Ultra-high reliability reflect generation ai and usage system thereof

Abstract

To provide a reflect generation AI system and a generation AI system utilization method to improve the accuracy of Japanese-language generation AI.SOLUTION: A reflect generation AI system improves accuracy of a response 31 by questioning AI2 again with an answer 3 obtained from a question 1 to AI2. More specifically, in addition to the AI's answer sentence and an original question word or sentence, it adds an input questioning whether truly appropriate, or not.SELECTED DRAWING: Figure 2

Description

[Technical field]

[0001] This invention relates to groundbreaking improvements in the accuracy of generative AI's problem-solving methods and means of using that AI efficiently, namely, methods for providing advice on problems, remembering the deceased, election campaign methods that even the elderly can use, election campaigns that allow voters to use AI on their smartphones, explanations of the contents of books, and instructions for using products. [Background technology]

[0002] Currently, it is possible to create various documents and paintings using generative AI. However, there are problems with accuracy, and the drawback of generative AI is that it "tells plausible lies." This is because generative GPT only compares and arranges AI vectors in several stages, making it poor at judging truth and falsity, and depending on how past AI data was selected. Furthermore, the generative AI has little Japanese data and is undertrained, making it inaccurate. When using Japanese generation AI, questions and answers are mainly conducted by text input, and oral consultations and questions can only provide simple answers in limited areas and the accuracy is low. When using AI to provide advice on problems, Japanese people ask verbally in Japanese, but when speech recognition is used, the meaning can shift and the speech can become text. Regarding election campaigns that utilize AI, a law amendment (legislator bill) in 2013 lifted the ban on election campaigns using the internet, etc. Based on this, activities using AI are also being carried out. While young people are easily accustomed to the Internet and can access information online, older voters are less likely to access information via the Internet than via paper or posters, and it is even more difficult for them to obtain information through question and answer sessions with AI. And when looking at the entire generation of voters, more voters are exposed to television, election posters, election advertisements in newspapers, and election bulletins than to election information on the internet. For example, in the 2020 Tokyo gubernatorial election, at most 30% of people used the special website set up by the election management committee or referred to information on the internet, while around 50% looked at the election bulletins and election posters distributed to their homes, and around 50% also referred to television and radio election coverage and political broadcasts. The current situation is that the use of the internet is still low. Paper, and the visual and audio of the candidates themselves still have a large influence. And elderly candidates do not run for office. This is because even people with clear minds and experience need to go through very hard election campaigns, which is physically difficult for elderly people. For example, the maximum age for a gubernatorial election is 88 years old when the term of office expires. By using the AI ​​of this invention remotely, elderly people can also campaign and serve as lawmakers. The present invention aims to eliminate these drawbacks, significantly improve the accuracy of AI, and make it easier to use. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Special table 2022-517292 [Patent Document 2] 2015―052992 [Patent Document 3] Patent Publication No. 2008-250367 [Patent Document 4] Patent Publication No. 2002-041718 Summary of the Invention [Problem to be solved by the invention]

[0004] Below are the questions for which answers were provided from the generating AI's own database. 1. There is little data in Japanese, so answers to questions in Japanese will be incorrect or will not be able to be answered. 2. The model may be biased, which may reduce its accuracy. 3. Still different from human language: Although natural language processing technology is used, the response is not perfect. 4. Does not use out-of-context information: Responses are based on algorithms, so they can only determine past conversation information, etc. 5. Logical inconsistencies: Responses sometimes have logical inconsistencies. 6. I feel unnatural when communicating with others.

[0005] Patent Document 1 describes a technology that improves the semantic expression of characters and increases the accuracy of text processing by adding a knowledge graph in machine learning. However, current algorithms for finding optimal answers fall short in this regard.

[0006] Patent Document 2 describes a technology for improving the accuracy of Japanese text processing in machine learning.

[0007] Patent document 3 is an invention filed by the present inventor, which discloses a tool for conducting election campaigns in which a two-dimensional code is placed on a sash or the like and an email address is read into it, allowing communication between voters and candidates via email. Moreover, in Patent Document 4, each candidate creates a database, and voters can access a designated information aggregation site via the Internet and ask questions and exchange opinions with the candidates through email and information about the candidates.

[0008] However, we now have high-speed Internet networks that enable high-speed video distribution, and we live in an age where everyone, even the elderly, has a smartphone.In addition, the Public Offices Election Act has been relaxed, so the means of transmitting information and exchanging opinions should also change. What is needed is a system that allows all voters, including the elderly, to use their smartphones to easily and safely answer questions individually over the Internet, and to learn more about candidates and political organizations.

