Ultra-reliable reflect generative ai and utilization system therefor

Abstract

[Problem] There is a need to increase the accuracy of generative AI in relation to the Japanese language. [Solution] In the present invention, reflect generative AI and generative AI having improved accuracy in Japanese / English translation facilitate, inter alia, problem consultation using a microphone and a speaker, conversations with the deceased, election activities using a connection function such as a QR code (trade name), and also facilitate content consultation related to books and magazines and explanations related to product manuals.

Description

Ultra-reliable AI for reflect generation and its application system

[0001] The present invention relates to a groundbreaking improvement in the accuracy of generative AI problem-solving methods and means for efficiently utilizing this AI, namely, consultation services, methods for 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 book contents, and product usage instructions.

[0002] Currently, it is possible to create various documents and paintings using generative AI. However, accuracy is an issue, and a drawback of generative AI is its tendency to "tell plausible lies." This is because generative GPTs only compare and align AI vectors in several stages, resulting in poor accuracy in determining truthfulness and depending on the selection of past AI data. Furthermore, generative AI has limited Japanese data and is undertrained, resulting in inaccuracies. When using Japanese generative AI, questions and answers are primarily conducted via text input. Oral consultations and Q&A sessions only provide simple answers in limited areas and have low accuracy. When using AI for consultations, Japanese people usually speak in Japanese, but speech recognition can sometimes result in mistranslation and the resulting text. Regarding election campaigns using AI, a 2013 law amendment (legislative bill) allowed election campaigns via the internet and other methods. Based on this, AI-based campaigns are also being carried out. While young people are easily accustomed to the internet and access information online, older voters are less likely to browse information online than through paper or posters, and even more difficult to obtain information through AI-powered question-and-answer sessions. Looking at the entire electorate generation, more voters are likely to view election information on television, election posters, newspaper election advertisements, and election bulletins than online. For example, in the 2020 Tokyo gubernatorial election, only about 30% of voters used the special website set up by the Election Commission or consulted online information, while only about 50% read the election bulletins and election posters distributed to their homes. About 50% also consulted television and radio election coverage and political broadcasts. The current situation is that online utilization is still low. Paper and the visual and audio influence of candidates remain dominant. Furthermore, older voters are less likely to run for office. This is because even those with sharp minds and experience require extremely intense campaigning, which is physically demanding for older voters. For example, in a gubernatorial election, the maximum age at the end of one's term of office is 88 years old. By using the AI ​​of this invention remotely, even elderly people can participate in election campaigns and parliamentary activities. This invention eliminates these drawbacks, dramatically improves the accuracy of AI, and makes it easier to use.

[0003] Patent Publication No. 2022-517292, Patent Publication No. 2015-052992, Patent Publication No. 2008-250367, Patent Publication No. 2002-041718

[0004] The following are problems with generating answers from the AI's own database. 1. There is little Japanese data, so answers to Japanese questions may be incorrect or may not be able to be answered. 2. There is a possibility that the model may be biased, which may reduce the accuracy of the model. 3. It is still different from human language: Although natural language processing technology is used, the response is not perfect. 4. It does not use information outside of the context: As the response is based on an algorithm, it can only determine information such as past conversations. 5. Logical inconsistencies: The response may sometimes have logical inconsistencies. 6. It feels unnatural when communicating with people.

[0005] Patent Document 1 describes a technology that adds a knowledge graph to machine learning to improve the semantic representation of characters and increase the accuracy of text processing. However, this is insufficient for current algorithms that find optimal answers.

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

[0007] Patent document 3, an invention filed by the inventor of the present application, discloses a tool for conducting election campaigns in which a two-dimensional code is attached to a sash or the like, an email address is read, and voters and candidates communicate via email. Also, Patent document 4 discloses a system in which each candidate creates a database, and voters can access a designated information aggregation site via the Internet and ask questions or exchange opinions with candidates through email and information on the candidates.

[0008] However, with the development of high-speed internet networks that can stream videos at high speeds, and the current era in which everyone, even the elderly, has a smartphone, and the Public Offices Election Act has been relaxed, the means of transmitting information and exchanging opinions must also change. What is needed is a system that allows all voters, including the elderly, to use smartphones to easily and safely answer questions individually in an internet environment and learn more about candidates and political organizations.

