, and the database 20 stores this data.

[0212] In this embodiment, the information providing device 10 plugs in to the database 20 (step S1-5), and learns a language model based on the data stored in the database 20 (step S2-5).

[0213] As described above, the data stored in database 20 is constantly updated, so a language model trained using various data stored in database 20 can generate answers based on the latest data.

[0214] In this embodiment, a proposal is made for a discussion using a language model. Specifically, as shown in Fig. 13, the information providing device 10 sets multiple accounts in the language model by inputting prompts to the language model, and then holds a discussion using the multiple accounts (step S3-5).

[0215] The example shown in Figure 13 shows a scene where a total of three accounts, Accounts A to C, have been set up. Also, for example, as shown in Figure 13, Accounts A to C may each be made up of multiple accounts, and discussions may be held in each account before conclusions are reached as Accounts A to C.

[0216] Next, an outline of a discussion by each account will be described with reference to Fig. 14. Fig. 14 is a diagram showing an outline of a discussion according to an embodiment. In this embodiment, a method for more efficiently conducting a discussion by each account using a language model is proposed.

[0217] Specifically, as shown in Fig. 14, first, account A states its opinion as account A on a specific topic and asks account B for its opinion (step S11). Account B, having been asked for its opinion by account A, states its opinion as account B and asks account C for its opinion (step S12).

[0218] In this case, Account B will agree with the parts of Account A's argument that it agrees with, explaining the reasons for doing so, and will state its own argument from a different perspective than Account A.

[0219] When account C is asked for its opinion by account B, it states its argument as account C and then asks account A for its opinion (step S13). At this time, account C agrees with the arguments of account A and account B to the extent that it agrees with them, explaining the reasons for doing so, and states its argument as account C from a different perspective than account A and account B.

[0220] For example, Account C will theoretically agree with Account A or Account B, or both, on points where it can agree, and if it makes an argument from a completely different perspective within that range of agreement, it will state its argument as Account C.

[0221] In addition, if there is no other point of view, Account C will clarify which opinion it agrees with, Account A or Account B.

[0222] Account A then states its argument in response to the claims of Account B and Account C and asks for Account B's opinion. At this point, Account A will agree with the parts of Account B and Account C's argument that it can agree with, provide further logical explanation for Account A's initial argument, or state its argument from a different perspective than Account B and Account C.

[0223] Through this loop of discussion, accounts A to C each make logically explainable claims from different perspectives, but when no logically explainable claims remain, they agree with the claims of the other accounts.

[0224] Ultimately, each account will agree with the assertion made by one of the accounts. In this embodiment, a state in which all accounts have reached an agreement is referred to as a state in which consensus has been reached. The information providing device 10 then provides information about such discussions (such as text related to the responses of each account) to users, etc. For example, by making decisions based on information about such discussions, users can avoid conflicts caused by assertions that differ from those of others.

[0225] In this way, in the information providing device 10 according to the embodiment, each account expresses their opinion on the topic from various angles, and the discussion continues until a consensus is finally reached.

[0226] Therefore, the information providing device 10 according to the embodiment can perform a simulation of a discussion for reaching a consensus among any group of people with different opinions, and can therefore effectively utilize a language model.

[0227] Next, an example of a prompt to be input to a language model will be described with reference to Fig. 15. Fig. 15 is a diagram showing an example of a prompt according to the sixth embodiment. As shown in Fig. 15, the prompt is text information regarding the prerequisites for a discussion between multiple accounts (here, three genius scientists).

[0228] As shown in Figure 15, for example, genius scientist A states his opinion and then asks for B's opinion, saying, "This is what I think. What do you think, B?" Similarly, genius scientist B asks for C's opinion, saying, "This is what I think. What do you think, C?"

[0229] Just like A and B, C also states his opinion and then asks A for his opinion, saying, "This is what I think. What do you think, A?" At this point, A to C will each either agree with the other person's opinion or state an opinion that differs from the other person's.

[0230] Through these discussions, a consensus will ultimately be reached until no more differing opinions emerge, that is, until all accounts agree with one of the opinions of Mr. A, Mr. B, or Mr. C.

[0231] With such prompts, each account set in the language model will make its own argument and continue the discussion until a consensus is reached. While the example in FIG. 3 shows a case where multiple accounts are "genius scientists," this is not limiting. Any arbitrary account may be set. In this case, the arbitrary accounts may have the same attributes or different attributes. Specifically, for example, assuming a discussion about a patient's medical treatment, each account may be a doctor, a patient, a family member, etc., and in this case, the doctor may be further subdivided into a surgeon, an internist, a dermatologist, etc.

[0232] The information providing device 10 repeats discussions until a consensus is reached among the multiple accounts after each of the multiple accounts set in a language model that generates an answer to an input prompt asserts its opinion and then requests the opinions of other accounts.

