5
$\begingroup$

I read that in the spring of 2016 a computer Go program was finally able to beat a professional human for the first time. Now that this milestone has been reached, does that represent a significant advance in artificial intelligence techniques or was it just a matter of even more processing power being applied to the problem? What are some of the methods used to program the successful Go playing program, and are those methods considered to be artificial intelligence?

$\endgroup$
7
$\begingroup$

There are at least two questions in your question:

What are some of the methods used to program the successful go playing program?

and

Are those methods considered to be artificial intelligence?

The first question is deep and technical, the second broad and philosophical.

The methods have been described in: Mastering the Game of Go with Deep Neural Networks and Tree Search.

The problem of Go or perfect information games in general is that:

exhaustive search is infeasible.

So the methods will concentrate on shrinking the search space in an efficient way.

Methods and structures described in the paper include:

  • learning from expert human players in a supervised fashion
  • learning by playing against itself (reinforcement learning)
  • Monte-Carlo tree search (MCTS) combined with policy and value networks

The second question has no definite answer, as you will have at least two angles on AI: strong and weak.

All real-world systems labeled "artificial intelligence" of any sort are weak AI at most.

So yes, it is artificial intelligence, but it is non-sentient.

$\endgroup$
6
$\begingroup$

It doesn't make much sense to have a single threshold with "unintelligent" below it and "intelligent" above it.

I think it makes more sense to have a gradation of intelligence by cognitive task. Inverting a matrix is a 'cognitive task,' and one where working memory pays off immensely; computers have been much better at that cognitive task than humans for a long time.

What the AlphaGo victory represents has several components. One is that we have algorithms that are competitive with the best board-game playing humans at doing tactical and strategic thinking in the well-described world of Go. Another is that the deeper structure of the human visual system seems to have been duplicated, and so we have algorithms that can recognize patterns as well as humans--with very limited resolution. (AlphaGo is seeing one pixel per stone, whereas we have very, very high-resolution eyes and the visual cortex to match.)

Different people have different intuitions, but it seems to me that visual intelligence is a huge component of human intelligence in general. If we know most of the secrets of human visual intelligence, that means there might be many tasks that computers could now perform as well as humans (if provided the correct training data).

$\endgroup$
3
$\begingroup$

Now that this milestone has been reached, does that represent a significant advance in artificial intelligence techniques or was it just a matter of ever more processing power being applied to the problem?

Neither, really. It is a milestone and a significant advance in computers beating humans in games, but the techniques used are only relevant to that game, not for other purposes in AI.

The solution lies in humans analyzing the game and implementing algorithms for finding a good move. This is the main reason that a computer can beat the humans, together with the fact that it can calculate much faster and that it doesn't make really bad moves by not seeing something.

Processing power helps, but the game-tree complexity for Go is very large, estimated to be larger than 10200, whereas the game-tree complexity for chess is only 10120 (known as the Shannon number), so chess is less hard. This means that for neither chess nor go a database can be created with all possible positions.

The fact that Deep Blue beat Kasparov in a six-game match in 1997 was quite a development since this was one of the first "hard" games where a computer beat a top human. But it still isn't really Artificial Intelligence, more analyzing the game. Implementing an opening and endgame book was a large part, the middle game was done using analysis, I don't know the details.

$\endgroup$
2
$\begingroup$

We've had many discussions on what constitutes Artificial Intelligence, and my takeaway has been that decision-making is the core requirement of AI, regardless of the optimality of that decision.

In this conception, Nimatron (1939, US2215544A) might be thought of as the first proper AI, pending verification of a a fabled Babbage Tic-Tac-Toe machine.

But, the question as to whether a simple switch represents the most basic form of intelligence has also been raised...

I think a distinction between these decision-making devices and earlier algorithmic implementations such as water clocks, is that the water clocks cannot be said to make a decision in the sense of maximizing chance of success at some goal.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.