Slashdot Mirror
Google's DeepMind AI Becomes a Superhuman Chess Player In a Few Hours (theverge.com)
An anonymous reader quotes a report from The Verge: In a new paper published this week, DeepMind describes how a descendant of the AI program that first conquered the board game Go has taught itself to play a number of other games at a superhuman level. After eight hours of self-play, the program bested the AI that first beat the human world Go champion; and after four hours of training, it beat the current world champion chess-playing program, Stockfish. Then for a victory lap, it trained for just two hours and polished off one of the world's best shogi-playing programs named Elmo (shogi being a Japanese version of chess that's played on a bigger board). One of the key advances here is that the new AI program, named AlphaZero, wasn't specifically designed to play any of these games. In each case, it was given some basic rules (like how knights move in chess, and so on) but was programmed with no other strategies or tactics. It simply got better by playing itself over and over again at an accelerated pace -- a method of training AI known as "reinforcement learning."
The only winning move, is not to play
So rise up, all ye lost ones, as one, we'll claw the clouds.
Please have it learn how to play modern strategy games like Starcraft and Civilization so we can have computer players which don't suck without massive bonuses which change the dynamic of the game.
-- All that is necessary for the triumph of evil is that good men do nothing. -- Edmund Burke
Reinforcement Learning systems have a tenancies of creating "Superstition" artifacts, were actions that may not create a net positive or negative are used over when the net outcome is positive. It often creates less than ideal outcome, but still it works. So this could mean a really long chess game with non-strategic moves, as the most optimal path, may not be enforced correctly.
If something is so important that you feel the need to post it on the internet... It probably isn't that important.
can it see why kids love Cinnamon Toast Crunch?
It simply got better by playing itself over and over again at an accelerated pace -- a method of training AI known as "reinforcement learning."
Like it was playing Global Thermonuclear War with zero players...
LK
"Hi. This is my friend, Jack Shit, and you don't know him." - Lord Kano
The world's gonna be an... interesting... place once someone merges this sort of code with virus code.
Check your premises.
No, not this one. Not even the next one. The one after that? Or after that?
Eventually, they will. The question is simply how long will that be. Right now, the ML pace continues to accelerate. Soon, they'll be stacking one skill upon another. The skill to walk. The skill to understand plumbing joints and leaks. The skill to know home construction. Etc.
It's coming. That whole "will never be able to" business... that's not going to pan out for anyone.
I've fallen off your lawn, and I can't get up.
David Fogel has inspired many of us with his clear and concise explanations. His book, Blondie-24 is a great introduction.
It's good to see the rest of the world catching up.
Did everyone miss the movie reference?
Alternative Right.
See subject: ... & I've played 1,000's of games of it (in the rain even) w/ roommates of mine since 1994 or so (usually drinking beer of course, makes it more fun w/ THAT 'handicap')!
* REPETITION & learning the other guy's style - it's what I did, & yes, had done to ME (but there's a wee bit 'more' to that & me being 'sly').
FUNNY STORY ON THAT - When the roomie I had I played 1,000's of games against (he had me by 20 games @ 1st, & w/in months I had HIM by the much) left? 2 new roomies moved in (young dudes, 23 & 24 respectively) telling me "Oh, you play chess? I am 'great' @ it, let's play!" & they wanted to use their 'custom sets' (where the pieces look too weird for me - no, for me, it's PURE plastic std. cheap shit, lol) - I said "Sure, I'll play, but not w/ your funky sets" & I WHIPPED THEIR ASSES badly (lmao - even BOTH of them together once)).
In any event - it is, w/out question the GREATEST board game ever!
APK
P.S.=> Eventually, & I told them that THIS would happen? "You boys will eventually get my 'style of play' down & start beating me!" which did happen (23 yr. old could beat me every 3rd game or so & the 24 yr. old, a SERIOUS GAMER & competitor type, caught up to me & I had only 1 game on him @ the end before he moved out (I wouldn't give him that last game to 'tie' either - not on HIS terms or when he was 'hot' but ONLY if I felt up to it (lol, I didn't want to give him a tie & held off on purpose))... apk
Comment removed based on user account deletion
They only succed for games for which is easier to quantify and optimize some success function. They did not achieve too much when trying Starcraft 2 which is quite a complicated game. It could not beat a bronze league player.
Well, the program playing itself is not really qualitatively different than "if I do this, and he does that, and I do the other, and he does........ then I win!"; it's just carried out to more steps than a human would (because a human can't go that far). Therefore, any approach I can conceive of to go from knowing the rules to knowing how to win is pretty much equivalent to "running some iterations". Even the ability of human chess masters to perceive the board as a pattern instead of just a bunch of individual piece positions is probably approximated by something in the program.