[0009] The inventor has found a way to dramatically improve the accuracy of the insufficient performance of AI, and invented a method to dramatically improve its performance for the Japanese language.Furthermore, using the AI ​​of the present invention, he has invented an election campaign system that allows even elderly people to run for office, allows candidates to efficiently campaign without moving around using AI on the Internet, and enables voters of all generations to fully understand candidates and political organizations and gain a great deal of trust from voters. [Means for solving the problem]

[0010] The present invention, which solves the above-mentioned problems, is characterized by obtaining a generating AI answer that reflects the AI ​​answer infinitely and increases it to the infinite highest accuracy. In other words, the answer obtained by the generative AI is then further queried, such as "Is that okay?" or "Give us a better answer?", and by repeating this process n times, an even more accurate answer is obtained. Furthermore, the system features a means where input is done in Japanese, which is then automatically converted into English, and the AI's most accurate answer is then provided in Japanese. Using the generation AI of the present invention, Furthermore, by using the AI ​​of the present invention, which has been trained by incorporating information such as the candidate's claims, policies, career, information on awards, invention information, conversations and posts on SNS, interview information, family information, political claims, and career, it is possible to hold bulletin boards without election posters or street speeches without the candidate. In other words, when a voter uses a dedicated app on their smartphone to recognize the candidate's face, name, two-dimensional code, character, code, frame image, or character written in the bulletin board title, the candidate's data is identified via the Internet, and documents, images, videos, and audio are automatically downloaded and viewed, and after viewing and listening, the system is connected to the AI ​​of the present invention, and the voter can ask questions by voice and receive accurate responses. This allows even elderly people, those who are in poor health, those who cannot speak, or those who are unable to walk (hereinafter referred to as "elderly people, etc.") to run for office and campaign with minimal physical burden, and by utilizing the AI ​​of the present invention, voters of all ages can fully understand elderly candidates. Effect of the Invention

[0011] This has the effect of obtaining the most accurate AI answer that is likely to be the most correct. In addition, it is possible to input questions to the AI ​​in Japanese and receive the most accurate answer, also in Japanese, that is deemed to be the most correct. In addition, by using AI trained on candidate information in elections, (1) There will be no need to put up huge amounts of posters or hand out flyers. (2) Reduces the manpower and costs required for posting posters. (3) Candidates themselves will no longer need to put in the effort of campaigning. (4) The AI ​​will also answer questions on behalf of candidates or political organizations. (5) It saves candidates effort and expense. (6) Even the elderly and physically handicapped can easily run for office. They can utilize their capable brains. They can lower the hurdles of election campaigning. And as people with long experience, they can effectively utilize their knowledge and brighten up Japan. It is effective. [Brief description of the drawings]

[0012] [Figure 1] Publicly known generating AI (GPT) [Diagram 2] Flow diagram showing a first embodiment of the present invention. [Diagram 3] Flow diagram showing a second embodiment of the present invention. [Figure 4] A flow diagram showing a third embodiment of the present invention. [Diagram 5] A flow diagram showing a fourth embodiment of the present invention. [Figure 6]Flow diagram showing a fifth embodiment of the present invention. [Figure 7] Flow diagram showing a sixth embodiment of the present invention. [Figure 8] A flow diagram showing a seventh embodiment of the present invention. [Figure 9] FIG. 13 is a top view showing an eighth embodiment of the present invention. [Figure 10] FIG. 13 is a front view showing a ninth embodiment of the present invention. [Figure 11] A front view showing a tenth embodiment of the present invention. [Figure 12] Flow diagram showing an eleventh embodiment of the present invention. [Figure 13] FIG. 12 illustrates a twelfth embodiment of the present invention. [Figure 14] FIG. 13 illustrates a thirteenth embodiment of the present invention. [Figure 15] FIG. 13 is a diagram showing a poster example according to the thirteenth embodiment of the present invention. [Figure 16] FIG. 23 is a diagram showing a processing flow according to a thirteenth embodiment of the present invention. [Figure 17] FIG. 23 is a diagram showing a processing flow according to a fourteenth embodiment of the present invention. [Figure 18] FIG. 15 is a diagram showing an example of the publication of the 15th embodiment of the present invention. [Figure 19] FIG. 23 is a diagram showing a processing flow according to the fifteenth embodiment of the present invention. [Figure 20] FIG. 23 is a diagram showing a processing flow according to the sixteenth embodiment of the present invention. [Figure 21] FIG. 17 is a diagram showing the processing flow of the 17th and 18th embodiments of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0013] Figure 1 shows a known generation AI. In November 2022, OpenAI strongly requested the Japanese Prime Minister to adopt generation AI. It is expected to be used in various fields using big data, and regulations are also being considered, but in this invention, we focused on accuracy. From the database, we found that the publicly known generation AI has the following problems: 1. Models may be biased: Generative AI generates responses based on the data used to train it, so there may be bias. This can make the model less accurate. 2. Still different from human language: Generative AI is trying to understand human language using natural language processing technology, but it is not yet perfect and cannot generate perfect responses. 3. No out-of-context information: Generative AI generates responses based on algorithms, so it can only get information from the context of past conversations. Therefore, it cannot use out-of-context information. This results in delayed optimization of the latest information. 4. Logical inconsistencies: The responses generated by the generative AI may contain grammatical and logical inconsistencies. 5. Some people feel uncomfortable with machine language. Generative AI is not yet perfect, so it sometimes generates responses that sound like machine language. This can make it feel unnatural when communicating with people. Furthermore, even generative AI that has learned a huge amount of data can choose appropriate answers to questions and "seriously lie." 6. Because the information context is mainly English, the accuracy of response generation for Japanese is poor. The above problems are well known and have been pointed out.