[0233] For example, the information providing device 10 sets multiple accounts in the language model by inputting a prompt such as that shown in Fig. 15 into the language model, and starts a discussion using the multiple accounts. In this case, the multiple accounts may be set by the user or may be set by the language model. Furthermore, the topic of the discussion may also be set by the user or may be set by the language model.

[0234] For example, when multiple accounts are set up in a language model, a discussion is conducted according to the following steps 1 to 5. The following describes a case where multiple doctor accounts discuss the medical treatment policy of an actual patient.

[0235] Step 1: Setting up multi-functional virtual experts. There are hundreds of sets of virtual experts. For example, specific characters such as "Neuro-Expert A," "Neuro-Expert B," and "Neuro-Expert C" are set up for neurology, and "Cardiologist A," "Cardiologist B," and "Cardiologist C" are set up for cardiology.

[0236] Step 2: Accelerating information exchange These experts exchange information based on patient data and debate billions of times per second, enabling them to instantly make decisions about the patient's condition, symptoms, and treatment options.

[0237] Step 3: Data Handling The virtual experts share and discuss various types of information, such as MRI data, genetic data, and blood data, allowing them to quickly arrive at optimal solutions from multiple perspectives.

[0238] Step 4: Self-learning and updating If new knowledge, information, or treatment methods are discovered during the discussion, they are immediately reflected in the database 20 and used in future discussions. This allows for diagnosis and treatment to be provided based on the latest medical information.

[0239] Step 5: Forming a consensus. Discussion leads to the formation of a final "consensus" through the clash of different opinions and perspectives. This allows information and knowledge from multiple sources to be comprehensively incorporated when making a diagnosis or treatment proposal for the patient. There may be multiple treatment proposals.

[0240] Through these discussions, for example, a consensus can be reached among multiple doctor accounts regarding the patient's medical care policy. Note that, for example, multiple accounts involved in the patient's medical care, such as the patient account and family accounts, may participate in the discussions.

[0241] The information providing device 10 provides information about the above-mentioned discussion to users, etc. For example, the information providing device 10 provides information about the discussion in the form of responses output by the language model as each account.

[0242] This allows users to check the content of the discussion and make decisions based on this content.

[0243] Next, a processing procedure executed by the information providing device 10 according to the sixth embodiment will be described with reference to Fig. 16. Fig. 16 is a flowchart showing a processing procedure executed by the information providing device 10 according to the sixth embodiment.

[0244] 16, the information providing device 10 first inputs a prompt to the language model (step S301), and then the information providing device 10 conducts a discussion using the language model (step S302).

[0245] Next, the information providing device 10 determines whether a consensus has been reached (step S303). If the information providing device 10 determines that a consensus has been reached (step S303; Yes), it ends the process. If the information providing device 10 determines that a consensus has not been reached (step S303; No), it returns to the process of step S302 and repeats the discussion.

[0246] According to this embodiment, a language model trained based on data stored in a database is used to carry out a discussion until a consensus is reached, so that the language model can be effectively utilized.

[0247] Although the present invention has been described above using embodiments, the technical scope of the present invention is not limited to the scope described in the above embodiments. It will be apparent to those skilled in the art that various modifications and improvements can be made to the above embodiments. It is clear from the claims that such modifications and improvements can also be included within the technical scope of the present invention.

[0248] It should be noted that the order of execution of each process, such as operations, procedures, steps, and stages, in the devices, systems, programs, and methods shown in the claims, specifications, and drawings is not specifically stated as "before," "prior to," etc., and that the processes can be performed in any order unless the output of a previous process is used in a subsequent process. Even if the operational flow in the claims, specifications, and drawings is described using "first," "next," etc. for convenience, this does not mean that the processes must be performed in this order.

[0249] REFERENCE SIGNS LIST 1 Information provision system 10 Information provision device 11 Communication unit 12 Storage unit 13 Control unit 20 Database 20a Legal case database 20b Accounting case database 20c Medical case database 20d Policy information database 20e Investment information database 31 Acquisition unit 32 Generation unit 33 Discussion unit 34 Judging unit 121 Language model information

Claims

1. An information providing device comprising: a discussion unit in which a language model is generated by learning data stored in a database relating to legal cases, the discussion unit holding discussions on legal cases using a plurality of lawyer accounts each making different claims set in the language model which generates answers to input prompts; and a judge unit which judges the merits of the claims based on the results of the discussion by the discussion unit using a judge account set in the language model.

2. The information providing device according to claim 1, wherein the discussion section conducts the discussion including a prosecutor account set in the language model.

3. The information providing device according to claim 1, wherein the judging section makes a judgment by referring to relevant laws or past precedents stored in the database.

4. The information providing device according to claim 1, wherein the discussion section discusses whether or not a client company should file a lawsuit.

