『[English] Do Todays AI Systems Really Understand Anything?』のカバーアート

[English] Do Todays AI Systems Really Understand Anything?

[English] Do Todays AI Systems Really Understand Anything?

無料で聴く

ポッドキャストの詳細を見る
[Preview books] [Borrow books] [Pause] Anyone who has worked with ChatGPT, Gemini, or any other artificial intelligence system of that kind has probably felt that they do understand. An AI communicates with us much like an ordinary human being. They make jokes, and they respond to our sarcasm and minor objections in ways that seem remarkably human.If you ask one of them to prepare presentation slides for your lecture, it may do a better job than you would yourself. I have even heard that many students are now using AI to complete their school and college assignments.So then, isn't it obvious that they understand things? Certainly not. The reason lies in the very way today's AI systems have been designed. In reality, they possess no power of understanding whatsoever.All an AI does is this: based on what you have said, or on what it was previously trained on, it simply performs pattern matching and predicts what the most appropriate answer to your question is likely to be.That may not be quite as bad as it sounds. After all, many of us do exactly the same thing. Most people function very much like pattern-matching prediction machines. We rarely make the effort to understand things deeply.So then, what exactly is involved in genuine understanding?To put it in a highly simplified way, it means connecting a newly encountered word to something we already know. Or, in other words, discovering the meaning of a new word through something that is already familiar to us.But this connection need not be limited merely to words. It can go much further than that.For example, the moment someone says the word "cat," our mind associates that word with a soft furry creature having four legs, a long tail, and a tendency to purr. In reality, we connect a word with the entire description of the living being it represents.Nor is our understanding limited merely to visual experience.If you have ever traveled to certain Southeast Asian countries, the mere mention of the word "durian" may immediately bring many details to mind: that sharp smell that can almost make one nauseous, and yet the surprisingly pleasant taste that lingers in the mouth.In other words, understanding is not simply matching one word with another. It involves connecting that word to the perceptions of all our senses, to past experiences, and to knowledge that has already been accumulated. But remember, these connections are not permanent. They can change over time as new information becomes available. And they can later be recalled and used again.Are AI systems incapable of doing this?In the AI systems currently available, that is certainly the case. An AI is fundamentally a machine confined to language. Its world consists primarily of words, sentences, and a vast storehouse of knowledge.If an AI is taught that "a durian is a fruit with a strong smell," it merely associates the word "durian" with a description of that smell. But this is possible only during its training phase. Only its creators can teach it in that way. You and I cannot do so afterward.Are you surprised?Perhaps not. You already know very well that no matter how intelligent an AI may appear, it is ultimately just a computer program. Yet, have you ever wondered how an inanimate program can accomplish so much?Let us look a little deeper into the origins of these AI systems.Today's AI systems are called "Large Language Models." They operate entirely around human language. The beginnings of these programs were quite simple. Their original purpose was merely to translate from one language into another.Most of us learned new languages in school by studying grammar, vocabulary, and similar things. But none of us learned our mother tongue in that manner. Yet we can speak it fluently and with relatively few grammatical mistakes. How did that become possible?It was not through conscious study or deliberate understanding.Research has shown that a child begins learning its mother tongue while still in the mother's womb. Even before birth, the sounds of people speaking outside can be heard by the fetus.Although the brain is still incomplete at that stage, it already begins trying to identify the boundaries between words in the language being spoken around it. However, it has no ability to understand what it is hearing.How does it manage this?That is the miracle performed by the neurons in its brain!A neuron is like a tiny biological computer inside our brain. There are billions of such neurons within us. As the fetus grows into a child, some of these neurons are still forming. Others have already formed and are preparing themselves for specific tasks. Still others are in the process of establishing their roles within the brain. These neurons are the real actors behind this remarkable phenomenon.Some clever researchers observed this process. They attempted to imitate nature itself. This led them to the idea of an "artificial neural network," modeled loosely on the human ...
adbl_web_anon_alc_button_suppression_t1
まだレビューはありません