[0014] The key points of publicly known generative AI (GPT) are 1. Deep Learning Model (1) Analyze it from the beginning to find its meaning. (2) The transformer improves the accuracy of determining the order of the data. 2. Vectorize each word. 3. Attention mechanism (processing information while paying attention to important elements) Something is important. AI derives relevance to help understand the meaning of the question. 4. The configuration of the transformer is shown in Figure 1(B). Repeat this process over and over again to understand the meaning and increase your strength, rather than just asking. And the information at the lower layers of a program is insufficient. It's only when you get to the top of the program that you get a good answer. And I'm not good at abstract things. It has the above functions. And the well-known generative AI (GPT) is characterized by using a large-scale language model with an attention mechanism and creating a table of contents for important parts rather than simply piling up a series.

[0015] Specifically, the algorithm (processing method) of the known generative AI, GPT (Generative Pretrained Transformer), is pre-trained with a large amount of text data using a transformer with the above-mentioned neural network structure. As shown in Figure 1(B), the transformer divides the input text into a string of words and characters, each of which is assigned positional information, which is then compressed, aggregated, and stratified as appropriate by the encoder. In order to understand the meaning and context of the text input to the encoder, the attention mechanism (self-attention function or multi-head attention function) calculates the relationships between each word and character, and using the information received from the encoder and self-attention, predicts the next word or character to be generated through the decoder, stratifies it, and outputs the answer. GPT then generates sentences that follow the input text. In the pre-training stage, GPT learns linguistic knowledge and patterns from large amounts of text data on the Internet (such as Wikipedia and news articles), and then performs the task of masking (hiding) parts of the text and predicting the masked parts from the remaining parts, thereby mastering the rules and grammar of general languages. After pre-training, a transfer learning stage is performed in which the model is fine-tuned to a specific task (such as a dialogue or summarization) and learns the knowledge and skills required for the task from text data related to the task (such as dialogue history and summaries), enabling the model to handle the specific task. Transfer learning involves feedback learning (so-called supervised learning) that prepares an output (answer) for an input (question), and further learning within GPT itself based on the above pre-learning (so-called unsupervised learning), thereby improving accuracy in multiple fields and enabling the generative AI to be applied in any field. For specialized matters, the answers generated by the GPT model will be evaluated by humans to further fine-tune the model.

[0016] The present invention is a technology that aims to improve the accuracy of answers to questions by using generative AI after it has gone through this type of learning. In particular, the present invention aims to convert language processing and translation processing into Japanese, which is highly reliable and allows sufficient communication with Japanese, and to generate highly accurate responses.

[0017] Hereinafter, an embodiment of the present invention will be described in detail with reference to the drawings. The idea behind this invention is to increase the reliability of an AI by repeatedly learning from a single question.

[0018] For example, a single convex lens can only magnify to a certain extent. However, if two convex lenses are stacked together, it becomes a microscope, allowing you to magnify even the smallest details. This invention is similar to stacking two convex lenses, and by stacking two or more huge amounts of AI data, it is possible to achieve higher accuracy than existing generative AI.

[0019] Furthermore, the conversation and investigation functions of known AI generators use big data to produce sentences similar to human conversations in English. In the case of Japanese, there is little accumulated data and there are problems with accuracy. In other words, the natural language processing technology of generative AI has insufficient Japanese language learning, and the importance of Japanese words is insufficient. There are also issues with accuracy and learning.

[0020] The deep learning (machine learning) method used in publicly known generative AI is a natural language processing model that is created and trained on an English database, so even if it is accurate in assigning importance to English words, it is thought that its learning for Japanese is insufficient. For example, even in OpenAI's multilingual model, the error rate for Japanese is 26% higher than the error rate for English. The present invention is a groundbreaking method for solving the above problems using a generative AI device.

[0021] Furthermore, the highly accurate Japanese response generation AI of the present invention can be used to develop a variety of activities and products. A specific processing form using this generation AI will be explained in the eighth and subsequent embodiments.

[0022] [First embodiment of the present invention] FIG. 2 shows a first embodiment of the present invention, which is an invention that improves the accuracy of answer 31 by asking AI a question, and then asking AI2 the answer 3 obtained by [word or character input] 1, in order to further increase the accuracy compared to known generation AIs. Then, the AI's answer is asked again, and in addition to the text of the AI's answer and the original question or text, an input is added to ask again whether it is really correct. In other words, words or text documents such as "Is this really correct?", "Re-investigate", and "Give us another reliable answer" are added. This changes the weighting of the words, leading to a higher-level data and a more accurate answer. Unlike known AI, the question setter does not simply ask a question and the AI ​​simply gives a low-level answer, nor does the question setter prepare the answer and have the AI ​​learn the errors. By having the AI ​​itself look up the content in more detail in the database, it can match the perception of the researcher as closely as possible, resulting in a more reliable answer31. This was named the reflection AI method.

[0023] [Second embodiment of the present invention] FIG. 3 shows a second embodiment of the present invention, characterized in that the input language of questions to the AI ​​2 is a language 11 other than English, such as Japanese 11. This is because the majority of publicly available generation AI data is in English12, and there is very little data in Japanese. The present inventor has invented a method for obtaining highly accurate answers using this Japanese input 11, and has filed a separate patent application for this invention. This second embodiment utilizes the inventor's invention to further improve the accuracy of the known generation AI.

