Predictive information generation program

Abstract

We provide a predictive information generation system that can automatically generate predictive information that forecasts a future in which the created idea can be effectively promoted. [Solution] The method includes a prompt input step that accepts prompt input to a base model, which is a large-scale language model that has learned text data describing the content of an idea and future prediction information; an information acquisition step that acquires prediction information consisting of text data about the future that can appeal to the idea, output from the base model in response to the prompt, and generates request information about things to be expected regarding the subject of the idea based on the acquired prediction information; a request information input step that inputs the request information to an idea creation base model that has been pre-trained with patent information; and an idea acquisition step that acquires an idea that has been refined according to the prediction information, output from the idea creation base model in response to the request information.

Description

Technical Field

[0006] ,

[0001] The present invention relates to a prediction information generation program that utilizes a foundation model and automatically generates prediction information for predicting a future in which the created ideas can be appealed.

Background Art

[0002] Patent Document 1 proposes a problem-solving support system that can dig deep into numerous peripheral deformation forms from the seeds of the created invention and further strengthen the technology group by reflecting information regarding market needs.

[0003] Patent Document 2 proposes a creation support program suitable for supporting the creation of innovations by effectively presenting specific examples in which the solution concept is reflected to the creator.

[0004] In the disclosed technologies of these Patent Documents 1 and 2, when learning technical information including patent information by artificial intelligence and inputting information regarding problems to be solved and issues related to the invention to be newly created, an optimal solution concept is proposed via artificial intelligence.

Prior Art Documents

Patent Documents

[0005]

Patent Document 1

Patent Document 2

Summary of the Invention

Problems to be Solved by the Invention

[0006] However, creative support tools that focus on commercializing the resulting solution concepts and inventions have not been proposed until now. If commercialization is the goal, then obtaining predictive information that forecasts a future where the invention idea can be appealing would allow for prior consideration of how to develop the idea and connect it to a business in the future, but a method for doing so has not been proposed until now.

[0007] Therefore, the present invention was devised in view of the above-mentioned problems, and its objective is to provide a predictive information generation program that can automatically generate predictive information that forecasts a future in which the created idea can be effectively communicated, using a base model (a large-scale model trained on a large amount of data and adaptable to various tasks). [Means for solving the problem]

[0008] The predictive information generation program according to the present invention is characterized by causing a computer to execute the following steps: a prompt input step in which a prompt including text data describing the content of an idea is input to a base model, which is a large-scale language model that has learned text data describing the content of an idea and known future prediction information that has been predicted about the future; an information acquisition step in which predictive information consisting of text data about the future that can appeal to the idea, output from the base model in response to the prompt input accepted in the prompt input step, is obtained and requests information about things to be expected for the subject of the idea based on the obtained predictive information; a request information input step in which the request information generated in the information acquisition step is input to an idea creation base model that has been pre-trained with patent information; and an idea acquisition step in which an idea is obtained from the idea creation base model in response to the request information input in the request information input step, which has been refined according to the predictive information. [Effects of the Invention]

[0009] According to the present invention, it is possible to automatically generate predictive information that forecasts a future in which an idea can be effectively communicated, by utilizing a base model (a large-scale model trained on a large amount of data and adaptable to various tasks). [Brief explanation of the drawing]

[0010] [Figure 1] Figure 1 is a schematic diagram showing the configuration of a network including an idea evaluation system to which the present invention is applied. [Figure 2] Figure 2 is a diagram illustrating the operation flow of the present invention. [Figure 3] Figure 3 shows examples of how each simulator can be used. [Figure 4] Figure 4 shows an example of generating request information for new ideas to be created based on acquired evaluation information. [Figure 5] Figure 5 shows an example of generating request information using a large-scale language model. [Figure 6] Figure 6 is a diagram illustrating the second embodiment. [Modes for carrying out the invention]

[0011] The idea evaluation system to which the present invention is applied will be described in detail below with reference to the drawings.

[0012] First Embodiment Figure 1 is a schematic diagram showing the configuration of a network including an idea evaluation system 1 to which the present invention is applied. As shown in the figure, the server 10 is connected to a user terminal 30 via a communication network 20 such as the Internet. In Figure 1, only one user terminal 30 is shown, but the server 10 may be connected to two or more user terminals 30. The server 10 provides evaluation results specified by the user who is operating the user terminal 30.

[0013] First, the hardware configuration of server 10 will be described. Server 10 is configured as a general computer and includes a computer processor 11, a memory 12, an input / output I / F 13, a communication I / F 14, and a storage 15, as shown in FIG. 1. These components within server 10 are electrically connected via a bus or the like not shown in the figure.

[0014] The computer processor 11 is configured as a CPU, GPU, or the like, reads various programs stored in the storage 15 or the like into the memory 12, and executes various instructions included in the programs. The memory 12 is constituted by, for example, DRAM or the like.

