AI can now win at Go

[Crispy]

Google's AlphaGo AI defeats human in first game of Go contest

In the first of 5 games, Google's DeepMind has defeated Lee Sedol, one of the top Go players in the world. The computer runs several massive neural networks to analyse thousands of experts' games, evaluate its own play, and narrow the almost infinite set of possible moves down to a more manageable number. It's an amazing achievement and it is now inevitable that the machines will destroy/save us.

 

[JimW]

Still can't rob you blind at mah jong while spitting out sunflower seed husks.

 

[Pickman's model]

yeh. but how is this marvellous machine at the likes of cluedo or monopoly?

 

[discokermit]

your jokes show you do not understand the gravity of this situation. or the complexity of go.
we are doomed.

 

[Crispy]

2-0

Dig your graves now.

 

[2hats]

it is now inevitable that the machines will destroy/save us.

Unless we tie them up for many years getting them to ponder the answer to a really difficult question (in traditional sci-fi fashion), which is clearly the intent of the designers here given the name they have chosen for their issue...

 

[bi0boy]

Bring it on. By the time I'm 65 I want an exoskeleton wired into my brain if they don't have full limb and organ replacement via stem cells already.

 

[Santino]

How is it at kick-boxing?

 

[DotCommunist]

Unless we tie them up for many years getting them to ponder the answer to a really difficult question (in traditional sci-fi fashion), which is clearly the intent of the designers here given the name they have chosen for their issue...

great c clarke (think its him) short story ,they get a thinking machine to count/say all the names of god. When its done the stars start to go out.

but yeah, go. I always thought it was just complicated Reversi but holy shit is it worse than that.

 

[discokermit]

Bring it on. By the time I'm 65 I want an exoskeleton wired into my brain if they don't have full limb and organ replacement via stem cells already.

it'll be a bit like that except you will be the exoskeleton.

 

[bi0boy]

it'll be a bit like that except you will be the exoskeleton.

Hell yeah. I bet the robots will be better than the Tories.

 

[discokermit]

great c clarke (think its him) short story ,they get a thinking machine to count/say all the names of god. When its done the stars start to go out.

but yeah, go. I always thought it was just complicated Reversi but holy shit is it worse than that.

they reckon if go is like war, chess is a knife fight in a phonebox.
and chess is very very complicated.

 

[beesonthewhatnow]

As long as we can pull out the plug we should be fine

 

[discokermit]

As long as we can pull out the plug we should be fine

the plug will go into your skull, just behind your ear.

 

[editor]

The human could still fight back*!

Google’s Go-playing machine has scored a second victory against the best human player.

The victory by AlphaGo against South Korea’s Lee Sedol, the winner of 18 world championships, puts the machine’s owners one victory away claiming a $1m (£700,000) prize.

AlphaGo’s first win against Lee, on Wednesday, shook the Go-playing world, marking a milestone in the development of artificial intelligence.

Many had believed it would take another decade for computers to conquer the ancient Chinese board game, one of the most creative games ever devised.

After his first loss, Lee said he was in shock as he had not expected to lose. Google’s team compared AlphaGo’s win to landing on the moon.

The five-game series is scheduled to run until Tuesday.

Google's AlphaGo wins second game against Go champion

*unlikely, it seems.

 

[DotCommunist]

the combined hubris here is staggering:

Lee said he was in shock as he had not expected to lose. Google’s team compared AlphaGo’s win to landing on the moon.

I'm on team lee.

 

[Jay Park]

Lee pulls it back to 3 - 1

 

[littlebabyjesus]

the combined hubris here is staggering:


I'm on team lee.

Garry Kasparov was confident he'd never lose.

I don't know the details of this Go-machine, but we need to be careful not to overhype the achievement. I posted this on another AI thread:

Here's a chess puzzle which a computer that had beaten grand masters could not solve.



White to move.


Its failure shows a lack of that slippery thing 'understanding'.

 

[DotCommunist]

well I suppose you don't get to the top of any competitive sport/game with the attitude of 'I will lose this one'.

 

[Jay Park]

Garry Kasparov was confident he'd never lose.

I don't know the details of this Go-machine, but we need to be careful not to overhype the achievement. I posted this on another AI thread:

Here's a chess puzzle which a computer that had beaten grand masters could not solve.

View attachment 84642

White to move.


Its failure shows a lack of that slippery thing 'understanding'.

Dare I say, king?

 

[littlebabyjesus]

well I suppose you don't get to the top of any competitive sport/game with the attitude of 'I will lose this one'.

He then worked out better ways of playing machines. You can't bully a machine. You can't bluff it. In chess, the only way to beat a good machine is to play a long, rather dull developmental game. I only vaguely know the rules of Go, but I'm sure it's similar.

 

[Buckaroo]

Lee pulls it back to 3 - 1

AlphaGo has decided it can't be arsed to play the last game because it's already made it's point and has other shit to do.

 

[littlebabyjesus]

Dare I say, king?

Yep. Shove your king around wherever you like for as long as you like. Black can't break through and you can't lose. You also can't win, clearly.

The computer took the rook.

 

[DotCommunist]

I looked up the rules of Go some time ago after encountering it in a rather shit thriller paperback where it was used to show that a western badass had truly absorbed chinese culture and was now a blend of occident and orient etc. rubbish. But yeah it looks intricate.

Lee pulls it back to 3 - 1

excellent news

FUCK YOU CYBERDYNE! YOUR BOY TOOK A HELL OF A BEATING! A HELL OF A BEATING!

 

[Jay Park]

Yep. Shove your king around wherever you like for as long as you like. Black can't break through and you can't lose. You also can't win, clearly.

The computer took the rook.

Did they manage to uncover why it continued ploughing pawns forward, and how different was the machine Kasparov played to Alphago?

 

[littlebabyjesus]

Lee pulls it back to 3 - 1

For me, even if the computer wins the series, this shows that computers still haven't cracked go. Not properly - if they'd done it properly, they'd win every time. And I would also wonder how the human is learning as they go on, as Kasparov did - the human needs to learn humility to beat machines.

 

[Jay Park]

He's a confident boy is Lee ( or so i'm told ). I'm calling it 3-2 over the series.

 

[Lord Camomile]

But can it do it on a wet Tuesday night in Stoke?

 

[littlebabyjesus]

Did they manage to uncover why it continued ploughing pawns forward, and how different was the machine Kasparov played to Alphago?

That's a good question - the puzzle was devised specifically to trip the computer up - by Bill Hartson (who presented the Chess show on the bbc in the 80s, if you remember that). This was late 90s. No doubt computer chess has improved since, but this was the state-of-the-art machine at the time.

One answer is that the computer did not realise it couldn't win. That's because the computer doesn't know it is playing a game and doesn't know what 'winning' is. But I don't know exactly what lessons were learned from this puzzle.

 

[Signal 11]

The computer took the rook.

Crafty said Kc3 after less than a minute.