[0024] [Third embodiment of the present invention] FIG. 4 is a flow diagram showing a third embodiment of the present invention. By repeating the step of “asking the AI ​​for its answer again” 4 in the first embodiment n times, answer n becomes more substantial and more accurate than answer 1. When "asking the AI ​​for its answer n times," it is also possible to break down the words in the AI's answer, highlight those words, and ask the AI ​​again. The questioner can decide which words are important, or they can use text mining techniques to determine this, and then highlight them and ask the AI ​​again. You may also add the above mentioned question again. Furthermore, in known natural language processing, the present invention also includes redirecting the embedding vector (weight) of a word with a small vector (weight) and investigating the language. According to the third embodiment of the present invention, even a difficult answer can be recognized more easily.

[0025] [Fourth embodiment of the present invention] FIG. 5 is a flow chart showing a fourth embodiment of the present invention, which is an invention for obtaining a more accurate answer by using AI2a, 2b where a and b are different. The AI2a of a and the generated AI2 (not shown) repeat the question and answer 5. At the same time, the AI2a of a also asks the AI2b of b the same question and answer document 5 from the generated AI2 n times. If the answer 5 obtained when the AI2b of b is used differs from the answer 5 given by the generated AI2 using the AI2a of a, the AI2a and AI2b continue asking each other. This invention allows the generation AI2 to obtain an answer with the highest accuracy using the input information.

[0026] [Fifth embodiment of the present invention] FIG. 6 shows a fifth embodiment of the present invention, in which a question 61 is input to the generation AI in Japanese, which is then passed through an automatic translation 621 from Japanese to English, and input 63 to the generation AI in English. The AI's response 65 is then passed through an automatic translation 622 from English to Japanese, and the response 666 is given in Japanese. The present invention does not involve a trainer (teacher) simply providing words and having the translation model learn the errors. Since the generating AI responds based on a huge amount of data in English, the present invention is able to provide highly reliable questions and answers.

[0027] According to the fifth embodiment of the present invention, the accuracy of AI responses can be improved even when input is made in Japanese.

[0028] [Sixth embodiment of the present invention] FIG. 7 shows a sixth embodiment of the present invention, in which input 63 is made to an AI in English via a Japanese to English translation function 671 with improved translation accuracy, and the output 65 from the AI ​​is output 66 in more accurate and better Japanese by a translation function 672 with improved translation accuracy from English to Japanese.

[0029] In order to achieve a higher level of accuracy than known translation functions, a question is posed to the translation function, and the answer obtained in English or Japanese is posed again to the translation function. In other words, re-examining the translation means repeatedly back-translating and forward-translating the translated words. By repeatedly translating and adding importance, the meaning and purpose of the words are clarified and the accuracy is improved. This results in different weighting of the words, leading to a more accurate and precise translation.

[0030] According to the sixth embodiment of the present invention, the response accuracy of a known AI device can be significantly improved when inputting Japanese. It is expected that the accuracy of any known generation AI will improve. Moreover, it will be easier to input.

[0031] [Seventh embodiment of the present invention] FIG. 8 shows a seventh embodiment of the present invention, which is a processing method in which high-accuracy response systems 6101, 6102 are added to the speech recognition systems 681, 682 whose accuracy has been improved as described above. A Japanese utterance 611 is input, and translation 681 and character conversion 691 are performed with improved speech recognition accuracy. The output text, that is, question 63, is then subjected to reverse speech recognition again into a Japanese utterance. Based on the second Japanese utterance, the questioner further speaks in Japanese, emphasizing the important points. For example, by inputting three different English sentences into the generation AI 6101, the meaning and intention of the translated words from Japanese are added to improve accuracy.

[0032] The automatic translation may be a character and / or speech translation model with improved translation accuracy as in the seventh embodiment, or any known translation model is also included in the present invention. In addition to Japanese, the present invention also includes languages ​​such as Chinese, Korean, Spanish, and German.

[0033] According to the seventh embodiment of the present invention, questions can be asked in Japanese and anyone can obtain highly accurate answers extremely easily.

[0034] The eighth and subsequent embodiments show specific embodiments of the present invention that utilize the highly accurate Japanese response generation AI of the present invention.

[0035] [Eighth embodiment of the present invention] The eighth embodiment of the present invention is as shown in FIG. 9, and specifically, is an invention for what may be called a "confession room" or a "consultation center" such as a church's "confession room" or a toilet room 761. As shown in FIG. 9, a box-shaped room 6 such as a toilet 761 with a sealed door 762 enclosed on all sides or a space partitioned off with curtains is prepared, and when a person seeking advice 763 sits on the toilet 761 and talks about his or her worries, a microphone 773 at the back collects the sound and inputs it into the device main body 771 of the present invention installed under the toilet water tank, and the answer to the trouble is spoken back to the person seeking advice 763 through a speaker 772 at the back. The content of the present invention is not about atonement for the past, but rather, it allows one to obtain answers to questions about unknown matters such as future hopes and unknown anxieties, which makes the present invention popular. The location of the installation device of the present invention may be an unenclosed area. This allows not only past data but also future information to be stored in the database, further increasing accuracy. The device 77 of the present invention (main body 771, microphone 773, and speaker 772) is a device having a server main body with a highly accurate Japanese response generation AI trained on life consultation cases such as those shown in embodiments 1 to 7, and it is preferable that the main body, microphone, and speaker are integrated into one unit so that the device can be installed in a private room, etc. A QR code (registered trademark) can be placed in this sealed room, read by a smartphone, and connected to the generating AI installed in the main body of the device of the present invention installed in another location, allowing wireless consultation and vocal responses. In more detail, the device 7 of the present invention also includes, as a unit, an Internet-connected device (such as a personal computer or smartphone) having an application that reflects answer queries, a display device such as a display, and power devices such as a battery and a power cord. When this is used for business purposes, a coin box 78 is provided as an example of the device of the present invention as shown in FIG. 9, and when a user inserts money into this box, the device main body 771 of the present invention is activated.

