Google's DeepMind AI Becomes a Superhuman Chess Player In a Few Hours (theverge.com) 93
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."
Strange game (Score:3)
The only winning move, is not to play
Re:Strange game (Score:4, Funny)
Re: (Score:3)
Why would you expect 99% unemployment? This AI will never be able to:
-fix your plumbing
-rack you servers
-move your furniture
-change your spark plugs
-etc, so forth, and so on.
Re: (Score:2)
Don't be so sure [popsci.com]
As soon as a humanoid robot [boston.com] is perfected, the only value people will have is knowing where the aim points are.
Re: (Score:1)
This one? Probably not.
You really think humans are so special that no other AI could ever do all those things? Better hope the AIs never hunt people for sport in order by Slashdot ID.
Re: (Score:1)
Better hope the AIs never hunt people for sport in order by Slashdot ID.
Well, hopefully they will start in reverse order. :P
Re: (Score:2)
So you aspire to be a mechanical actuator?
At any rate, how are you sure that an AI won't eventually be able to teach itself mechanical control system skills that rival humans? Mice and birds with pea-sized brains are able to navigate the physical world rather effectively.
Because we understand progress (Score:5, Insightful)
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.
Re:Strange game (Score:4, Insightful)
Candiate 1: "I traded stocks on Wall Street for 20 years prior to having my job automated."
Candiate 2: "I operated a fork lift in a warehouse for 8 years before the facility was automated."
Candiate 3: "I drove semi trucks for 15 years before the robots came in."
And so on.
The thing about AI and automation is that as human workers are displaced, they shift to job types that are financially unattractive to automate-- like those categories you cite. With the flood of displaced workers in these job areas, wages are diluted. "A plumber always makes a good living" will no longer be a true statement as the plumber job market becomes oversaturated by workers displaced by automation.
Re: (Score:3)
The thing about AI and automation is that as human workers are displaced, they shift to job types that are financially unattractive to automate-- like those categories you cite.
Those jobs aren't financially unattractive to automate, they're still a way beyond our current level of tech.
1983 And I Am Dreaming Of Nuclear Warfare (Score:1)
Did everyone miss the movie reference [imdb.com]?
Re: (Score:2)
The only winning move, is not to play
When algorithms will be clever at societal things instead of games, it will become more difficult not to "play".
Teach it Starcraft Civilization (Score:3)
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.
Re: (Score:3)
Please have it learn to play politics, so we can....
Re: (Score:2)
They have [wired.com]. Facebook too [engadget.com].
Those games have a hell of a lot more complexity too, so it's no wonder it's a hard problem to solve. Resource management, army counter/order management, base creation, etc...
Re:Teach it Starcraft Civilization (Score:4, Interesting)
They are working on this.
https://deepmind.com/blog/deepmind-and-blizzard-open-starcraft-ii-ai-research-environment/ [deepmind.com]
Re: (Score:3)
I agree. If AI machines can play these games, then the gamers will be freed for a more productive use of their time.
Re: (Score:1)
I suspect 4x style games will be among the hardest computer games for AI to tackle for several reasons. Just off the top of my head.
Re: (Score:2)
I'd love to see what kind of city designs and road systems it might come up with in Cities: Skylines.
Micro transactions (Score:2)
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 imp
Super Human? (Score:3, Insightful)
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.
Re:Super Human? (Score:5, Insightful)
>, were actions that may not create a net positive or negative are used over when the net outcome is positive.
Which is still a net improvement over humans, who may stick with actions that are actually net negative despite proof if they initially miscategorized them as positive.
What they should get the AI to do to minimize such artifacts is have a meta-analysis going where the positive associations are re-evaluated whenever the overall victory is judged to not be at stake in the event the action was correctly evaluated in the first place.
Re: (Score:2)
"It didn't teach itself to play at all. The rules are programmed in. What's a legal move. What's a win. All programmed, not learned."
Isn't that what humans do? We're taught what's a legal move, what's not, why they're not, what's a win and why it's win. We're programmed as well, our interface and language is different.
As for watching to learn... You can watch to learn, but for most humans, there's an explanation as to why things are and aren't allowed, what happened, why, who scored and why, etc...
Re: Super Human? (Score:2)
Re: (Score:2)
"a really long chess game with non-strategic moves, as the most optimal path, may not be enforced correctly."
Well, at least with humans, pushing a game very long will stress the other opponent in to making a mistake if they're not as well equipped for it as you are. That's not sub-optimal, just longer. If you can't beat them straight up in logic, then you switch to alternate tactics.
Now, this may not apply to an AI as they don't tire like we do.. However, it's entirely possible, that if their logic systems
In other words... (Score:2)
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
And next: (Score:5, Insightful)
The world's gonna be an... interesting... place once someone merges this sort of code with virus code.
Re: (Score:2)
Re:Is it AI? (Score:4, Insightful)
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"...
Re: (Score:2)
Re: (Score:2)
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 quit
Re: Is it AI? (Score:2)
Re: (Score:2)
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.
Re: (Score:2)
Cool. (Score:2)
Re: Cool. (Score:1)
Re: (Score:2)
what the fuck is it with people feigning absurd levels of majestic boredom about an article on a piece of tech that blows out of the water anything in the field from 3 years ago.
Re: (Score:2)
Re: (Score:2)
I'm pointing out that intelligence requires awareness and motivation that is not programmed by an external source
That would disqualify your own brain. Your motivation to survive is programmed by external sources.
Re: (Score:2)
What was AlphaZero running on... (Score:2)
Re: (Score:2)
Re: (Score:2)
Can it run on a desktop? (Score:2)
But can it play Mario? (Score:1)
Re: (Score:1)
I do think this is most interesting if we don't allow the AI to adjust the game clock at all: if there's too large a delay between input and output, it will simply fail.
There have been AI's designed explicitly to play Super Mario World (see here [youtube.com], for instance). Where this becomes cool is with an AI that doesn't get an abstract representation of the level, but has to interpret the pixels that are displayed. That's a far more complicated problem, as the information displayed on even a simple game like Super
Not Just A Bigger Board (Score:2)
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.