Neural-net powered AI "crushes" top chess engine


#1

Originally published at: https://boingboing.net/2018/12/07/neural-net-powered-ai-crushe.html


#2

The only winning move is not to play.

“I have never been beaten in Chess by the neural net AI. I’m just that good.”

“You’ve never played Chess.”

“Exactamundo”


#3

Computer chess has killed the game. It’s boring now.


#4

Impressive, but this is still a state-of-the-art expert system designed to do one thing.
An actual AI could invent a game.


#5

I am reminded of this passage from John Brunner’s 1975 novel The Shockwave Rider, a dialog between a high-tech interrogator and the captive protagonist. The novel (set indeterminately in the 2010 to 2020 range) has already introduced the fictional game of Fencing:

[Interrogator:] “You’re a good fencing player, I believe.”

“Care for a game to break the monotony?” A ghost of old defiance tinted the words.

“I’m a poor player myself; it would be a mismatch. But why did fencing appeal to you rather than, say, go, or even chess?”

“Chess has been automated,” was the prompt reply. “How long is it since a world champion has done without computer assistance?”

“I see. Yes, I understand nobody has yet written a competent fencing program. Did you try it? You had adequate capacity.”

“Oh, using a program to play chess is work. Games are for fun. I guess I could have spoiled fencing, if I’d spent a year or two on the job. I didn’t want to.”

“You wanted to retain it as a nondeterminate analogy of your own predicament, because of its overtones of captivity, enclosure, secure ground and the like — is that it?”

“Think of it in any way you choose. I say the hell with it. One of the worst things wrong with people like you is inability to enjoy themselves. You don’t like the idea that there are processes that can’t be analyzed. You’re the lineal descendant, on the sociological side of the tree, of the researchers who pithed cats and dogs because even their personalities were too complex for comfort. Which is fine for studying synapse formation but no damned good for studying cats.”

https://full-english-books.net/english-books/full-book-the-shockwave-rider-read-online-chapter-16


#6

Well, they limited your ability to become world champion. Were you close?


#7

Chess is just the egg-head sudoku. I do not understand the relevance of this game to general intelligence.


#8

Poohs to that.

People were trying to wrote chess games in the 1950’s, sixty ears ago. The Kasparov .vs. Deep Blue game was twenty years ago and really Kasparov .vs. Deep Blue plus a large team of IBM chess players and programmers. On the last year or two we have had computers teaching themselves by playing games against themselves, and (apparently) learning by following the same paths as humans do. So, almost from the first computers, we have had the idea that they ought to be able to make themselves smarter, and it never worked.

And now, finally, it has begun to work. And it has also worked in a human-like manner, unlike the early chess programs whose advantage lay in incredible sacrificial combinations because they could look at all combinations 7 or 8 moves ahead which human. This suggests that their emerging intelligence is, perhaps, similar to ours, instead of being wholly alien (which it might have been).

The nice thing about games is that the computer could train itself. If you really want it to invent a game, then unless you can specify the properties of the ‘best game’, then someone is going to have to test a lot of duds. Imagine testing a billion variants on Monopoly. No thanks.

I can’t answer for you, but this for me is like seeing a child becoming an adult.


#9

It’s also made for a lot more boring robot battles than the ones I always imagined…


#10

The last chess engine I’d heard of was Fritz, which is apparently ancient news at this point.

It would be neat if there was some kind of rock-paper-scissors thing, whereby engine A could beat engine B but not engine C, but engine B would win against engine C, or something. But I suppose it’s simpler than that.
http://www.computerchess.org.uk/ccrl/4040/


#11

The dirty secret you probably never see is that each one these games probably cost $10,000 in compute time to win.


#12

You are using the traditional definition: “Real Artificial Intelligence is anything that a computer hasn’t done yet.”


#13

Personally I’m brushing up on my poker skills. When the AI have taken over most jobs, raising pre-flop to nits who run when they don’t have aces will be the only industry left.

(Apparently it’s much harder to automate a poker bot do to the number of possible plays/combinations being much bigger than chess)


#14

I wonder how long it will be until chess is solved, the perfect strategy found. Can’t be long now.


#15

a couple of hours of self training?

we are so fucked.


#16

Neural-net powered AI “crushes” top chess engine

Time to put it in the arena against a real opponent then!


#17

Amazingly, none of the actual games I play with other actual people have been remotely affected by computers being brilliant at chess.

Maybe at the highest levels we can blame an increase in drawish games on computers, but but that surely just makes watching the game boring, not the game boring.


#18

AIs are already beating the best humans at heads up. https://www.zdnet.com/article/researchers-reveal-how-poker-playing-ai-beat-the-worlds-top-players/


#19

Previously:


#20

Well, in HS I was a provisional high B-class player (you really have to be able to play in a number of tournaments to get a full rating; it is not easy). In practice, this meant I could stomp on most people, which was fun. But the best game I ever played was in a tournament where right off I was seated against the top rated person there, a Master. We played the French Defense, a seriously closed opening. I fought hard, but eventually lost. Still, my best game ever. You’ve no idea how grueling tournament play is.