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

[0010] The present invention, which solves the above-mentioned problems, is characterized by obtaining a generative AI answer that infinitely reflects the AI's answer and increases it to the highest possible accuracy. In other words, the answer obtained by the generative AI is further queried, such as "Is that good enough?" or "Give me a better answer?", and this process is repeated n times to obtain an even more accurate answer. Furthermore, the present invention features a means for inputting input in Japanese, automatically converting the input to English, and then returning the AI ​​answer in Japanese with the highest possible accuracy. Using the generative AI of the present invention, the AI ​​can be trained by incorporating various information, such as a candidate's claims, policies, career, honors, invention information, conversations and posts on social media, interview information, family information, political claims, and career history, enabling the AI ​​to be used for bulletin boards without election posters or street speeches without candidates present. In other words, when a voter uses a dedicated app on their smartphone to recognize a candidate's face, name, two-dimensional code, character, symbol, frame image, or character written in a bulletin board title, the candidate's data is identified via the Internet, and materials, images, videos, and audio are automatically downloaded and viewed, after which the voter connects to the generation AI of the present invention, and questions are asked by voice, allowing the voter to receive accurate responses. This allows even elderly, sickly, mute, or disabled people (hereinafter referred to as "elderly, etc.") to run for office and campaign with minimal physical strain, and by using the AI ​​of the present invention, voters of all ages can fully understand elderly and other candidates.

[0011] This has the effect of enabling us to obtain the most accurate and most precise AI answers. Furthermore, we can input questions to the AI ​​in Japanese and receive the most accurate and most precise answers, also in Japanese. Furthermore, by using AI trained with candidate information in elections, (1) the need for a huge number of posters and flyers will be eliminated. (2) The manpower and costs required for postering will be reduced. (3) Candidates themselves will no longer need to go to the trouble of campaigning. (4) Furthermore, the AI ​​will answer questions on behalf of the candidate or political organization. (5) This will reduce the candidate's labor and costs. (6) Even physically disabled people, such as the elderly, can easily run for office. They can utilize their capable minds. 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's effective.

[0012] Publicly known generative AI (GPT)Flow diagram showing a first embodiment of the present inventionFlow diagram showing a second embodiment of the present inventionFlow diagram showing a third embodiment of the present inventionFlow diagram showing a fourth embodiment of the present inventionFlow diagram showing a fifth embodiment of the present inventionFlow diagram showing a sixth embodiment of the present inventionFlow diagram showing a seventh embodiment of the present inventionTop view of an eighth embodiment of the present inventionFront view of a ninth embodiment of the present inventionFront view of a tenth embodiment of the present inventionFlow diagram showing an eleventh embodiment of the present inventionA diagram showing a twelfth embodiment of the present invention.A diagram showing a thirteenth embodiment of the present invention.A diagram showing a poster example of the thirteenth embodiment of the present invention.A diagram showing a processing flow for the thirteenth embodiment of the present invention.A diagram showing a processing flow for the fourteenth embodiment of the present invention.A diagram showing a publication example of the fifteenth embodiment of the present invention.A diagram showing a processing flow for the fifteenth embodiment of the present invention.A diagram showing a processing flow for the sixteenth embodiment of the present invention.A diagram showing processing flows for the seventeenth and eighteenth embodiments of the present invention.

[0013] Figure 1 shows a known generative AI. In November 2022, OpenAI strongly requested the Japanese Prime Minister to adopt generative AI. Big data is expected to be used in various fields, and regulations are also being considered. However, this invention focuses on accuracy. Database analysis revealed the following problems with known generative AI: 1. Potential for model bias: Generative AI generates responses based on the data used for training, which may result in bias. This can reduce the accuracy of the model. 2. Not yet identical to human language: Generative AI attempts to understand human language using natural language processing technology, but it is not yet perfect and cannot generate perfect responses. 3. Does not use out-of-context information: Generative AI generates responses based on algorithms, so it can only obtain information from the context of past conversations. Therefore, it cannot use out-of-context information. This delays the optimization of the latest information. 4. Logical inconsistencies: Responses generated by generative AI may have grammatical and logical inconsistencies. 5. Some people find machine language unfamiliar. Generative AI is not yet perfect, so it sometimes generates responses that sound like machine code. This can make it feel unnatural when communicating with people. Furthermore, even generative AI that has learned a huge amount of data can choose random answers to questions and end up "seriously lying." 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] Key features of publicly known generative AI (GPT) are: 1. Deep learning model (1) Analyzes from the beginning to find meaning. (2) A transformer improves the accuracy of understanding the order of the sentence. 2. Vectorizes each word. 3. It is important to have an attention mechanism (processes information while focusing on important elements). It derives relevance to make it easier for the AI ​​to understand the meaning of the question. 4. The transformer is configured as shown in Figure 1(B), but it repeatedly improves its ability by understanding meaning rather than just asking. Information in the lower layers of a multi-layered program is insufficient. It only reaches the upper layers of the program to finally arrive at a good answer. It is not good at abstract things. It has the above functions. Publicly known generative AI (GPT) is characterized by using a large-scale language model with an attention mechanism and a table of contents for important parts, rather than stacking them in a series.

