Date of last update:
AI technology has developed in recent years, and AI has come to be used in various places, but “AI with emotions” has not yet been achieved.
AI with emotions refers to “general purpose AI”, which can judge the situation on its own and perform various roles. All of the AI used in “An example of expert AI is the artificial intelligence “AlphaGo” developed by Google, which is also known for winning against a professional Go player. Although expert AI is said to be more than abilities human, the development of general purpose AI is said to be currently not feasible.
▶ What is artificial intelligence (AI)? A thorough explanation of definitions, history, trends and future prospects >>“class=”glossaryLink”> Specialized AI”, which specializes in a single field or field.
Among them, this time, we have many achievements in machine conversation, text mining, language analysis, and word-of-mouth analysis, and we are working hard to realize “AI with emotions” with a logical approach. We interview Mr. Koji Nishimoto.
Mr. Nishimoto, CEO of Bazgraph
About Bazgraph Inc.
It is now required to read more information and verify the necessary text data in a limited period of time.
Buzzgraph has developed “Tanteki AI Sentence Summary” which can be used for analyzing and selecting large amounts of text data by making full use of its own natural language processing engine.
Tanteki uses a unique system for analyzing sentences, and instead of relying on a large number of dictionaries to summarize, it is an eco-friendly AI that can create its own dictionary and learn on its own.
“Natural Language Processing
Natural language processing is a technology that allows computers to understand the words that humans use every day.
For example, Nico Nico Douga uses this technology to analyze comments, enabling predictive conversion by analyzing comments unique to Nico Nico Douga one character at a time.
▶ What is “Natural Language Processing (NLP)” you can’t ask now? >>
▶ Nico Nico Douga, machine learning to reduce monitored comments by 75%! Dwango Data Use Case Seminar Report on “Supernatural Language Processing”>>” class=”glossaryLink”> Even in difficult areas like natural language processing and AI, it is easy to understand and familiar to everyone from children to adults, and everyone loves and needs it. AI is n being able to think like a human being.” is our aim.
Summary and About Tanteki Series
ーー Could you explain the Tantech series please?
Tanteki series is a service created by making full use of our own natural language processing AI “HAKASE”, focusing on news articles. The data we use now is basically news articles. Two years is the number of news articles used so far, and the number is around 4 million articles. Among them, 1.4 million articles are relearned, and points are given to all the words that make up the article every three hours. The feature word list shows the degree of importance of words, and summarizes based on this.
Click here for the article on Tanteki
ーー How is the business model going?
Right now, we provide an API, which we’d like to fund, but it’s very difficult to wrap it up in terms of a business model.
In the first place, what we basically use for summarization is not the BERT model, but our own model, which applies unsupervised learning and does not use any sentences summarized by humans. So, if you ask me if I can sum it up in the same way as a human being, I really can’t. In terms of what we focus on, we focus on developing a system of thinking, judging which areas have impact and are important to human beings.
For example, there are businesses that need a lot of abstracts. Humans do the abstracting, but if you ask if it can be replaced by machines, there will still be problems somewhere, so it’s hard to do business with.
What I’m aiming for at the current level is when mining call center logs, etc., where other companies’ mining can’t catch up, for example, by summarizing and compressing parts that humans have to judge , important parts Since it is drawn, it becomes very easy to understand. In such a place, we are now thinking in the direction of monetization.
Is it possible to express emotions in words alone?
ーー Recently, a Google researcher claimed that “the large-scale language model LaMDA has emotions” and became a hot topic. What do you think about that?
Basically, I don’t think so. As for what is necessary to have emotions, the basic reason is rational judgment. Humans see the ability to judge the situation and understand its extent as an emotion.
For example, even if you create a neural structure and build something like a BSRT model, in order to create emotions there, how risky is the situation you’re in now? Otherwise, there is basically no choice but to quantify how close you are to achieving your goals with a computer. However, I believe that it can only be done to a certain extent through writing.
ーー What exactly does it mean to be able to do it with just text?
Both need an understanding and a logical degree of the situation. Since sentences are there to convey situations to people, there is basically logic in sentences. As long as a sentence expresses a situation, it must contain logical judgments, and a sentence contains a combination of logical judgments.
Then what is logic is the order in which things happen, that is, the chronological order. Since sentences contain chronological sequences within them, logical constructions can be drawn from sentences.
Now, the question is whether there is anything that can be done with natural language processing.
In shortBy disassembling and consolidating the three sentences, “Ruptured internal organs were serious injuries”, “He was seriously injured due to ruptured internal organs” and “He died of serious injuries”, the logic of “Ruptured”. internal organs → serious injuries → death””
This is why we deal with news articles AI researchers and AI companies often analyze Wikipedia. However, the crucial difference between Wikipedia and news articles is that news articles can simulate what humans experience. News articles have “5W1H”, which can be said to be experience points. And big news is written by many reporters, each judging the important points for human beings. Logic is formed by analyzing and extracting features of news articles, such as experience value and degree of importance to humans, in large quantities.
Moreover, if you step into emotions, if you automatically create a graph structure with a specific risk and a specific purpose at both ends, for example, when you get an infectious disease,We can calculate how far an infectious disease is from the risk of death by the number of articles about it and the thickness of the article content.
And by inserting various things on both ends of this graphic structure, we can see how much of an effect a certain phenomenon has on people, and I think that will be the basis of emotions.
ーー How do you see the approach of loading large amounts of data into deep learning, Mr. Nishimoto?
I think that is one approach, but I don’t think it will solve everything. I feel that no matter how much deep learning is followed, human beings cannot be created.
Of course, we use deep learning, but if we apply deep learning to the finished product based on logic, we can create a clean, fast filter with even less waste.
Originally, my idea was that plants grow their branches arbitrarily in the direction of light and wear leaves to grow efficiently, and that slime molds gradually aim for the shortest distance. This is where I felt I could not do it.
So, to put it in a nutshell, what BuzzGraph is trying to do is to first build something that adds self-growing artificial intelligence and a database, and then create such a system.
ーー Finally, could you tell us about your plans for the future?
First of all, what we are ultimately aiming for is a society where computers can solve problems together by making full use of natural language and thought for humans.
Most recently, I’ve been focusing on developing mining tools, and I’m trying to create mining tools that not only allow us to understand the past, but also allow us to see the future based on past data.
last
This time, we interviewed Mr. Koji Nishimoto of Bazgraph Co., Ltd. Mr. Nishimoto’s thoughts on the question “How can we make AI feel emotions?” It may not be mainstream in the industry, but it was persuasive and I felt an unwavering conviction.
Future Buzz Graph trends are covered, such as developing a summary system that can think and make human-like judgments through text analysis, and developing mining tools that can understand the future such as problem solving and prediction.
I am studying advertising and public relations at university.
I like watching and playing football.
I would like to deepen my knowledge about AI and submit easy to understand articles!