[0036] [Ninth embodiment of the present invention] In the ninth embodiment of the present invention, as shown in FIG. 10, a main body 771 of the device of the present invention, a microphone 773, and a speaker 772 are placed on top of a grave 79 or inside a stone, and when visiting the grave, one can talk to the deceased and get a response, and can also talk to the deceased about their current worries, listen to their concerns, and exchange love. This invention will completely change the concept of graves in the future. When installing the device 77 of the present invention at a grave, it may be a portable device that can be moved by integrating the main body 771, microphone 773, and speaker 772. Or, when installing it as an all-weather type that can withstand wind and rain, a system that can supply power from solar power is provided. In addition, a QR code (registered trademark) could be placed on the gravestone, which could then be read by a smartphone and used to connect to a generating AI installed in the main body of the device of the present invention installed in another location, allowing people to talk wirelessly or ask for advice about their worries.

[0037] [Tenth embodiment of the present invention] The tenth embodiment of the present invention is a Buddhist altar. As shown in Fig. 11, this is a completely new invention of a Buddhist altar 710 consisting of the main body of the device of the present invention 771, a microphone 773, a speaker 772, a memorial tablet, and a portrait of the deceased. This is an invention of a revolutionary Buddhist altar that allows you to talk with the deceased in the same way as the tomb described above. As with the above invention, this invention will bring about a religious revolution with the "room", "tomb" and "altar". A QR code (registered trademark) can be placed on the Buddhist altar of the present invention, which can then be read by a smartphone and used to connect to a generating AI installed in the main body of the device of the present invention installed in another location, allowing users to talk wirelessly or seek advice on problems.

[0038] [Eleventh embodiment of the present invention] The eleventh embodiment of the present invention, as shown in FIG. 12, is an invention in which the business model of a generation AI implementing an embodiment of the present invention is such that, upon payment of a fee set lower than the subscription fee, a consultation is received from a client via voice, the information is converted into text and proofread, and an answer is obtained by the AI ​​of the present invention implementing an embodiment of the present invention, which is then converted into voice and spoken to the client. According to the eighth embodiment of the present invention, there is an advantage that a person seeking advice only needs to make a request and can immediately receive a voice response from an expert.

[0039] The method of the embodiment of the present invention may be used as a method for training an AI. The automatic document translation and automatic speech translation in the present invention are character translation and speech translation with AI functions, but known technologies may also be used.

[0040] [Twelfth embodiment of the present invention] FIG. 13 is an overall view showing a twelfth embodiment of the present invention, illustrating a specific method of using the AI ​​2, 64 generated by the present invention in an election campaign 80. Basically, candidates do not move around. Voters use smartphones. The feature of this system is that they can work remotely. Furthermore, candidates do not have to put up posters on bulletin boards. They do not have to ride in campaign cars. Candidates do not have to give street speeches in person. That is, the election campaign will involve question and answer communication with each voter's smartphone, and the AI ​​of the present invention will automatically generate written or voice responses via the Internet. This will enable remote working of candidate election campaigns.

[0041] An overall diagram of a method using this generated AI of the present invention is shown in Figure 13. This method is characterized in that the generated AI of the present invention is connected to the generated AI 2,64 system by holding a voter's smartphone, which has a dedicated app installed, over the display of a link connection function such as a QR code (registered trademark) provided on election-related items such as posters and election bulletins during the election campaign 80, or over the face of a candidate or the numbers, frames, or letters at an election poster display area, and the generated AI 2,64 system of the present invention, which has been trained with candidate information, responds in text or voice to questions asked by candidates or other questioners. First, the differences between known election campaigns and the election campaign 80 of the present invention are listed in a table. TIFF2025009955000002.tif206137

[0042] Regarding the construction of the software, previously, only policies were posted on the website, Candidates will create dedicated smartphone apps capable of facial recognition, image recognition, character recognition and 2D code recognition. Then they will distribute it to voters and encourage them to install it. The server constructs the training AI 2, 64 of the present invention using the candidate information. By using the present embodiments 1 to 7, the AI ​​2, 64 with high accuracy can be generated, which can answer questions by voice (in Japanese). As for the bulletin board, basically, even if a voter does not put up a poster in the "poster display area 81", they can just start up a dedicated app on their smartphone and hold it up to the bulletin board, and download and answer questions using the AI2, 64 generated by the present invention. This will have the effect of reducing labor costs, as candidates will not have to go to the trouble of creating large numbers of posters and pasting them up in several thousand locations across Tokyo, for example. Regarding the election bulletin 83 and the election advertisements in newspapers 82, the names and photographs of the candidates, two-dimensional codes, and characters can be converted into AI 2.64 generated by the present invention, or into a homepage, database, image, video, or other data using a dedicated app.