5. The information providing device described in claim 1, wherein the discussion department discusses the contents of anticipated litigation between the client company and other companies and the contents of a contract to avoid or win litigation, and the judging department prepares a draft of the contract contents based on the judging results.

6. An information provision method executed by an information provision device, comprising: a discussion process in which a legal case is discussed by a plurality of lawyer accounts making different claims, the plurality of lawyer accounts being set in a language model that is generated by learning data stored in a database relating to legal cases and that generates answers to input prompts; and a judging process in which a judge account set in the language model judges the merits of the claims based on the results of the discussion in the discussion process.

7. An information provision program that causes a computer to execute the following steps: a discussion step in which a legal case is discussed by a plurality of lawyer accounts making different claims, the plurality of lawyer accounts being set in a language model that is generated by learning data stored in a database relating to legal cases and that generates answers to input prompts; and a judging step in which a judge account set in the language model judges the merits of the claims based on the results of the discussion in the discussion step.

8. An information providing device comprising: a discussion section in which accounting-related discussions are conducted by a plurality of accountant accounts making different claims set in a language model that is generated by learning data stored in a database relating to accounting cases and that generates answers to input prompts; and a judging section that judges the merits of the claims based on the results of the discussion by the discussion section using a judge account set in the language model.

9. An information providing device comprising: a discussion unit in which a discussion regarding a patient's treatment plan is held among a plurality of doctor accounts set in the language model, the discussion unit being a language model generated by learning data stored in a medical database and generating answers to input prompts; and a proposal unit in which a judge account set in the language model judges the results of the discussion by the discussion unit and proposes the treatment plan.

10. An information providing device comprising: a discussion section in which a discussion regarding a patient's treatment plan is held among a plurality of doctor accounts set in the language model, the discussion section being generated by learning data stored in a medical database and generating responses to input prompts; and a proposal section in which a judge account set in the language model judges the results of the discussion by the discussion section and proposes the treatment plan, the proposal section generating minutes of the discussion that led to the treatment plan and providing the minutes together with the treatment plan.

11. An information providing device comprising: a discussion unit which holds discussions regarding a patient's treatment plan among a plurality of doctor accounts set in the language model, the discussion unit generating responses to input prompts, and a proposal unit which judges the results of the discussion by the discussion unit using a judge account set in the language model and proposes the treatment plan, wherein the discussion unit restricts access to the data in the language model during the discussion.

12. An information providing device comprising: a collection unit that collects answers output by a plurality of language models that output answers regarding a patient's treatment plan based on an input prompt, the answers being set so that the answer tendencies of the multiple language models differ from each other, and text input by a user; a presentation unit that presents the collected answers to the user; and an output control unit that inputs the collected answers and the text to the multiple language models as prompts, and further causes the answers to be output.

13. An information providing device comprising: a discussion unit in which a language model is generated by learning data stored in a database related to corporate activities, and in which a discussion on a company's management policies is held using a plurality of CEO accounts set in the language model that generates answers to input prompts; and a proposal unit that judges the results of the discussion by the discussion unit using a judge account set in the language model and proposes the management policies.

14. An information providing device comprising: a discussion section in which a language model is generated by learning data stored in a database related to politics, the discussion section discussing policy directions using a plurality of chairperson accounts set in the language model that generates answers to input prompts; and a proposal section that judges the results of the discussion by the discussion section using judge accounts set in the language model and proposes the policy directions.

15. An information providing device comprising: a discussion unit that holds discussions on investment policies using a plurality of investor accounts set in a language model that is generated by learning data stored in a database related to investments and that generates answers to input prompts; and a proposal unit that judges the results of the discussion by the discussion unit using a judge account set in the language model and proposes the investment policies.

16. An information providing device comprising: a discussion unit in which an investment policy is discussed by a plurality of investor accounts set in a language model generated by learning data stored in a database related to investment, the language model generating answers to input prompts; and a proposal unit in which a result of the discussion by the discussion unit is judged by a judge account set in the language model and the investment policy is proposed, wherein the proposal unit generates minutes of the discussion that led to the investment policy and provides the minutes together with the investment policy.

17. An information providing device comprising: a discussion unit in which a language model is generated by learning data stored in a database related to investments, the discussion unit holding discussions on investment policies using a plurality of investor accounts set in the language model that generates answers to input prompts; and a proposal unit that judges the results of the discussion by the discussion unit using a judge account set in the language model and proposes the investment policy, wherein the discussion unit restricts access to the data in the language model during the discussion.

18. An information providing device comprising: a discussion unit in which multiple accounts set in a language model that generates answers to input prompts each state their opinion, and then the discussion is repeated until a consensus is reached among the multiple accounts by soliciting opinions from other accounts; and a providing unit that provides information regarding the content of the discussion by the discussion unit.

Images