[0015] The input / output I / F 13 includes various input / output devices for exchanging information with an operator or the like. The input / output I / F 13 includes, for example, information input devices such as pointing devices such as a keyboard, a mouse, and a touch panel, voice input devices such as a microphone, and image input devices such as a camera. The input / output I / F 13 also includes image output devices such as a display and voice output devices such as a speaker.

[0016] The communication I / F 14 is implemented as hardware such as a network adapter, various communication software, or a combination thereof and is configured to enable wired or wireless communication via a communication network 20 or the like.

[0017] The storage 15 is constituted by, for example, a magnetic disk or a flash memory or the like. The storage 15 stores various programs including an operating system and various data and the like. For example, as shown in FIG. 1, the storage 15 has an evaluation information management table 152 or the like that manages evaluation information with an evaluation attached to an idea.

[0018] In addition, the storage 15 stores a server - side program 40 according to an embodiment of the present invention. The program 40 is a program for causing the server 10 to function as all or part of a system for providing services of an idea evaluation system. At least part of the server - side program 40 may be configured to be executed on the user terminal 30 side via a terminal - side program 42 described later.

[0019] In the present embodiment, the server 10 may be configured by using a plurality of computers each having the above - described hardware configuration. In such a case, the server 10 is composed of a plurality of server devices.

[0020] The server 10 configured as described above can be configured to have functions as a web server and an application server, execute various processes in response to requests from the user terminal 30, and transmit screen data and control data composed of HTML data or the like according to the results of the processes to the user terminal 30. On the user terminal 30, a screen based on the received data is output.

[0021] Next, the hardware configuration of the user terminal 30 will be described. The user terminal 30 is configured as a general computer and includes, as shown in FIG. 1, a computer processor 31, a memory 32, an input / output I / F 33, a communication I / F 34, and a storage 35. These components are electrically connected via a bus (not shown).

[0022] The computer processor 31 is configured as a CPU or a GPU or the like, reads various programs stored in the storage 35 or the like into the memory 32, and executes various instructions included in the programs. The memory 32 is composed of, for example, DRAM or the like.

[0023] The input / output interface 33 includes various input / output devices for exchanging information with the operator, etc. The input / output interface 33 includes, for example, information input devices such as keyboards, mice, and pointing devices such as touch panels; audio input devices such as microphones; and image input devices such as cameras. Furthermore, the input / output interface 33 includes image output devices such as displays and audio output devices such as speakers.

[0024] The communication interface 34 is implemented as hardware such as a network adapter, various communication software, and a combination thereof, and is configured to enable wired or wireless communication via the communication network 20, etc.

[0025] The storage 35 is composed of, for example, a magnetic disk or flash memory. The storage 35 stores various programs, including the operating system, and various data. The programs stored in the storage 35 can be downloaded and installed from an application market or the like. The storage 35 also stores the terminal-side program 42 described above. This program 42 is configured as a web browser or other application and can be configured to execute at least a portion of the server-side program 40, as described above.

[0026] The user terminal 30 may be configured as a mobile phone, smartphone, tablet, wearable device, or personal computer, etc.

[0027] A user operating the user terminal 30 configured in this way can utilize the idea evaluation system service provided by the server 10 by communicating with the server 10 via the terminal-side program 42 installed on the storage 35, etc.

[0028] Next, the functions of the server 10 configured in this way will be described. As shown in Figure 1, the computer processor 11 of the server 10 is configured to function by executing instructions contained in a program loaded into memory 12 or at least a part of the server-side program 40.

[0029] The server 10 then acquires evaluation information based on information received from the user terminal 30, or information generated on the server 10 side. At this time, the server 10 inputs a prompt regarding the created idea to the base model, and acquires evaluation information that has been generated by the base model and includes an evaluation of that idea. The base model is an AI neural network trained on a large amount of unlabeled datasets, and is a type of machine learning model trained to perform various tasks, from text translation to medical image analysis. The base model is programmed to understand and function within the general context of patterns, structures, and representations. This fundamental understanding of communication methods and patterns identification methods creates a baseline of knowledge, which can be further modified and fine-tuned to perform tasks in specific areas of almost any industry. Many base models, especially those used in natural language processing (NLP), computer vision, and audio processing, are pre-trained using deep learning techniques. The base model may also be a model trained to generate, identify, classify, predict, and translate natural language, still images, moving images, and music through transfer learning, fine-tuning, and distillation. The foundational model may be a model trained to detect ethically problematic information, discriminatory and harmful information, information not based on facts, or simply useless information. The foundational model may be a model trained to detect information not based on facts in natural science using a natural science simulator. Since the foundational model may include hallucination, it may be a model that has been fine-tuned to detect hallucination through reinforcement learning from human feedback (RLHF). The foundational model may also be a model that reflects "human value standards" such as human preferences and intentions through reinforcement learning from human feedback.

