2016년 3월 20일 일요일

The meaning of AlphaGo, the AI program that beat a Go champ - interview of Geoffrey Hinton


AI의 아버지 정도 되는 전문가의 인터뷰.

http://www.macleans.ca/society/science/the-meaning-of-alphago-the-ai-program-that-beat-a-go-champ/


Q: So, why is it important that AI triumphed in the game of Go?

A: It relies on a lot of intuition....The neural networks provides you with good intuitions, and that’s what the other programs were lacking, and that’s what people didn’t really understand computers could do.


바둑은 직관에 의존하고, 알파고는 훌륭한 직관을 제공하고, 그것은 다른 프로그램이 갖지 못한 것이고, 컴퓨터가 할 것이라고 사람들이 이해하지 못했던 것이었다.


Q: In 2014, experts said that Go might be something AI could one day win at, but the common thinking was that it would take at least a decade. Obviously, they undershot that estimate. Would you have guessed then that this was possible?


2014년(구글이 딥마인드를 인수하던 시절)에도 10년은 걸릴 것이라고 하던 일을 2년 만에 해낸 것이라고...



Q: How important is the power of computing to continued work in the deep learning field?

In deep learning, the algorithms we use now are versions of the algorithms we were developing in the 1980s, the 1990s. People were very optimistic about them, but it turns out they didn’t work too well. Now we know the reason is they didn’t work too well is that we didn’t have powerful enough computers, we didn’t have enough data sets to train them. If we want to approach the level of the human brain, we need much more computation, we need better hardware. We are much closer than we were 20 years ago, but we’re still a long way away. We’ll see something with proper common-sense reasoning.

지금 사용하는 딥 러닝의 알고리즘은 80년대, 90년대에 개발된 것이지만, 당시에는 잘 작동하지 않았다. 지금은 그 이유가 충분히 강력한 컴퓨터와 학습에 필요한 충분한 데이타가 없어서 그런 것이었다는 것을 안다.



Q: In the ’80s, scientists in the AI field dismissed deep learning and neural networks. What changed?

A: Mainly the fact that it worked. At the time, it didn’t solve big practical AI problems, it didn’t replace the existing technology. But in 2009, in Toronto, we developed a neural network for speech recognition that was slightly better than the existing technology, and that was important, because the existing technology had 30 years of a lot of people making it work very well, and a couple grad students in my lab developed something better in a few months. It became obvious to the smart people at that point that this technology was going to wipe out the existing one.

80년대의 인공지능과학자들은 딥러닝과 신경망을 무시했는데, 무엇이 바뀐 것인가?
무엇보다 그것이 작동을 한다는 것이다. 당시에 그것은 중요한 ai문제들을 풀지 못했고 기존의 기술을 대치하지 못했다. 그러나 2009년에 음성인식 신경망을 개발했고, 기존의 기술보다 조금 우수했는데 그것이 중요했다. 기존의 기술은 잘 작동하기 위해서 30년동안 많은 사람들이 필요했지만, 우리는 몇개월동안 두 명의 대학원생이 더 좋은 것을 개발했다.


  1) There’s life in old AI approaches: deep learning + tree search

  2) Polanyi’s paradox isn’t a problem: “We know more than we can tell.”
  http://www.nytimes.com/2016/03/16/opinion/where-computers-defeat-humans-and-where-they-cant.html?

  3) AlphaGo isn’t really AI:
  http://spectrum.ieee.org/automaton/robotics/artificial-intelligence/why-alphago-is-not-ai

  4) AlphaGo is pretty inefficient

  5) Commercialization isn’t obvious


http://www.goratings.org/


http://www.geekwire.com/2016/geek-trash-talk-facebook-ai-chief-dismisses-googles-alphago-victory-lee-sedol/

LeCun is apparently skeptical about whether AlphaGo is actually learning how to play, or is simply processing millions of potential Go moves that have been programmed into its memory.