[0015] Specifically, the algorithm (processing method) for the GPT (Generative Pretrained Transformer), a known generative AI, is pre-trained on 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 strings of words and characters, each of which is assigned position information, and then appropriately compressed, aggregated, and stratified by an encoder. 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 relationship between each word and character. Using the information received from the encoder and self-attention, the decoder predicts the next word or character to be generated, stratifies it, and outputs it. The GPT then generates a sentence following the input text. During the pre-training phase, GPT learns linguistic knowledge and patterns from large amounts of text data on the internet (e.g., Wikipedia and news articles), masking (hiding) portions of text and predicting the masked portions from the remaining text to master general language rules and grammar. After pre-training, the model undergoes a transfer learning phase, fine-tuning it to a specific task (e.g., dialogue or summarization). It learns the knowledge and skills required for the task from task-related text data (e.g., dialogue history and summaries), enabling it to handle the specific task. Transfer learning involves feedback learning (so-called supervised learning), which prepares an output (answer) for an input (question), and further learning (so-called unsupervised learning) within GPT itself based on the pre-training. This improves accuracy in multiple fields, enabling generative AI to be applied in any field. For specialized topics, humans evaluate the answers generated by the GPT model and further fine-tune the model.

[0016] The present invention is a technology that aims to improve the accuracy of answers to questions from the generated AI after this kind of learning. In particular, it converts it into Japanese language processing and translation processing that is highly reliable and can communicate with Japanese, resulting in the highly accurate response generation of the present invention.

[0017] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described in detail below with reference to the accompanying drawings. The concept of the present invention is to increase the reliability of an AI by repeatedly learning a single question.

[0018] For example, a single convex lens can only magnify to a certain extent. However, stacking two convex lenses creates a microscope that can magnify even the smallest details. This invention is similar to stacking two convex lenses, and by stacking two or more massive pieces of AI data, it is possible to obtain higher accuracy than existing generation AI.

[0019] Furthermore, the conversation and research functions of known generative AI are performed in English using big data and sentences similar to human conversation. Even in Japanese, there is little accumulated data and accuracy issues. In other words, the natural language processing technology of generative AI has insufficient Japanese language learning, and the importance of Japanese words is insufficient. Consequently, there are issues with accuracy and learning.

[0020] The deep learning (machine learning) methods used in known generative AI are natural language processing models created and trained based on an English database. Therefore, even if they are accurate in assigning importance to English words, their training for Japanese is considered insufficient. For example, even with 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 that solves the above problem using a generative AI device.

[0021] The highly accurate Japanese response generation AI of the present invention can be used to develop various activities and products. Specific processing forms using this generation AI will be explained from the eighth embodiment onwards.

[0022] [First Embodiment of the Present Invention] Figure 2 shows a first embodiment of the present invention. This invention aims to further improve accuracy over conventional generative AI, for example, by re-asking AI2 the answer 3 obtained by [word or character input] 1. The AI's answer is then re-asked, and in addition to the AI's answer and the original question, an additional input is added to question whether the answer is correct. That is, re-asking words or characters such as "Is this really correct?", "Re-investigate," or "Give us another reliable answer." This changes the weighting of words, leading to a higher-level data hierarchy and a more accurate and precise answer. Unlike conventional AI, where a question setter simply asks a question and the AI ​​simply produces a low-level answer, the question setter does not prepare the answer and have the AI ​​learn the errors. By having the AI ​​itself examine the content in more detail from the database, it can match the perceptions of the researcher as closely as possible, resulting in a more reliable answer 31. This has been named the Reflection AI method.

[0023] [Second embodiment of the present invention] Figure 3 shows a second embodiment of the present invention, characterized in that the input language of questions to the AI ​​2 is a language other than English 11, such as Japanese 11. This is because the data of publicly known generative AI is mostly in English 12, and there is very little data in Japanese. The present inventor has invented an invention that obtains highly accurate answers using this Japanese input 11, and the inventor has filed a separate patent application for this invention. This second embodiment further improves the accuracy of publicly known generative AI using the inventor's invention.