[0030] A foundational model trained in natural language processing can recognize context, grammar, and language structure, and generate and extract information from the data used for training. For example, a foundational model trained in natural language processing is a trained model that performs natural language processing in response to instructions and generates a response. The instructions may contain at least information about the created idea. The foundational model may be an interactive, chat-type, or conversational model that alternately receives instructions and generates responses. The natural language model may be a large-scale language model (LLM) trained on a large amount of text data, or a model that has been transfer-learned from such a large-scale language model. Furthermore, the natural language model is not limited to being built within server 10, but may be built in a system outside of the idea evaluation system 1. The foundational model may also be a diffusion model used in image generation AI.

[0031] Furthermore, this foundational model may learn M&A information regarding partnerships, mergers, and acquisitions between companies. This M&A information could include, for example, information obtained from news articles on the internet about which companies have collaborated or partnered in business, or information about mergers and acquisitions between companies. Additionally, by analyzing patent information, companies that have filed joint patent applications may be considered to have a partnership relationship, and this information may be included in the M&A data.

[0032] Furthermore, this foundational model may be trained with future prediction information. This future prediction information may include, for example, publicly available future prediction information published on the internet with titles such as "Predicting the Future in X Years," "Global Megatrends in X Years," or "Society and Corporate Scenarios in X Years." This future prediction information is likely to be highly reliable because it is predicted by experts based on data from government agencies, think tanks, various industry associations, and private research companies, and this information is incorporated and used to train the foundational model. In addition to information from the internet, future prediction information may also be incorporated as electronic data obtained by reading media such as books and documents. If these books are sold or provided as electronic data, they may be incorporated as is. This M&A information and future prediction information may be incorporated and trained based on, for example, RAG (Retrieval-Augmented Generation).

[0033] A large-scale language model is a deep learning model that pre-trains a language model, which models human spoken language based on its probability of occurrence, using a vast amount of data. In other words, a large-scale language model is a natural language processing model trained using a large amount of text data, taking text as input and outputting text. When a large-scale language model is applied to a question-and-answer system, a question is input into the large-scale language model, and the model outputs an answer.

[0034] When server 10 receives the text data (prompt) of a request, it uses a large-scale language model to statistically estimate the probability of generating the next word from the text contained in the received prompt, and sends the estimation result to the requester. As the large-scale language model, for example, publicly known technologies described on internet sites such as "https: / / huggingface.co / datasets / databricks / databricks-dolly-15k" and "https: / / huggingface.co / datasets / togethercomputer / RedPajama-Data-1T" can be used. Alternatively, as the large-scale language model, for example, GPT-4 (registered trademark) provided by OpenAI, Inc. in the United States may be used.

[0035] As shown in Figure 2, prompts regarding the idea are entered. These prompts consist of text data describing the content of the idea. These prompts may consist of items such as an overview of the invention (e.g., "Adopt a vertical-axis wind turbine to capture wind energy from all directions"), a problem to be solved (e.g., "Conventional horizontal-axis wind turbines have limited wind exposure"), a solution (e.g., "Introduce a vertical-axis wind turbine that captures wind energy from all directions"), and the effects of the invention (e.g., "Improved flexibility to changes in wind direction"), or they may be a single-word expression of the invention concept or solution, such as "Use a nested structure".

[0036] This prompt may be conceived by the user through their own creative activities, or it may be based on an idea generated by artificial intelligence based on well-known technologies, such as those shown in Patent Documents 1 and 2. In such cases, a foundational model for idea generation that searches for solutions to input text data is prepared in advance by training it with patent information. This foundational model for idea generation may utilize an LLM. Then, by inputting text data such as "How can we improve the power generation efficiency of wind power generation?" into the trained model, an idea consisting of an outline of the invention, the problem to be solved, the solution, and the effects of the invention is output. This outputted idea may be a concise expression of the invention concept or solution. This idea may be output as text data, but is not limited to this; the content of the output text data may also be displayed as a diagram or drawing. In such cases, a foundational model that utilizes well-known generative AI to output the content as a diagram instead of text for the input text data may be used.

[0037] Text information of ideas created by the user themselves or output from a trained model is input as a prompt. The prompt input may be entered via the user terminal 30, or if it is already stored in the storage 35 of the user terminal 30, it may be used. Alternatively, it may be entered on the server 10 side, or if it is already stored in the storage 15 of the server 10, it may be used.

[0038] In particular, ideas generated by pre-trained models may already be stored on the user terminal 30 or server 10, so they can be used as is.

[0039] Next, the input prompt is fed into the large-scale language model, and evaluation information with an assessment of the idea is obtained. This operation is basically performed via the large-scale language model on the server 10 side. For this reason, the prompt obtained at the user terminal 30 is sent to the server 10 via the communication network 20. The server 10 may display the evaluation information obtained via the large-scale language model via the input / output interface 13, or it may send the evaluation information to the user terminal 30 via the communication network 20 and display it via the input / output interface 33 on the user terminal 30.

[0040] In addition to being embodied as such an idea evaluation system 1, the present invention may also be embodied as a program capable of performing these functions.