[0043] In the past, candidates were only seen and heard at election cars, street speeches 85, political broadcasts 84, individual lectures, and debates 86 at the venue, but these will now be basically remote, and voters will access the AI ​​2.64 generated by the present invention on their smartphones via a link provided by a two-dimensional code, the candidate's face, or the text of the candidate's campaign promises, to answer questions. In addition, an image of the candidate may be displayed, or a clone android (registered trademark) or avatar may be created that resembles the candidate, and the candidate may answer questions using voice using the AI ​​generated by the present invention 2.64. Additionally, audio may be communicated as a live two-way broadcast between candidates and voters at the venue.

[0044] The above-mentioned dedicated apps may be linked to apps of known communication tools (LINE, YouTube, Instagram, X (formerly Twitter (registered trademark)), Facebook (registered trademark), TikTok, WhatsApp, WeChat, LinkedIn, Snapchat, Pinterest, or KakaoTalk). The communication means may be text and / or video and / or audio, and may be used for communication between voters and candidates themselves. Furthermore, the generated AI 2, 64 is linked to a publicly known AI such as a chat GPT AI, and trained with candidate information using an embodiment of the present invention (using the server of the linked AI system), and a question and answer session is conducted with this AI.

[0045] By using the above-described method and device to carry out an election campaign 80, and by providing a microphone, camera, database storage device, communication server computer, and other electronic devices in the home of a candidate, even an elderly person can run for office, make a claim, and be elected. In addition, the database storage device and / or communication server computer and other electronic devices may be provided in the cloud.

[0046] [Thirteenth embodiment of the present invention] 14, 15, and 16 are diagrams showing a thirteenth embodiment of the present invention, which is a more specific version of the twelfth embodiment of the present invention. FIG. 14 shows an election poster bulletin board 81, FIG. 15 shows an example of an election poster with a two-dimensional code 87, and FIG. 16 shows the processing flow of the bulletin board. First, the voter starts a dedicated application installed on his / her smartphone, and the posted number (code) and information on whether or not an election poster is attached are displayed from a database 89. The smartphone camera starts up, and when the smartphone is held up over the election poster display area 80, the image of the square display frame 813, the code (for example, a small number) 812 written in the frame, the code (a large number) 811 written in the center of the candidate's frame, or the text title 814 of "Poster Display Area" or "XX Governor Election" is captured and recognized as a shape, number, or character. Also, if a candidate's election poster is posted, the two-dimensional code 87 on the poster is recognized. Then, by comparing with the database 89, the server determines where the voter is looking at what (e.g., election poster board, street speech, election advertisement in newspaper, etc.), and automatically downloads data designated in advance for the location to the smartphone. For example, the processing of this dedicated app at the poster board location can, for example, download and display a document file such as a candidate's poster, and then have a voice question and answer session with the AI ​​88 of the present invention trained with the candidate's information. The processing procedure of such a dedicated app may be varied depending on the conditions as appropriate, or may be determined in advance by the voter. This makes it possible to obtain highly accurate candidate information from the generation AI of the present invention. As mentioned above, even if candidates do not put up posters, voters can easily obtain a lot of information about candidates or political organizations.

[0047] In addition, the app downloaded to the smartphone may be a dedicated app exclusively for a particular candidate or political organization, or it may be an app issued, for example, by the election management committee that allows links to all candidates or political organizations. When the poster is up and running, the link can be read using a smartphone's well-known 2D code reader. Also, a question and answer session may be held by voice with the AI ​​88 generated by the present invention directly from the candidate's homepage.

[0048] [Fourteenth embodiment of the present invention] FIG. 17 is a process flow diagram showing a 14th embodiment of the present invention, which is a more specific example of the 12th embodiment, and shows the process flow of a dedicated application for an election car or a street speech 85. Voters hold their smartphone camera over the candidate's face (candidate's face photo, portrait, name, or 2D code) after launching a dedicated app, and the server compares the face and character recognition with the database 89, and the server determines where the voter is and what they are looking at, and sends them the materials. For example, if the candidate is making a street speech, it automatically and instantly downloads election bulletins and videos of policies. It then connects to the AI ​​88 of the present invention, which has been trained (data accumulated) with the candidate's information, and answers questions by voice. In addition, after linking with this dedicated app, you can also communicate with an avatar of the AI88 generated by the present invention that is trained with candidate information that resembles the virtual candidate. It can also be broadcast live with remote candidates.

[0049] By using the AI ​​88 of the present invention trained on candidate information, it is possible to provide highly accurate question and answer responses, and the physical burden on the candidate himself can be significantly reduced.

[0050] The candidate information includes knowledge information or information on political organizations, such as the candidate's career, honors, invention information, conversations and posts on SNS, interview information, family information, claims of political organizations, career, etc. Each of these pieces of information is input to the AI ​​88 of the present invention, and the AI ​​is trained using the methods of the first to seventh embodiments to enable it to respond appropriately.

[0051] [Fifteenth embodiment of the present invention] 18 and 19 show a fifteenth embodiment of the present invention, in which FIG. 18 is an example of an election bulletin 83 having a two-dimensional code 87, and FIG. 19 is a processing flow diagram from an election advertisement 82 published in a newspaper and the election bulletin 83. When a dedicated app installed on a smartphone is started and the smartphone is held over the face, name, and two-dimensional code 87 on the list of candidates in the election bulletin, the image or text is recognized, and the server determines that the voter is looking at a document such as the election bulletin by referring to a candidate information database 89, and instantly and automatically begins sending that candidate's poster materials, political policy materials (election bulletin, newspaper advertisements), and video materials to the smartphone, after which it connects to the present invention's generation AI 88, which has been trained on the candidate information, and a question-and-answer session is held with the voter. In addition, this process may involve having the voter's smartphone recognize the facial image of a specific candidate in a televised political broadcast, connecting to the present invention's generated AI 88 trained on candidate information, and conducting a question-and-answer session with the voter.