[0024] Third Embodiment of the Present Invention FIG. 4 is a flow diagram illustrating a third embodiment of the present invention. By repeating the process of "asking the AI ​​for its answer again" (4) n times in the first embodiment, answer n becomes more substantial and more accurate than answer 1. When "asking the AI ​​for its answer n times," the words in the AI's answer may be broken down and emphasized before being queried again. The questioner may determine which words are important, or may use text mining techniques to derive them, emphasize them, and queried again. The aforementioned words to be queried again may also be added. Furthermore, in known natural language processing, the present invention also encompasses the redirection of embedding vectors (weights) of words with small vectors (weights) and the investigation of language. According to the third embodiment of the present invention, even difficult answers can be recognized more clearly.

[0025] [Fourth Embodiment of the Present Invention] Figure 5 is a flow chart showing a fourth embodiment of the present invention, which uses different AIs 2a and 2b (a and b) to obtain more accurate answers. The AI ​​2a of a and the generation AI 2 (not shown) repeat the question and answer 5. At the same time, the AI ​​2a of a also asks the AI ​​2b of b the same question and the answer document 5 from the generation AI 2 n times. If the answer 5 obtained using the AI ​​2b of b differs from the answer 5 obtained by the generation AI 2 using the AI ​​2a of a, the AI ​​2a and AI 2b continue to ask each other. This invention then allows the generation AI 2 to obtain the most accurate answer using the input information.

[0026] [Fifth Embodiment of the Present Invention] Figure 6 is a diagram showing 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 to the generation AI in English 63, and the AI's response 65 is passed through an automatic translation 622 from English to Japanese and returned as a response 666 in Japanese. The present invention does not involve a trainer (teacher) simply preparing words and having a translation model learn the errors. The generation AI responds based on a huge amount of data in English, so the present invention allows for highly reliable questions and answers to be obtained.

[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 is a diagram showing a sixth embodiment of the present invention, in which input 63 is made to AI in English via a Japanese-to-English translation function 671 with improved translation accuracy, and output 65 from the AI ​​is output 66 in more accurate and better Japanese by a translation function 672 with improved English-to-Japanese translation accuracy.

[0029] To further improve accuracy compared to known translation functions, the translation function is queried and the resulting English or Japanese answer is queried again. Queriing the translation means repeatedly translating the translated words back and forth. Repeated retranslations, with weighting added, clarify the meaning and purpose of the words, improving accuracy. This results in different weightings for words, leading to more accurate and precise translations.

[0030] According to the sixth embodiment of the present invention, the accuracy of answers from known AI devices can be significantly improved by inputting Japanese. The accuracy of any known AI generation device can be improved. Furthermore, input becomes easier.

[0031] 8 is a diagram showing a seventh embodiment of the present invention, which is a processing method in which high-accuracy answer systems 6101, 6102 are added to the speech recognition systems 681, 682 with improved accuracy as described above. A Japanese utterance 611 is input, and translation 681 with improved speech recognition accuracy and character conversion 691 are performed. The output text, that is, question 63, is then subjected to reverse speech recognition again into a Japanese utterance, and based on this second Japanese utterance, the questioner speaks in Japanese, further emphasizing the important points. By inputting 6101 three different English sentences, for example, into the generation AI, 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 a known translation model. Furthermore, languages ​​other than Japanese, such as Chinese, Korean, Spanish, and German, are also included in the present invention.

[0033] According to the seventh embodiment of the present invention, anyone can ask questions in Japanese and obtain highly accurate answers very 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, as shown in FIG. 9 , is an invention for a "confession room" or toilet room 761 in a church, specifically a "consultation center." As shown in FIG. 9 , a box-shaped room 6, such as a toilet 761, with a sealed door 762 enclosed on all four sides, or a curtained space is prepared. A counselor 763 sits in the toilet 761 and talks about their concerns. A microphone 773 at the back collects the voice and inputs it into the device main body 771 of the present invention, which is installed under the toilet water tank. The answer to the problem is then audibly replied to the counselor 763 through a speaker 772 at the back. This invention is appreciated because it provides answers to questions about unknown matters, such as future hopes and other unknown anxieties, rather than past atonement. The device of the present invention may be installed in an unenclosed location. This allows not only past data but also future information to be stored in a database, further improving accuracy. The device 77 of the present invention (main unit 771, microphone 773, and speaker 772) is a device having a server main unit with a highly accurate Japanese response generation AI trained on life consultation cases such as those described in embodiments 1 to 7. Preferably, the main unit, microphone, and speaker are integrated into a single unit so that it can be installed in a private room, etc. A QR code (registered trademark) can be placed in the enclosed room, scanned by a smartphone, and connected to the generation AI installed in the main unit of the device of the present invention installed in another location, allowing wireless consultation and spoken responses. More specifically, the device 7 of the present invention also includes an internet-connected device (such as a personal computer or smartphone) with an application that reflects answer questions, a display device such as a display, and power devices such as a battery and a power cord, all integrated into a single unit. When used for business purposes, a coin box 78 is provided as an example of the device of the present invention, and the device main unit 771 of the present invention is activated when the user deposits money into it, as shown in FIG. 9.