[0041] Furthermore, if the function for generating evaluation information based on a large-scale language model in response to input prompts is provided on the server 10 side, these operations are performed via the storage 15 and the server-side program 40. Alternatively, the function for generating evaluation information based on a large-scale language model in response to input prompts may be provided on the user terminal 30 side instead of the server 10. In this case, these operations are performed via the storage 35 and the terminal-side program 42 on the user terminal 30. The prompts acquired by the user terminal 30 are then input directly into the large-scale language model within the user terminal 30 without being sent to the server 10.

[0042] Furthermore, the evaluation of an idea may include at least one of the following: the resources or costs necessary to provide the product or service containing the idea to customers, the marketability of the idea, or the customer satisfaction level of the idea.

[0043] Details, outlines, specifications, manufacturing methods, manufacturing man-hours, required materials, import and export destinations, and logistics information necessary for transporting products and materials, including ideas, can also be obtained through a large-scale language model. After exploring this information, the resources and costs necessary to provide the product or service to the customer can be calculated from manufacturing man-hours, labor costs, material costs, etc., using pre-prepared unit prices as templates. Alternatively, these resources and costs can also be obtained through the large-scale language model, or they can be directly calculated from input prompts. Furthermore, cost and resource evaluation can be performed by preparing existing cost calculation programs and man-hour calculation programs in advance, inputting parameters into them, and using the results calculated for evaluation.

[0044] Furthermore, in addition to simply outputting the necessary resources or costs through this large-scale language model, it may also evaluate the actual business potential itself based on the quantity of those resources and the magnitude of the costs, and include this in the evaluation information.

[0045] Marketability indicates the likelihood that an idea will be accepted by customers in the market, in other words, the likelihood that customers will purchase the idea, and the market share that can be gained. The marketability of an idea is determined from prompts about the idea through a large-scale language model, and, if necessary, based on information from the internet, and the results are output.

[0046] In this case, a market simulator 51 that performs marketability simulations may be used, as shown in Figure 3.

[0047] Market Simulator 51 is a tool that performs the above-mentioned market simulations based on the input parameters. This Market Simulator 51 may utilize any existing tool. Examples of information implemented in Market Simulator 51 include financial results information of listed companies, expert comments and opinions on futures forecasts and reasons for stock price increases and decreases published in analyst reports and newspaper articles. Other examples of information implemented in Market Simulator 51 may include expert opinions on the Nikkei 225 futures as a whole, or opinions on specific segments or industries, or even opinions on individual futures, as well as comments and forecasts of increases or decreases from experts (analysts) published on the internet. Furthermore, examples of information implemented in Market Simulator 51 may include information such as interest rates, futures, exchange rates, stock prices of individual securities, price movements of crude oil, precious metals, Bitcoin, etc., sequential charts and line graphs, Bollinger Bands, trading volume, MACD, and moving averages. Examples of information implemented in the market simulator 51 may include various data related to politics, economics, society, technology, etc., such as GDP, employment statistics, industrial production index, capital investment, labor force survey, business cycle index, consumer spending, new car sales, and consumer price index.

[0048] The market simulator 51 performs a marketability simulation based on well-known simulation methods, using parameters corresponding to input prompts or information based on ideas output from a large-scale language model. The market simulator 51 may also obtain evaluation information that assesses the marketability of the idea based on the simulation results. In this case, the simulation may also consider examples of similar markets, products, or services to determine the extent to which the idea will be accepted in the market. Furthermore, the market simulator 51 may also refer to data on future market, social, political, economic, and technological forecasts to simulate marketability. Additionally, the market simulator 51 may determine the rarity of the idea based on the extent to which similar products or services exist, and then simulate marketability based on the results of this determination.

[0049] In addition to outputting the simulation results themselves from the market simulator 51 as evaluation information, the large-scale language model may also perform a solution search based on those simulation results and output evaluation information. In such cases, the evaluation information is obtained by inputting a prompt that reflects the results of the simulation by the market simulator 51 into the large-scale language model.

[0050] Customer satisfaction with an idea may be evaluated by referring to examples of similar markets, products, or services. In this case, the level of customer feedback received in the past from examples of similar markets, products, or services may be determined from sales and satisfaction data. In such cases, the degree of empathy may also be referred to based on information posted on social media and the internet, for example, the number of "likes" posted online.

[0051] Furthermore, the evaluation of an idea may include its feasibility. Here, feasibility refers to whether or not it is technically feasible, but is not limited to this; it may also include whether or not it is viable as a business model, or the possibility that it may become unfeasible due to factors other than technology, such as politics, economics, or society. In such cases, the feasibility of the idea will be evaluated using a large-scale language model that has been trained on information related to politics, economics, and society.

[0052] In particular, when evaluating feasibility based on whether or not it is technically feasible, the feasibility of an idea is assessed using a large-scale language model trained on information related to natural science. This information related to natural science includes all papers in the natural sciences, non-patent literature such as conference presentations and international conference proceedings, as well as patent literature and all technical information publicly available on the internet.

[0053] In this case, a natural science simulator 52, which performs simulations of natural sciences, may be used, as shown in Figure 3.