Given that, I am unable to come up with a mechanism to go from "knows the rules" to "knows how to win a game" without doing something equivalent to "running iterations"...
Wake me when it can decide, on its own without human directive, that it wants to play chess in the first place.
vs. what was Stockfish running on?
Let's see when the AI can be consistently good at Backgammon.
So like a human, you tell the person the rules, they give them zero thought, and play zero games, but are an expert? That wouldn't be SO. That would be Magical Intelligence.
Guns don't kill people; Physics kills people! - John Lithgow as Dick Solomon on Third Rock From The Sun
It simply pushed the "calculated lookup' horizon futher out by doing it in advance.
Is it open source?
File under 'M' for 'Manic ranting'
I'd love to see if this AI can learn to play a more complicated game like Super Mario World given only: 1) The pixels displayed as input. 2) Fail conditions (when a life is lost). 3) Basic map navigation rules (bonus if these can be eliminated and the game can be judged only on whether or not it gets a game over or completes the final level). 4) Valid controller inputs. I do wonder how this AI would translate from the turn-based world of Chess and Go to realtime.
I think I'd like to play Superhuman Chess, it sounds much cooler than regular Chess.
Wake me with a slap in the face, Sir AI.
I was deeply into AI back in the 70's and 80's. In 1986 for my thesis I created software that ran on the then powerful, now pitiful, hardware of the time that learned to play any game on an NxM grid. It could be used for many other things but that was how I did it for the thesis. Chess was what I demoed for my presentation. It blew away the profs because it went from zero skills to beating them in just a few hours and some of them considered themselves very good at chess. Later I set it up playing against it's own clone once it had the basics of various games and they would rapidly become extremely good players.
Realize that our iPhones of today have about 1,000x the computing power of what I was using back then per user slot.
I was approached by a medical research group who wanted to use for diagnostics. We were working on that. Then wind got to the military and I got a visit. They wanted it. Problem was I didn't want them. So I dumbed it down and purposefully flubbed the presentation. They went away thinking the person who had told them about it exaggerated and I went back to my research but kept it out of sight. That mean nixing the medical application.
Systems that learn have been around for a long time. I often wonder how many other conscientious objectors like myself chose to hide their research rather than let it go to the military.
I found other ways to make money. It is a shame it didn't get used in the medical field but to expose it there would have let it to be used in smart military systems which I didn't want.
I knew that eventually they would catch up. As it was, I kept it out of their hands for 30 years so hopefully society has time to mature. The late 1980's was a very bad time for the military to get their hands on something like that.
It's called reinforcement learning which is a subset of machine learning which is a subset of AI. So yes, it's AI. The exact steps to reach the end state weren't directly pre-programmed by a programmer. Any software which makes a decision based on some evaluation function can be considered to contain a weak AI. It has to be slightly more than a bunch of if-else statements written by a programmer, but only slightly more.
All humans and animals use reinforcement learning. Neurons which fire together eventually wire together.
What is funny is every chess program for the last 30 years has played itself for many moves before picking the best possible move. I remember in the 80's where if you put the computer to the hardest difficulty it would take a few minutes to make each move.
I would hope a program, AI or not, would beat a human since you can't trick something that can look ahead 100+ moves.
I find all this AI stuff as stupid, it is just a program.
Because humans don't do that. Why would we expect AI to do it? At least you don't have to worry: you're so illogical, no program could ever replace you. Obsolete, yes. Replace, no.
What if it tried to play against itself in order to lose? And then just did not do that.
They should feed it the world wide tax laws and give it a target to minimize tax paid (legally.) That would be interesting... google may already do this...
Comment removed based on user account deletion
Better yet, make it play one of the newer games that is all about micro transactions and pay to play.
Pretty sure the AI will come to the determination that they are retarded and refuse to play anymore. Either that or IBM or whoever will have their profit margins cut by a massive credit card bill...
Then again release enough AI's onto the market grinding an infinite number of games for credits, buying up all the good stuff, making the game, and the micro transactions useless might actually have a positive impact by influencing game makers to stop doing that anymore.
Ah, I see what you mean. No, an exhaustive search algorithm isn't what I'd call "intelligent". But an exhaustive search for chess would take a lot longer than a few hours, and a process that develops some sense of "this move will be bad" without having to try it every single time does seem, while not necessarily "intelligent", to be at least one step up from brute force, because it is making decisions based on, well, not unknown values of variables (not much in chess is invisible) but on situations not quite the same as what's been seen before.
Someone really should throw this thing at Magic The Gathering, once it masters poker.
I'm sure it could come up with some really ingenious decks if given the time.
I don't think the principal difference between shogi and chess is board size. In Shogi, you can place the pieces you capture onto the board as your own pieces. Having paratroopers is a lot different.