[0036] Ninth Embodiment of the Present Invention In the ninth embodiment of the present invention, as shown in FIG. 10 , a device main body 771, a microphone 773, and a speaker 772 of the present invention are installed on a grave 79 or inside a stone. This allows visitors to talk to the deceased, receive feedback, talk about their current worries, listen to their voices, 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 integrates the device main body 771, microphone 773, and speaker 772. Alternatively, if the device is installed as an all-weather device that can withstand wind and rain, a solar-powered system may be installed. Alternatively, a QR code (registered trademark) may be installed on the grave, scanned with a smartphone, and connected to a generating AI installed in the device main body of the present invention installed elsewhere, allowing users to talk wirelessly and seek 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 Figure 11, this is a completely new invention of Buddhist altar 710, consisting of a main body 771 of the device of the present invention, a microphone 773, a speaker 772, a memorial tablet, and a portrait of the deceased. This is a groundbreaking invention of a Buddhist altar that allows one to communicate with the deceased, just like the grave described above. As with the present invention, this invention brings about a religious revolution through the "room," "grave," and "altar." A QR code (registered trademark) can be provided on the Buddhist altar of the present invention, which can be read by a smartphone. This code can be used to connect to a generating AI installed in the main body of the device of the present invention installed in a different location, allowing for wireless communication and consultation about worries.

[0038] [Eleventh Embodiment of the Present Invention] The eleventh embodiment of the present invention is an invention as shown in Figure 12, in which a business model of a generation AI that embodies an embodiment of the present invention is an invention in which, upon payment of a fee set lower than the subscription fee, a consultation is received from a client via voice, and the information is converted into text and proofread, and an answer is obtained by the AI ​​of the present invention that embodies an embodiment of the present invention, which then converts the answer into voice and speaks it to the client. The eighth embodiment of the present invention has the advantage that the client only needs to generate a consultation and can receive a voice response from an expert on the spot.

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

[0040] [Twelfth embodiment of the present invention] Figure 13 is an overall diagram showing the twelfth embodiment of the present invention, illustrating a specific method of using the AI ​​generated by the present invention 2, 64 in an election campaign 80. Candidates do not generally move around. Voters use smartphones. These activities are characterized by being carried out remotely. Furthermore, posters do not need to be posted on bulletin boards. Voters do not need to ride in campaign cars. Candidates do not need to personally give street speeches. In other words, election campaigns involve question-and-answer communications with each voter's smartphone, with automatically generated written or audio responses being provided via the Internet by the AI ​​generated by the present invention. This makes candidate election campaigns remote working.

[0041] The overall diagram of the method using the AI ​​generated by the present invention is shown in Figure 13. The method is characterized in that the AI ​​generated by the present invention system is connected to the AI ​​generated by the present invention 2,64 system by holding a voter's smartphone with a dedicated app installed over the display of a link connection function such as a QR code (registered trademark) provided on election-related materials 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. The AI ​​generated by the present invention 2,64 system, trained with candidate information, responds in text or voice to questions posed 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.

[0042] While traditionally, candidates only post their policies on their websites, candidates can create dedicated smartphone apps capable of facial recognition, image recognition, character recognition, and 2D code recognition. These apps are then distributed to voters and encouraged to be installed. The server is trained with candidate information to create the AI2, 64 of the present invention. Using embodiments 1 through 7, a highly accurate AI2, 64 capable of answering questions via voice (Japanese) is created. Regarding bulletin boards, voters can essentially download and answer questions using the AI2, 64 of the present invention by simply launching a dedicated app on their smartphone and holding it over the bulletin board, without having to post posters in the "poster display area 81." This eliminates the need for candidates to create large numbers of posters or post them in several thousand locations across Tokyo, thereby reducing labor costs. Regarding election bulletins 83 and newspaper election advertisements 82, the dedicated app can be used to convert the candidate's name, face photo, 2D code, and text into the AI2, 64 of the present invention, a website, a database, images, videos, and other data.