[0054] The Natural Science Simulator 52 is a tool that performs simulations related to the natural sciences described above based on the input parameters. This Natural Science Simulator 52 may utilize any existing tool. As the Natural Science Simulator 52, tools capable of simulating physical and chemical phenomena, and indeed all aspects of natural science, can be used, such as FEM simulators that analyze stress fields and strains based on the finite element method, chemical reaction simulations, chemical engineering simulations including plants, electrical circuit simulations, fluid dynamics simulations, quantum mechanics simulations, etc.

[0055] The natural science simulator 52 performs natural science simulations based on known simulation methods, using parameters corresponding to input prompts or ideas output from a large-scale language model. Through this natural science simulator 52, it may also be determined whether the idea itself is technically feasible. For example, if the idea violates the law of universal gravitation, which is contrary to the laws of nature, this fact may be added to the evaluation.

[0056] In addition to determining whether or not it is scientifically feasible, if it is scientifically feasible, the magnitude of its effects and its feasibility may also be determined via the natural science simulator 52.

[0057] In addition to outputting the simulation results themselves from the natural science simulator 52 as evaluation information, the large-scale language model may also perform a solution search based on those simulation results and output evaluation information. In such cases, the evaluation information is obtained by inputting a prompt that reflects the results of the simulation by the natural science simulator 52 into the large-scale language model.

[0058] Furthermore, according to the present invention, evaluation information regarding the novelty or inventive step of the above idea may be obtained by using a large-scale language model trained on patent information. Specifically, patent information includes not only all text data of publicly available publications and patent gazettes (hereinafter collectively referred to as patent gazettes), but also patent classifications. The large-scale language model searches for relevant patent gazettes based on the obtained prompt. It then compares the patent information described in the searched public patent gazettes with the content of the obtained idea prompt. It then determines whether there is a difference between the patent information described in the public patent gazettes and the content of the idea prompt, and the degree of that difference, and performs an evaluation of novelty or inventive step based on the determination result.

[0059] In such cases, a pre-trained model may be pre-configured to search for similar patent publications in response to input text data by training it with patent information beforehand. The input prompt is then input to the pre-trained model. As a result, the pre-trained model searches for similar patent publications (patent documents) in response to the input prompt. That is, text data of patent publications similar to the created idea can be obtained from the pre-trained model. Next, it is determined whether there is a difference between the patent information described in the known patent publication and the content of the idea prompt, and the degree of that difference is determined, and novelty or inventive step is evaluated based on the determination result. The searched patent publications (patent documents) may be acquired as a single piece of evaluation information.

[0060] In this case, the prompt may be input into a large-scale language model, and idea-based text data output from it may be obtained first. Idea-based text data is information about the idea that has been organized from the prompt through the large-scale language model, or generated into wording suitable for solution search. By inputting this into a trained model, the novelty or inventive step may be evaluated based on the output text data of known patent publications.

[0061] In this invention, as shown in Figure 4, request information for new ideas to be created may be generated based on the acquired evaluation information. This request information becomes a prompt (input text shown in Figure 4) when generating ideas using artificial intelligence, utilizing a trained model that searches for solutions to the input text data. If higher quality request information is generated based on the evaluation information in this way, the artificial intelligence (trained model) can generate even higher quality ideas accordingly. In particular, if the evaluation results are unsatisfactory, it is possible to efficiently generate request information necessary to further refine the invention. For example, if the evaluation result indicates that the solution method of the idea, "multiple blades and rotor," lacks novelty and inventiveness, then request information necessary to further refine "according to wind speed" can be generated, such as, "What is a method to increase the power generation effect of wind power generation in a wind power generation system that combines a low-speed cross-flow wind turbine and a high-speed Darrieus wind turbine according to wind speed?" For example, if novelty and inventiveness are lacking, the idea can be further refined by generating request information necessary to demonstrate inventiveness.

[0062] Furthermore, this request information may be input into an idea generation platform model that utilizes a large-scale language model. The system may then output information about new ideas to be created based on the input request information, derived from the large-scale language model. In other words, a large-scale language model may be applied when generating ideas using artificial intelligence.

[0063] Furthermore, a large-scale language model may be used when generating request information from evaluation information. That is, as shown in Figure 5, the generated evaluation information is input into the large-scale language model. Then, the large-scale language model outputs request information based on the input evaluation information. In other words, a large-scale language model may also be applied when generating request information from evaluation information.

[0064] Thus, according to the present invention, it is possible to automatically evaluate ideas created using a large-scale language model from various perspectives such as feasibility, marketability, and patentability. This makes it possible to automatically evaluate solution concepts and inventions completed with the help of artificial intelligence, organize them when a large number of solution concepts and inventions are automatically generated by artificial intelligence, and easily consolidate them into the necessary technologies to link them to value provided to customers.

[0065] Furthermore, even if the evaluation results are unfavorable, the large-scale language model can generate the necessary request information to further refine the invention.

[0066] It should be noted that the present invention is not limited to the embodiments described above. For example, the invention may also involve searching for companies that would be encouraged to utilize the idea based on evaluation information.

