AlphaZero

Google's AI team is at it again. They published a new paper where they describe a generic deep learning algorithm that easily beats all state of the art engines in various board games including Chess, Shogi and Go. This algorithm learns from zero by playing against itself and requires no training data such as human games. It just needs to know the rules of the game.

arxiv.org/abs/1712.01815

Other urls found in this thread:

lichess.org/study/EOddRjJ8
theatlantic.com/health/archive/2016/11/the-link-between-autism-and-trans-identity/507509/
twitter.com/NSFWRedditVideo

For example, the AlphaZero program didn't lose a single game out of the 100 played against last year's computer chess champion Stockfish.

Once again another great work by white men.
Shitskins are worthless animals.

>engine elo is lower in chess
Does this mean chess is the harder game for engines?

Hyuck hyuck, dang right Cletus! Now we just need to roust these soyboys!

Houdini is a better engine than stockfish

>Antonoglou, Lai, Guez, Marc Kumaran, Simonyan, Hassabis
>white

>It just needs to know the rules of the game.

Too bad the world doesn't work like that. Rules are constantly changing and I doubt a machine can identify them. It had to be able to interpret the world like a human who grew up in this world and made experiences. Good luck with that.

are the games published anywhere? I'd like to see some of those wins, especially with black.

It's just another technique for solving a type of constrained optimization problem, nothing more. Or does anyone actually buy the general intelligence meme?

the next step would be to create an engine that figures out the rules of chess just by looking at piece movement and piece disappearane

at some point, all important decisions, whether executive, judicial, military, or financial, may be decided by AI, since it knows the optimal path for greatest gain.

Yet again, shifting the goalpost. Infering the rules is nothing more than another optization problem. Keep denying machine learning will replace most human workers in the next century, you snowflake.

I could be totally wrong but.. go oompared to chess has many more possible moves to compute, thus the elo ceiling might just be lower in chess because of that.

>Keep denying machine learning will replace most human workers in the next century, you snowflake.
Good job making your claim unscientific and unfalsifiable.
>someone brings major exception to claim
>but I didn't say all workers would be affected!!!!11!!

It's in the end of paper
lichess.org/study/EOddRjJ8

That's not hard to do

AI will be regulated
No then life would lose all purpose. Humanity won't let that happen
And what did you contribute?

very interesting that this beat Stockfish, I thought chess engines stopped getting much stronger

since humans tweaked the parameters of Stockfish to increase its strength, it seems impure in comparison

What? It's the most falsifiable claim of them all. You simply have to wait another 82 years. Either humans will have been replaced by then or not.
You massive faggot.

Does Stockfish favor moves to counter specifically human moves? Maybe AlphaZero, as in the case with Go, developed a moveset for chess, which deviates from human behavior.

>Too bad the world doesn't work like that. Rules are constantly changing and I doubt a machine can identify them.
This is probably the smartest post on Sup Forums.

Seems like we've hit a ceiling with chess computers, only processing power is the limit. Kind of boring. Probably explains why Carslen is the champ too, he's just more autistic than the competition and can rush for endgame knowing his autism will help him find the right moves

what hardware was used for stockfish?

Pretty much the only thing AI is good for is emulating autism.

theatlantic.com/health/archive/2016/11/the-link-between-autism-and-trans-identity/507509/

transgender AIs when?

>Game: Chess
>Black: AlphaZero
>Win: 3

Welp, that's it.
>pic related.

>Unfalsifiable, unscientific
How about this. AI will outperform human drivers in the next decade. AI will outperform human in investing in the next decade. AI will outperform human in compilation in graphics animation, texturing and to some extent modeling in the next decade. AI will outperform human in music in the next 3 decades. AI will outperform human in any game in the next decade. AI will outperform human in any management task in the next decade. I can go on and on. Some of those are very close to being true already, these are pessimistic guesses. Not many jobs are safe from current "generation" of AI.
AI will only be regulated in countries that want to be irrelevant. Do you think it will stay regulated in US if China goes full AI?

Yes, just like the many other international treaties

You can see Stockfish won 3 games.

elo doesn't have a ceiling, it's just the likelihood of a win when compared with the elo another engine/human
it's also completely independent of the size of the game tree
lower engine elo implies that there is a smaller likliehood of beating the average player than with a higher engine elo (usually 1500, and assuming the average is the same for all games in the test used)

it's because chess has so fucking many draws.

...

You're not reading the table right.

Hassabis is solid 56%

Seems definitely very impressive.

Notched

>iPhoneX user

When well they try ariimaa?

Computers can already beat humans in Arimaa.

If you make it play Tic-Tac-Toe does it realize that it's pointless to play?

chess needs a balance patch to buff black

Oh, I see, 'from AlphaZero's perspective'. Poorly made table if you ask me.

Just allow the person with the lowest score to choose the side. Problem solved.

No, it will play perfectly in a very short while.

It wasn't designed to be hard to beat?

Then why bother wasting time trying to beat humans in it?