[0043] Previously, candidates' appearances and voices were only seen and heard at campaign cars, street speeches 85, campaign broadcasts 84, personal lectures, and debates 86 at the venue, but now these events are primarily remote, with voters accessing the AI ​​2.64 of the present invention via a link on their smartphones, such as a two-dimensional code, the candidate's face, or the written promises, to answer questions. Images of the candidate may also be displayed, or a clone android (registered trademark) or avatar resembling the candidate may be used to conduct audio Q&A sessions using the AI ​​2.64 of the present invention. Furthermore, audio communication may be conducted as a live two-way broadcast between the candidate and the voters at the venue.

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

[0045] By using the above-described method and device to carry out election campaign 80 and installing electronic devices such as a microphone, camera, database storage device, and communication server PC in the candidate's home, even elderly people can run for office, make their case, and run for election. The database storage device and / or communication server PC may also be installed in the cloud.

[0046] [Thirteenth Embodiment of the Present Invention] Figures 14, 15, and 16 illustrate a thirteenth embodiment of the present invention, which is a more specific version of the twelfth embodiment. Figure 14 shows an election poster bulletin board 81, Figure 15 shows an example of an election poster with a two-dimensional code 87, and Figure 16 shows the processing flow of the bulletin board. First, a voter launches a dedicated app installed on their smartphone. A database 89 displays the posted number (code) and information on whether an election poster has been posted. The smartphone's camera then activates and holds the smartphone over the election poster display area 80. Images of the square display frame 813, the code (e.g., a small number) 812 written in the frame, the code (e.g., a large number) 811 written in the center of the candidate's frame, and the text title 814, such as "Poster Display Area" or "XX Governor Election," are captured and recognized as shapes, numbers, and letters. Furthermore, if a candidate's election poster is posted, the two-dimensional code 87 on the poster is recognized. The server then compares the results with the database 89 to determine where the voter is looking (e.g., election poster board, street speech, newspaper election advertisement, etc.) and automatically downloads pre-specified data to the smartphone based on the location. For example, the dedicated app's processing at a poster board location may involve downloading and displaying a candidate's poster or other document file, followed by an audio question-and-answer session with the AI ​​generated by the present invention 88 trained on the candidate's information. The processing procedure of such a dedicated app may be varied as appropriate depending on the conditions, or may be predetermined by the voter. This allows for highly accurate candidate information to be obtained from the AI ​​generated by the present invention. As described above, voters can easily obtain a wealth of information about candidates or political organizations, even if the candidate does not put up posters.

[0047] The app downloaded to the smartphone may be a dedicated app for a specific candidate or political organization, or it may be an app issued by, for example, an election management committee that provides links to all candidates and political organizations. When posters are posted, the link can be read using a well-known two-dimensional code reader on the smartphone. Furthermore, questions and answers can be exchanged via voice with the AI ​​88 generated by the present invention directly from the candidate's homepage.

[0048] [14th Embodiment of the Present Invention] Figure 17 is a processing flow diagram showing a 14th embodiment of the present invention, which further clarifies the 12th embodiment of the present invention. It illustrates the processing flow of a dedicated app for campaign cars and street speeches 85. After launching the dedicated app, voters hold their smartphone camera over a candidate's face (the candidate's face photo, portrait, name, or two-dimensional code) to compare it with a database 89 using facial authentication and character recognition. The server then determines where the voter is and what they are looking at, and sends them materials. For example, if the voter is giving a street speech, the server automatically and instantly downloads election bulletins and policy videos. The server then connects to the AI ​​generated by the present invention 88, which has been trained (data accumulated) with the candidate's information, and engages in audio Q&A. After linking with the dedicated app, voters can also communicate with a virtual avatar of the AI ​​generated by the present invention 88, which has been trained with candidate information similar to the candidate's. Live broadcasts with remote candidates are also possible.

[0049] By using the AI ​​88 generated by the present invention trained with 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 history, honors received, invention information, conversations and posts on social media, interview information, family information, claims and careers of political organizations, etc. Each of these pieces of information is then input into the AI ​​generation 88 of the present invention, and the methods of embodiments 1 to 7 are also incorporated to train it so that it can respond appropriately.