[0067] Companies that can utilize an idea are those that actually want to use that idea, or those that would benefit from collaborating through open innovation. In searching for such companies, the search is conducted based on publicly available information, namely the content of each company's patents and non-patent literature, as well as information from company websites and news releases. This search itself may be carried out using artificial intelligence, or even a large-scale language model. In such cases, evaluation information is input into the large-scale language model, and based on the aforementioned information, the large-scale language model outputs companies that would be beneficial to the utilization of the idea.

[0068] Furthermore, the acquired story information may be modified based on the evaluation information described above. In such cases, the well-known LLM technique may be used to list proposed modifications to the story information as modification candidates, or the story information itself may be output from the beginning with modifications based on this evaluation information.

[0069] Regarding the modification of story information based on evaluation information, for example, story information that results in a higher evaluation in the evaluation information may be generated based on LLM and proposed.

[0070] Furthermore, as evaluation information, evaluation information that evaluates the story from a moral perspective may be obtained using a base model trained on information related to moral insights. Information related to moral insights consists of data that describes exemplary behavior, thoughts, and philosophies from a moral or ethical standpoint. Examples of information related to moral insights include descriptions of ethical dilemmas in text format, offering different interpretations and solutions. For example, in response to the question, "Scenario: A train is out of control, and five people are standing in its path. You can switch it to another track, but one person is standing on that track. Will you switch the switch?", the model may be trained with answers such as "Switch the switch (a utilitarian perspective focused on results)" and "Do not switch the switch (a Kantian perspective emphasizing the morality of the action)." Alternatively, for example, in response to the question, "What criteria should an autonomous vehicle use to make decisions to avoid an accident?", data trained with the answer "Ethical perspective: Should pedestrian safety be prioritized, or should passengers be protected?" may be used. In addition to providing answers to such questions, one could also prepare a foundational model that has been trained on information about moral insights by accumulating training data consisting of documents that concisely and concretely explain various moral theories, ethics, and philosophical approaches. Alternatively, one could train the model on data that describes ethical dilemmas in text format and presents interpretations and solutions from different perspectives.

[0071] Second Embodiment The second embodiment will now be described. In the second embodiment, the configuration of the idea evaluation system 1 described in the first embodiment is used as is, and processing operations described in the first embodiment are combined and executed as appropriate.

[0072] In the second embodiment, a prompt regarding an idea is input to the base model, and story information regarding a business story that utilizes the idea is output from the base model in response to the input prompt.

[0073] A business story that leverages an idea, as shown in Figure 6, represents the setting or characters, the conflicts, the context, and the big idea that lead to the solution constructed from that idea.

[0074] The setting or characters represent the situation and background necessary to arrive at the idea, or solution. The characters are the actual people who appear in the story, such as customers, business partners, partner companies, employees, and stakeholders—people with whom the story's audience can empathize.

[0075] Context is formed through this setting or characters. For example, in the case of an idea (solution) regarding two-dimensional code technology, a story (text information) is generated with the setting of the two-dimensional code payment market, and the characters are those to whom this solution is proposed (e.g., a company that develops and produces two-dimensional codes). Note that the characters may be omitted when forming this context.

[0076] Furthermore, conflict represents the problems and unfortunate events that could arise if the current situation continues. For example, in the case of an idea (solution) regarding QR code technology, it generates a story that creates a mental conflict, suggesting that continuing to use the current QR code will lead to numerous security problems and incidents, and that something must be done to escape this situation. In other words, through conflict, you create excitement, leave a strong impression that the other person won't forget, break away from a boring story, and make the other person interested and lean forward to listen.

[0077] A big idea is not the solution itself, but rather a higher-level conceptualization of the solution, or a simplified expression of the solution using a catchy slogan or tagline. The big idea should be presented in a single catchy phrase that brings peace of mind and relief to the listener, and that will stick in their memory. This big idea can be generated not only from the idea itself, but also from the context and conflicts created in the preceding steps. Examples of big ideas include: "Introduce a system to measure biometric data XX without needles," "Build an investment platform for XX that is easy for XX investors to use," or "Equip the QR code itself with a XX prevention function." It can also be something that conveys what you want to do in a single phrase.

[0078] When generating such stories, it is also possible to suggest, through the generated stories, how to connect the generated ideas to a business and monetize them. Furthermore, suitable business models for realizing such stories may also be generated and suggested.

[0079] Let's explain an example of a generated story. For example, if the prompt is "Miniaturize the structure of fuel cells," the story for commercialization could be "Development and sales of ultra-small fuel cells for wearable devices," and the details could be "Developing an ultra-small fuel cell that utilizes nanoscale porous separator technology to achieve more than 10 times the operating time of conventional lithium-ion batteries. By providing it for wearable devices such as smartwatches and headsets, we will create a market for next-generation wearable devices that can be used for extended periods. Furthermore, we will develop a regular delivery service for fuel cartridges to build a sustainable revenue model."

