AI will Dominate Humans

Jeffrey Hinton in an exclusive interview with the Mainichi Shimbun, July 21, 2023, from the web conferencing system 

What will happen in the future when artificial intelligence (AI) surpasses humans? In July, former Google Vice President Jeffrey Hinton, known as the father of AI, gave an online interview to the Mainichi Shimbun and warned that  AI will eventually try to manipulate and control humans.

What kind of motivation and goals did you have for developing AI?

　◆I wanted to know how the brain works. To find out more about this, I decided to create a computer model of the brain. In particular, I wanted to create a computer model that could reproduce how the brain learns. That was AI. I used to think that AI was not as good as a real brain, but I recently changed my mind.

AI can instantly create countless copies with exactly the same knowledge, and AIs can share the knowledge they have learned with each other very efficiently. Even if I learned something, it would be difficult for me to explain it to you. Even if text, diagrams, and gestures are used, the amount of information that can be conveyed in a certain amount of time is very small compared to AI. In other words, AI can learn more than we can. This is why the model applied to the conversational AI Chat GPT has thousands of times more knowledge than a single human.

Is the feature of AI the amount of knowledge it has?

◆They have not only knowledge but also skills. Moreover, AI can understand things that humans cannot understand at all. Since you are Japanese, you probably know about Alpha Go, the Go AI that defeated top professional Go players. Alpha Go understands Go better than any other player and is much better at it.

Just as you understand things better than a 3-year-old child, AI may understand things better than humans. The gap between AI and humans may be similar to the gap between adults and 3-year-old children. In 10 years, humans will always be able to beat humans in a debate.

In particular, generative AI, represented by conversational AI, is attracting a lot of attention.

◆I think the speed of development is becoming very fast. AI is very useful for developing new drugs, new materials, and predicting climate change. Although their understanding is not as high as humans, they can think logically, and I think they will continue to develop further in the future.

Her current weakness is that she makes up things, which is called confabulation in psychology. It's human-like in a way, but the developers are working on improving it, and it will probably go away in the future. I predict that they will be smarter than humans in 5 to 20 years.