[0051] [Fifteenth Embodiment of the Present Invention] Figures 18 and 19 show a fifteenth embodiment of the present invention. Figure 18 shows an example of an election bulletin 83 with a two-dimensional code 87, and Figure 19 shows a processing flow diagram from a newspaper election advertisement 82 and an election bulletin 83. When a dedicated app installed on a smartphone is launched and held over the face, name, or two-dimensional code 87 on the candidate list in the election bulletin, the app recognizes the image or text. The server determines that the voter is viewing the election bulletin or other document by referring to a candidate information database 89. The server instantly and automatically begins transmitting the candidate's poster materials, political policy materials (election bulletin, newspaper advertisements), and video materials to the smartphone. The smartphone then connects to a generated AI 88 of the present invention trained on the candidate information and engages in a question-and-answer session with the voter. This process may also involve having the voter's smartphone recognize the face image of a specific candidate in a televised political campaign broadcast, connecting to a generated AI 88 of the present invention trained on the candidate information, and engaging in a question-and-answer session with the voter.

[0052] [16th Embodiment of the Present Invention] Figure 20 is a process flow diagram showing the 16th embodiment of the present invention, illustrating the process of hosting a candidate lecture or debate 86 online. Voters launch a dedicated app installed on their smartphone, click on a link to the lecture or debate information, and participate at a time notified in advance by the candidate or political organization. Voters can attend the lecture or debate in person at the venue, or they can participate at home using a dedicated app on their smartphone or computer. For example, if a voter is at a lecture venue, they can use the dedicated app to point their smartphone camera at the candidate's face or two-dimensional code. The lecture content is then instantly transmitted, and the voter is then connected to the AI ​​88 of the present invention trained with the candidate's information, allowing for a question-and-answer session. Furthermore, after linking with the dedicated app, they can communicate with a virtual avatar of the AI ​​88 of the present invention trained with candidate information that resembles the candidate. A live broadcast with a remote candidate is also possible.

[0053] For example, candidates can use a microphone and camera at home to provide video and audio to a video and audio display device set up at the lecture venue or debate, or to voters' smartphones or computers. Live communication can also be conducted using two-way video and audio. Candidates can record audio of their lectures in advance and distribute it to voters at the lecture or debate venue in person and on their smartphones, with voters only participating online and responding during question and answer sessions. Since they only need to respond, the burden on candidates is reduced.

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

[0055] [17th Embodiment of the Present Invention] Figure 21 shows the 17th embodiment of the present invention. Instead of a dedicated app, a dedicated website is prepared on a server. Voters can access the dedicated website on the server using a known reading link function, such as a two-dimensional code 87 on a candidate's flyer or poster, or by searching in their browser. They can then engage in a question-and-answer session with a trained AI 88 (e.g., ChatGPT) generated by the present invention, which incorporates the candidate's knowledge, political philosophy, policies, basic education, or political organization information (i.e., candidate data information). This trained AI's knowledge can respond to questions from people around the world and can represent the candidate's claims, particularly regarding politics. The inventor has named this system "Answer Mars."

[0056] [18th 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, and converts voice information from question and answer sessions held with a trained AI generated by the present invention 88 that has taken in candidate data, or information on desired issues stated by voters, into text and stores it in a database 91.The data is then organized and consolidated from the database 91 using another AI generated by the present invention 90 that has been trained in politics and policies, different from the trained AI that has taken in candidate data, to generate countermeasures and provide feedback to the candidate information database 89 and the trained AI generated by the present invention 88.

[0057] The details of the feedback system shown in Figure 18 are explained below. During the question-and-answer session of the trained AI 88 of the present invention, which incorporates voter and candidate data, the AI ​​determines whether the voter's voice is a question or concern for the candidate, or a request, complaint, issue, or problem for the entire group. In other words, it classifies the matter into one that requires an answer or one that does not. It then answers the question, and responds by listening to the details of requests, complaints, issues, and problems to elicit specific content from the voter. All of the conversation (text or audio) is converted into text and stored in a database 91. Another AI 90 of the present invention, trained on social information such as political policies, then organizes all information into questions, concerns, or complaints and requests, and further filters them by policy category. This other AI generates policy proposals. After the candidate confirms the proposed proposals, the AI ​​88 of the present invention revises the data 89 related to the candidate's policies and principles, and the AI ​​88 of the present invention, trained on the candidate information, further updates the AI ​​using the methods described in embodiments 1 to 7 of the present invention.

[0058] The above-mentioned circulating information can minimize the discrepancy between the candidate policies generated by the AI ​​of this invention and the requests and issues of voters, further improving the candidate policies and making it clear what election candidates should do politically, which has a very positive effect on the country.