[0080] If the underlying model is trained with M&A information, story information that incorporates this information will be generated. Therefore, users considering M&A, such as partnerships, mergers, and acquisitions between companies, can leverage their ideas while obtaining stories that take into account the details of their M&A activities.

[0081] Similarly, if future prediction information is learned, story information that incorporates this information will be generated. You can obtain a story that utilizes your ideas while taking into account predicted future developments in politics, economics, society, and technology.

[0082] Furthermore, the feasibility may be evaluated using the aforementioned natural science simulator, and the resulting evaluation information may be generated and attached to the proposed story.

[0083] When proposing this story, the evaluation information shown in the first embodiment described above may also be provided.

[0084] Alternatively, a market simulator may be used to perform a simulation, and story information may be obtained based on the results of that simulation. In such a case, if the simulation results show that the product or service is unlikely to sell well, the system may work to generate a story that creates a stronger, more impactful conflict.

[0085] Furthermore, prompts reflecting the results of a market simulator simulation may be input into the base model to generate a new story corresponding to the simulation results. If an evaluation of the novelty or inventiveness of the idea is obtained at this time, story information based on this evaluation may be acquired. For example, if there is no inventiveness, a corresponding context may be created.

[0086] Furthermore, based on the acquired story information, request information for new ideas to be created may be generated. In other words, if there are further solutions in the acquired story that would lead to a more effective resolution of the conflict, request information accordingly may be generated.

[0087] In this second embodiment, as shown in Figure 6, the basis may include information regarding the evaluation of the novelty or inventiveness of the idea, or patent information output in the first embodiment. This makes it possible to demonstrate that the solution, i.e., the idea itself, is patentable, and can also imply that the entire generated business story is protected by patents.

[0088] It should be noted that the present invention is not limited to the embodiments described above. As a prompt to be input into the base model, a prompt regarding technological seeds may be input as an alternative to a prompt regarding ideas.

[0089] The technological seeds referred to here can range from broad categories such as NFT technology, AI technology, nanotechnology, and metaverse technology, to more specific technologies such as nanoimprint technology and nanorobot technology.

[0090] Then, in response to prompts regarding such technological seeds, needs information related to business needs that can be utilized with the above-mentioned technological seeds is obtained from the base model. This needs information shows which marketing segments and customer groups actually have a need for the specific value that can be provided using those technological seeds.

[0091] Conversely, when a prompt for needs information is entered, the system could output information about technological seeds from the base model.

[0092] In both the first and second embodiments, the prompts input to the base model are not limited to text data, but may also be image data. In such cases, text information may be extracted from the image data using well-known techniques, or the semantic content of the image data may be vectorized and input to the base model as a prompt.

[0093] Similarly, in both the first and second embodiments, the information output from the base model is not limited to text data, sentences, or strings; it may also output figures, tables, image data, etc. In particular, when generating story information, it is sometimes easier to understand the story if it is output not only as text data but also as an image, or often easier to understand if images and text data are combined.

[0094] Third Embodiment The third embodiment will now be described. In the third embodiment, the configuration of the idea evaluation system 1 described in the first and second embodiments above is used as is, and processing operations described in the first embodiment are combined and executed as appropriate.

[0095] In the third embodiment, a prompt regarding an idea is input to the base model, and predictive information is obtained from the base model that predicts a future state in which the idea can be appealed to in response to the input prompt.

[0096] In this third embodiment, it is desirable that the base model is trained with future prediction information that has been predicted about the future.

[0097] In the third embodiment, instead of outputting an idea from the base model in response to an input prompt, predictive information is generated that forecasts future societies, lifestyles, businesses, etc., in which that idea can be appealed.

[0098] For example, one predictive piece of information is: "Develop a separator with a nanoscale porous structure to optimize gas flow while reducing its thickness to less than 1 / 10 of conventional separators. Further miniaturization can be achieved by integrating the anode and cathode separators." Another predictive piece of information is: "A future society that utilizes low-power, ultra-small fuel cells, in which medical devices are constantly implanted in the body for extended periods, and health checks can be performed automatically and continuously through these medical devices."

[0099] If the underlying model is trained with future prediction information, it becomes possible to obtain predictive information that incorporates future forecasts while still utilizing the ideas.

[0100] This future prediction information may include information about predicted needs in the future. This information about needs includes all kinds of needs, such as customer needs and societal needs. In such cases, the base model outputs predictive information that forecasts needs to which the idea can be appealed.

[0101] An example of a need that can be appealed to by an idea is when the prompt is "miniaturize fuel cells," and the idea to be generated is "develop a separator with a nanoscale porous structure, optimize the gas flow, and reduce the thickness to less than 1 / 10 of conventional separators. Further miniaturization will be achieved by integrating the anode and cathode separators." In this case, the generated need (predicted information) would be "the need for a lightweight VR device that can be used for extended periods," etc.