[0052] [Sixteenth embodiment of the present invention] FIG. 20 is a process flow diagram showing a sixteenth embodiment of the present invention, This is a flow chart showing the process of holding candidate lectures and debates online. Voters start up a dedicated app installed on their smartphones, click on the link to the lecture or debate information, and participate at the time notified in advance by the candidate or political organization. Voters can attend lectures or debates in person or at home using a dedicated app on their smartphone or computer. For example, if a voter is at a lecture venue, they can use a dedicated app to point their smartphone camera at a candidate's face or two-dimensional code, and the content of the lecture will be instantly transmitted. The AI ​​will then connect to the present invention's generation AI 88, which has been trained on the candidate's information, and a question-and-answer session can take place between the voter and the AI. In addition, after linking with this dedicated app, you can also communicate with an avatar of the AI88 generated by the present invention that is trained with candidate information that resembles the virtual candidate. It can also be broadcast live with remote candidates.

[0053] Candidates can use a microphone and camera at home to provide video and audio to a video and audio display device at a lecture hall or forum, or to voters' smartphones or computers. Live communication can be conducted via two-way video and audio. Candidates may choose to have pre-recorded audio of their speeches streamed to voters at the lecture or debate venue and on the candidates' smartphones, with voters only participating and responding online during question and answer sessions. This reduces the burden on candidates as they only need to respond.

[0054] Furthermore, by using the generated AI 88 and avatar of the present invention, the physical burden on the candidate himself can be significantly reduced.

[0055] [Seventeenth embodiment of the present invention] Figure 21 shows the 17th embodiment of the present invention, in which instead of a dedicated app, a dedicated website is prepared on a server, and voters can use a publicly known reading link function such as a two-dimensional code 87 on a candidate's flyer or poster, or use a browser search to go to the dedicated website on the server and engage in question-and-answer sessions with a trained AI 88 (e.g., ChatGPT) generated by the present invention that has imported the candidate's knowledge information, political philosophy, policies, basic education, or information on political organizations (i.e., candidate data information). With the above-mentioned trained AI knowledge, it can respond to questions from people all over the world and can represent candidates, especially in political matters. The inventor names this system "Answer Mars."

[0056] [Eighteenth embodiment of the present invention] The 18th embodiment of the present invention shown in Figure 21 is a system that further improves the accuracy of the 17th embodiment, in which information on questions and answers exchanged between a trained generation AI 88 of the present invention that has taken in candidate data and information on desired issues expressed by voters is converted from voice to text and stored in a database 91, and the data is organized and consolidated from the database 91 using another generation AI 90 of the present invention that has been trained in politics and policies and is different from the trained AI that has taken in the candidate data, to generate countermeasures, which are then fed back to the candidate information database 89 and the trained AI 88 of the present invention.

[0057] The feedback system of FIG. 18 will now be described in detail. In the question and answer session of the trained present invention generating AI 88 that incorporates voter and candidate data, the present invention generating AI judges whether the meaning of the voter's voice is a question or doubt for the candidate, or a request, complaint, issue, or problem for the whole. In other words, it classifies the matter into whether it should be answered or not. Then, it answers the question, and responds by asking for detailed information about requests, complaints, issues, and problems in order to draw out the specific content from the voter. All of the conversation (text or voice) is converted into text and saved as a database 91. Then, another present invention generating AI 90 trained with social information such as political policies organizes all the information into questions, doubts, or complaints, and further sifts it into policy matter classifications. Then, the other present invention generating AI generates a policy countermeasure plan. After the candidate checks the countermeasure plan, the candidate revises the data 89 related to the candidate's policies and guidelines, and the present invention generating AI 88 trained with the candidate information further updates the generating AI using the methods of the present invention embodiments 1 to 7.

[0058] By circulating the above information, it is possible to minimize the discrepancy between the candidate policies generated by the AI ​​of this invention and the wishes and issues of voters, further improving the candidates' policies and making it clear what election candidates should do as politicians, which will have a very beneficial effect on the country.

[0059] [Nineteenth embodiment of the present invention] In the ninth embodiment of the present invention, a two-dimensional code is placed on the cover or cover of a printed material such as a magazine, book, or other publication. When a reader holds a smartphone over the code, the reader is connected to a trained generation AI of the present invention that has incorporated the knowledge information, philosophy, and policies of the author or editor of the printed material, as well as a description of the book's contents, and can engage in a question and answer session. According to the nineteenth embodiment of the present invention, reading printed matter such as books can be made even more enjoyable.

[0060] [Twentieth embodiment of the present invention] In the twentieth embodiment of the present invention, a two-dimensional code is placed on a product or its instruction manual, and when a reader holds a smartphone over the code, the reader can connect to the generative AI of the present invention, which has been trained on detailed information, structure, and descriptions of the product, and can exchange questions and answers in text or voice. According to the twentieth embodiment of the present invention, the product is simple and easy to handle and understand.

