10
$\begingroup$

It was recently brought to my attention that Chess experts took the outcome of this now famous match as something of an upset.

See: Chess’s New Best Player Is A Fearless, Swashbuckling Algorithm

As as a non-expert on Chess and Chess AI, my assumption was that, based on the performance of AlphaGo, and the validation of that type of method in relation to combinatorial games, was that the older AI would have no chance.

  • Why was AlphaZero's victory surprising?
$\endgroup$
14
$\begingroup$

Good question.

First and foremost is that in Go deepmind had no superhuman opponents to challenge. Go engines were not anywhere near the highest level of the top human players. In chess, however, the engines are 500 ELO points stronger than the top human players. This is a massive difference. The amount of work that has gone into contemporary chess engines is staggering. We are talking about millions of hours in programming, hundreds of thousands of iterations. It is a massive body of knowledge and work. To overcome and surpass all of that in 4 hours is staggering.

Secondly it is not so much the result itself which is surprising to chess masters but instead its how AlphaZero plays chess. It's quite ironic that a system which had no human knowledge or expertise plays the most like we do. Engines are notorious for playing ugly looking moves, those lacking harmony etc. Its hard to explain to a non-chess player but there is such a thing as an "Artificial move" like the contemporary engines come up with often. AlphaZero does not play like this at all. It has a very human-like style where it dominates the opponent's pieces with deep strategic play and stunning position sacrifices. AlphaZero plays the way we aspire to, combining deep positional understanding with the precision of an engines calculation.

Edit Oh and I forgot to mention something about the result itself. If you are not familiar with computer chess it may not seem staggering but it is.

These days the margins of victory which separate the top contemporary engines are razor thin. In a 100 game match you could expect to see a result like 85 games drawn, 9 victories, and 6 losses to determine the better engine.

AlphaZero 28 wins and 72 draws with zero losses was otherworldly crushing and was completely unthinkable right up to the moment it happened.

$\endgroup$
  • $\begingroup$ Nice answer. Your point about comparison between AI's re Chess is interesting in terms of the limitation based on Chess' loopiness and the Win/Loss/Draw triad. (Possibly, in the future, we will need finite, intractable games that allow more granular analysis in terms of results.) I am familiar with the history of Chess engines, and the massive amount of effort and human knowledge that went into them, but the context of the lack of success re: the much more complex 19x19 Go had an opposite ramification to me. $\endgroup$ – DukeZhou Feb 7 '18 at 22:14
  • $\begingroup$ Specifically, my assumption was that if AlphaGo could beat the top humans in the significantly more complex game, it seemed reasonable that it would beat not only the top humans, but the top previous AI's in any other game. $\endgroup$ – DukeZhou Feb 7 '18 at 22:17
  • $\begingroup$ The bit about artifical moves is pretty important, and not something I've seen non-chess players talk about. +1 $\endgroup$ – Stella Biderman Feb 14 '18 at 7:25
4
$\begingroup$

MCTS for chess had been tried in the literature with little success. It was assumed AlphaGo's approach would never work on chess, maybe in Go but not in chess. Suddenly, Google announced the approach was working and it was beating the World's strongest chess program by a very signficiant margin.

Before Google, all chess programmers were taught crafting heuristics in engine programming was a better strategy than machine learning. No matter how you implemented neural networks, it would have never ran faster than a bunch of 64-bit bitboards instructions. AlphaGo was running quite slow, but it played strongest chess.

$\endgroup$
2
$\begingroup$

I see, based on the articles you provide, many levels of surprise in the victory:

Chess is hard game to master and the counter part had the world's best practices, AlphaZero had tabula rasa.

Learning took four hours and AlphaZero lost no match of 100.

Playing style was an alien mix of human and computer like moves, aggressive and some times seeming goofy with sacrifices that have no idea but are actually making future status more strong.

Amount of possiblities taken in account per move was less than counter part, AlphaZero had a mysterious gut feeling or intuition.

The upset feeling came from the amount of training material AlphaZero had built itself with and the time limit, that did not maybe give the traditional machine fair amount of time.

$\endgroup$
  • $\begingroup$ Aaah. So it stemmed from lack of confidence in the new AI method. That makes sense. $\endgroup$ – DukeZhou Feb 7 '18 at 22:17

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.