[0102] At this point, the system may allow the user to input request information as a prompt regarding expectations for the subject of the idea. This request information could include, for example, all expectations for the object on which the idea is to be refined, such as a "wind turbine," "fuel cell," "concrete structure," or "thermoelectric element." This request information could also include prompts for inputting any information expected of the object (method), such as its challenges, effects, expectations, and desired functions. It is not mandatory for such request information to be input as a prompt; the system may simply generate and display candidate request information to the user.

[0103] Candidate request information may be generated based on predictive information. In such a case, for example, if the predictive information generated is text data such as "a society in which medical devices are constantly implanted in the body for long periods of time, and health checks can be performed continuously and automatically through those medical devices," then from this, phrases such as "long-term implantation in the body" and "automatic health checks" may be selected as candidates for the next request information. An LLM may be used as appropriate to generate such phrases. This allows the user to generate ideas by inputting new request information as prompts based on the predictive content suggested by the predictive information, and then obtaining information about newly generated ideas output from the base model.

[0104] Furthermore, the generated request information may be input to the idea creation platform model described above. In this case, one or more candidate request information may be generated, and the request information selected by the user may be input to the idea creation platform model, or the generated request information may be input directly to the idea creation platform model without any selection or decision-making by the user to generate ideas. In other words, new ideas will be sequentially created by the idea creation platform model from the generated request information without the user's intervention.

[0105] In the third embodiment, company information may be input to the base model as a prompt. Examples of this company information include company name, year of establishment, business activities, location, representative, number of employees, capital, major customers, sales, corporate philosophy / vision, major products and services, history, IR information, etc. Such company information may be information found on the internet, or information read from other recording media, etc.

[0106] This company information can be entered manually as a prompt, or it can be imported as data. Furthermore, the company information may not only be publicly available information, but also internal documents that are kept confidential within the company.

[0107] In the third embodiment, in addition to prompts regarding ideas, company information is also input as a prompt. That is, in addition to prompts regarding desired ideas, such as "We want to miniaturize the structure of fuel cells," new ideas may be generated from the base model based on information about the future vision extracted from company information, such as "We have a five-year plan to apply fuel cells to the XX field," or predictive information that can promote those ideas may be generated. This predictive information can promote the ideas and also predicts the future state of the company based on the input company information.

[0108] In the third embodiment, the market simulator described above may be used. In this case, the market simulator is used to perform a simulation using the input prompts or forecast information output from the base model as parameters. Then, forecast information may be obtained based on the results of the simulation by the market simulator. [Explanation of Symbols]

[0109] 1. Idea Evaluation System 10 servers 11, 31 Computer Processors 12, 32 memory 15, 35 storage 20 Communication Networks 30 User terminals 40 Server-side programs 42 Terminal-side programs 51 Market Simulator 52 Natural Science Simulators 152 Evaluation Information Management Table

Claims

1. A prompt input step accepts prompt input containing text data describing the content of an idea for a base model, which is a large-scale language model that has been trained on text data describing the content of an idea and publicly known future prediction information that has been predicted about the future. In the prompt input step described above, predictive information consisting of text data about the future that can appeal to the above idea, output from the base model, is obtained in response to the prompt input received, and based on the obtained predictive information, request information regarding what is expected of the subject of the idea is generated. The above information acquisition step involves inputting the above request information into an idea creation platform model that has been pre-trained with patent information; The computer is to execute the following steps: in the request information input step, it takes the request information entered in the request information input step and, in the idea acquisition step, it acquires an idea that has been refined according to the prediction information output from the idea creation platform model, based on the request information entered in the request information input step. A predictive information generation program characterized by the following.

2. The system further includes an evaluation information acquisition step to obtain evaluation information that has been assigned an evaluation to the above idea, In the above information acquisition step, the above request information is generated based on the evaluation information acquired in the above evaluation information acquisition step. In the above idea acquisition step, further refined ideas are acquired based on the above evaluation information. A predictive information generation program according to claim 1, characterized by the following:

3. The above step of acquiring evaluation information involves acquiring evaluation information that includes an assessment of the marketability of the idea, or evaluation information that includes an assessment of the novelty or inventiveness of the idea. A predictive information generation program according to claim 2, characterized by the following:

4. In the prompt input step described above, the prompt is input to the base model that has learned the future prediction information, which includes information about predicted needs for the future. The above information acquisition step involves acquiring predictive information that forecasts the needs for which the above idea can be appealed, as output from the above base model. A predictive information generation program according to claim 1, characterized by the following:

5. In the prompt input step described above, further company information is entered into the base model. In the above information acquisition step, the above idea output from the above base model can be presented in response to the prompt entered in the above prompt input step, and predictive information that predicts the future state of the company based on the input company information can be obtained. A predictive information generation program according to claim 1, characterized by the following:

6. It further includes a market simulator that performs a simulation of marketability based on the input parameters, The above market simulator performs a simulation using the prompts entered in the above prompt input step, or the forecast information output from the above base model, as parameters. In the above information acquisition step, the above forecast information is obtained based on the results of the simulation by the above market simulator. A predictive information generation program according to claim 1, characterized by the following:

Images