[0061] The invention shown in the 20th embodiment of the present invention is not limited to products or documents having two-dimensional codes, but includes all information that needs to be explained to a third party with a smartphone, such as structure, handling, etc., that can be connected to a trained generation AI of the present invention via a smartphone and used for question and answering.

[0062] The AI ​​of the present invention may be trained by adding, for example, information on the same deceased individual or past factual information as an IC chip, as shown in Patent Application No. 2022-136374 invented by the present inventor. This allows, for example, in the case of an election, information such as the policies and examples of a long-standing political party to be input, enabling a highly accurate response as a political party.

[0063] All generation AIs that are modifications of the present invention are included in the present invention. [Industrial Applicability]

[0064] The present invention provides a foundation for further industrial development and has great industrial applicability. [Explanation of symbols]

[0065] 1 question 2 AI 2n n types of AI 3 answers 31 High precision answer 32 High-precision answers using English input High-precision answers by repeating 3n questions 4. Generate answer n-1 and question n (n=2,3,4...) 5 questions and answers 61 Questions in Japanese 611 Questions in Japanese (audio) 62 Machine Translation 621 Japanese to English automatic translation function 622 English to Japanese automatic translation function 63 English Questions 64 Generation AI 65 English Answers 66 Answers in Japanese 661 Answer in Japanese (audio) 67 High-precision translation 671 Japanese to English automatic translation function 672 English to Japanese automatic translation function 68 High-precision voice translation function 681 High-precision speech translation (Japanese to English) 682 High-precision speech translation (Japanese to English) 691 Voice to Transcription 692 Text-to-speech 6100 High precision answer function 6101 High precision answer (question) 6102 High precision answer (answer) 76 The invention of the "confession room" of the church, the toilet room 61, etc. 761 Toilet 762 Doors 763 Counselor 77 Device of the present invention 771 Main unit 772 Speaker 773 Mike 78 Fare Box 79 The Grave of the Invention 710 Buddhist altar of the present invention 711 Questions and answers in Japanese written or spoken language 712 Written questions and answers in English 713 Automatic document translation and automatic speech translation 80 Election Campaign 81 Bulletin Board 811 Poster number (large) 812 Poster number (small) 813 bulletin board 814 Text titles such as "··· Election Poster Display Area" 82 Newspaper election advertisements 83 Election Bulletin 84 Political Broadcasts 85 Election cars and street speeches 86 Lectures and discussions 87 2D Code 88 AI of the present invention trained (accumulated learning) with candidate information 89 Candidate Information Database 90 AI trained on politics, policy information, and current affairs 91 Requests and questions were categorized in detail. Voter Information Database

Claims

1. A reflect generation AI system that memorizes and stores the thoughts of respondents in advance, and outputs the answers from the AI ​​to inquires such as voters or product purchasers.

2. A reflect generating AI system as claimed in claim 1, characterized in that an intermediate code to the generating AI, such as a QR code (registered trademark), is provided on posters, election bulletins, products, etc.

3. The reflective generation AI system of claim 1 is characterized in that the answer obtained by the generation AI is further queried to the AI, such as "Is that okay?" or "Give me a better answer", to obtain an answer.

4. 4. The reflect generation AI system according to claim 3, wherein the question is repeated n times.

5. A reflective generation AI system as claimed in any one of claims 1 to 4, characterized in that a question in a language other than English, such as Japanese, is first converted into English through a translation system, the question is then asked in English to the generation AI, and the English answer is then output in a language other than English, such as Japanese.

6. A method of using a generative AI system in any one of claims 1 to 5, comprising providing a device having the generative AI system that stores life consultation information, and an object that can surround a microphone and speaker, so as to respond to the consultation of the client while taking privacy into consideration.

7. In claims 1 to 5, a method of using a generative AI system that allows communication with the deceased and ancestors by placing a device having the generative AI system that stores information about the deceased and ancestors, a microphone, and a speaker on a grave, inside a stone, or on a Buddhist altar.

8. A method of using a generative AI system in any one of claims 1 to 5, characterized in that a connection function such as a QR code (registered trademark) is provided on election-related materials such as posters and election bulletins during election campaigns, and the generative AI system responds to inquirers in text or voice.

9. A method of using a generative AI system in any one of claims 1 to 5, wherein the election-related item is an election poster display area, and the generative AI system responds to the inquirer in text or voice by holding a communication device such as a smartphone of a voter with a dedicated app installed over the numbers or frames on the display area, or the letters on the election poster display area.

10. A method of using a generative AI system in any one of claims 1 to 5, characterized in that a connection function such as a QR code (registered trademark) is provided on the cover or cover of a book or magazine, and the generative AI system responds to the inquirer in text or voice.

11. A method of using a generative AI system in any one of claims 1 to 5, characterized in that a connection function such as a QR code (registered trademark) is provided on a product or its instruction manual, and the generative AI system responds to the questioner in text or voice.

12. A method of using a generative AI system, as claimed in claim 5 or 6, characterized in that a first generative AI system classifies the questions of an inquirer as to whether or not they require a reply, answers those questions that require a reply, a second generative AI system classifies the questions that do not require a reply into categories, formulates and selects countermeasures for each category, accumulates information on the countermeasures in the first generative AI system, and improves the accuracy of the answers of the first generative AI system.

Images