Whether AI will become useful or not?   Eshita Masayuki (Professor, Meiji University)|電経新聞

Whether AI will become useful or not?   Eshita Masayuki (Professor, Meiji University)

We should just end the debate about whether AI will be useful or not. This is because it just depends on how it is used.
The text creation and image processing techniques for the generated AI are at a sufficiently practical level depending on the purpose. At the same time, the generated AI is not a content vending machine. If you make a request, you won’t be able to deliver something that is satisfying.
To show how unusable it is, many people still say, “Enter your name and let the AI ​​investigate,” and the number of teachers who are experiencing the silly reckoning of reports created by leaving it all to AI has increased significantly more than last year.
The former case simply means that there is a lack of training data, and the latter example is to generate statements with high probability of matching for questions that have no logic construction, since machine learning-based generation AI is not a logical inference device.
It was only natural to try and show that he was useless with a smug look on his face.
Incidentally, when I ask ChatGPT4, “Isn’t that good at inductive reasoning, but not deductive reasoning?”, he replies, “It can be said that it is generally a reasonable understanding,” and has presented the following as “Why is the generation AI considered to be weak to deductive reasoning?” that linguistic models are essentially stochastic prediction devices and their ability to guarantee the correctness of procedures is limited. I’m sure I’m the worst at mathematics proof problems. Proofs for textbook theorems such as the Three-Square Theorem often appear in the past, and the proof process is often explicitly described in natural language, meaning that it can be imported as training data, so it is often possible to prove it. However, ChatGPT4’s response was, “generated AI is resistant to reconstructing known patterns, and there are limits to how to construct deductive solutions to unknown problems.” The ability to logically explain one’s own weaknesses can be said to be a major advance in recent years in the generation AI.
About 40 years ago, computers gradually began to become popular in the world. In a manga from the time that depicted the near future, a boy who freely controls a computer and tries to “converse” with the computer and tries to uncover the answers to various problems.
Real computers only had the performance to do numerical calculations using BASIC programs and draw rough dot graphics, so the “conversation” scenes with the computer were futuristic enough. There was no computer to talk to.
Needless to say, conversations with computers have already reached a sufficiently practical level today through the use of AI generated by the AI.
In my case, I was able to save a lot of effort by summarizing sentences, proofreading, and interpreting long sentences in foreign languages ​​thanks to the generation AI. Sometimes they ask you to suggest ideas. However, in this case, it is necessary to provide sufficient information in advance, to show the target goal, and to repeat the questions. It can save labor, but it can’t reduce labor.

※Translating Japanese articles into English with AI