[0059] [Nineteenth Embodiment of the Present Invention] In the nineteenth embodiment of the present invention, a two-dimensional code is provided on the cover or obi of a book, a magazine, a book, a newspaper, sheet music, an instruction manual, a comic book, or on stationery such as a memo pad or notebook, or on a photograph, and when a reader holds a smartphone over the code, the reader is connected to a trained generative AI of the present invention that has accumulated knowledge information, philosophy, policies, and content description data of the author, supervisor, creator, or photographer of the printed material, and can ask questions by voice or text and receive answers from the generative AI of the present invention. The nineteenth embodiment of the present invention makes reading printed materials such as books even more enjoyable.

[0060] In the twentieth embodiment of the present invention, a two-dimensional code is provided on a product or its instruction manual, and when a reader holds a smartphone over the code, they can engage in a question-and-answer session using connected text or voice with the AI ​​generated by the present invention, which has been trained on detailed information, structure, and explanations of the product. According to the twentieth embodiment of the present invention, the product can be easily handled and understood.

[0061] The invention shown in the 20th embodiment of the present invention is not limited to products or documents that have two-dimensional codes, but includes everything that involves connecting information that needs to be explained to a third party with a smartphone, such as structure and handling, to a trained AI generated by the present invention via a smartphone and conducting a question and answer session.

[0062] The AI ​​of the present invention may be trained by adding, for example, information on deceased individuals or past facts 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 long-standing political parties to be input, enabling highly accurate responses from political parties.

[0063] All AIs produced by modifying the present invention are included in the present invention.

[0064] The present invention provides a foundation for further industrial development and has great potential for industrial application.

[0065] 1 Question 2 AI 2n n types of AI 3 Answer 31 High-precision answer 32 High-precision answer by inputting English 3n High-precision answer by repeating questions 4 Answer n-1 and question n generation (n = 2, 3, 4, ...) 5 Question and answer 61 Question in Japanese 611 Question in Japanese (audio) 62 Automatic translation 621 Automatic translation function from Japanese to English 622 Automatic translation function from English to Japanese 63 Question in English 64 Generating AI 65 Answer in English 66 Answer in Japanese 661 Answer in Japanese (audio) 67 High-precision translation 671 Automatic translation function from Japanese to English 672 Automatic translation function from English to Japanese 68 High-precision speech translation function 681 High-precision speech translation (from Japanese to English) 682 High-precision speech translation (from Japanese to English) 691 Speech-to-text conversion 692 Text-to-speech conversion 6100 High-precision answer function 6101 High-precision answer (question) 6102 High-precision answer (answer) 76 "Advice room for problems" of the present invention, such as a church "confession room" or toilet room 61 761 Toilet 762 Door 763 Adviser 77 Device of the present invention 771 Main body 772 Speaker 773 Microphone 78 Fee box 79 Grave of the present invention 710 Buddhist altar of the present invention 711 Question and answer in Japanese text or voice 712 Question and answer in English text 713 Automatic document translation and automatic voice translation 80 Election campaign 81 Bulletin board 811 Number for poster pasting (large) 812 Number for poster pasting (small) 813 Notice board frame 814 Text title such as "Election poster display area" 82 Newspaper election advertisement 83 Election bulletin 84 Political campaign broadcast 85 Campaign car and street speech 86 Lectures and debates 87 2D code 88 AI of the present invention trained (accumulated learning) with candidate information 89 Candidate information database 90 AI of the present invention trained with political, policy, and current affairs information 91 Voter information database categorized in detail into requests and questions

Claims

1. A reflective generation AI system that memorizes and stores the thoughts of respondents in advance, and then outputs the thoughts from the AI ​​to inquire about the voter or product purchaser, etc.

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

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

4. A reflective generation AI system according to claim 3, characterized in that the question is repeated n times.

5. A reflective generative 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 generative 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 as claimed 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. A method of using a generative AI system as claimed in any one of claims 1 to 5, comprising installing a device having the generative AI system storing information on the deceased and ancestral information, a microphone and a speaker on a grave or in a stone or on a Buddhist altar, and enabling communication with the deceased and ancestors.

8. A method of using a generative AI system as claimed 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 inquiries in text or voice.

9. A method of using a generative AI system as claimed in any one of claims 1 to 5, wherein the election-related object 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 as claimed in any one of claims 1 to 5, characterized in that a connection function such as a QR code (registered trademark) is provided on printed matter such as the cover or dust jacket of a book, magazine, newspaper, sheet music, instruction manual, etc., and the generative AI system responds to an inquirer in text or voice.

11. A method of using a generative AI system as claimed 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 a reply is required, and replies to those that require a reply, a second generative AI system classifies the items 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 replies of the first generative AI system.

Images