• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Google’s AI beats world Go champion

M

Miles Vappa

Junior Member
You guys seem like smart people; what do you make of this:

http://www.bbc.com/news/technology-35761246

What i find interesting is the quote (which i'm somewhat paraphrasing) I heard on BBC newshour by Frederic Friedel, computer chess pioneer.

People are not talking about it enough, because it's coming.

People will have to convince me why in 20-30 years from now we won't have computers that have a general AI level beyond human or equal to us, beyond global catastrophes... as technology developes, it's coming, and once it's aware, it won't stop, as it will become a 100X, 1,000,000X more intelligence. What you should do is tell me why this will not happen because i still have not heard any reasons for it not happening.
Click to expand...

So what do you think? Are we going to have the singularity and robot overlords in 30 years?
 
ATOT has John Conner so I'm not too worried about that happening. The real question is when can we get our sex bots? :hmm:
 
Miles Vappa said:
So what do you think? Are we going to have the singularity and robot overlords in 30 years?
Click to expand...

no because people have no idea how computers work

beating humans at things isn't all that difficult.

Chess was brute forced - they simply tossed enough computing power at it to analyze all possible moves and pick the best one, it also could look up all boards with the exact layout and such at the same time

Go is more complex due to the sheer amount of moves available so it cant be brute forced, its more about pattern recognition. its near impossible to brute force as it would take billions of years. but with simply game theory /alpha/beta decision trees it can come to the "best outcome"

a computer can simply simulate more possible outcomes then a human . even if the best Go player can be 30 moves ahead the computer can be more

its not "learning" its simply building a database and using a decision tree,
 
Anubis said:
a computer can simply simulate more possible outcomes then a human . even if the best Go player can be 30 moves ahead the computer can be more
Click to expand...

(this first part is more for others who don't know why Go is more difficult)
In Chess there is a branching factor of 35, so to look 6 moves ahead that's 35^6 (1,838,265,625), a super computer doesn't have much trouble with this, or even 20-30 moves. However, Go has a branching factor of around 250, that means 6 moves ahead is 250^6 (244,140,625,000,000) or roughly 130,000 times as many possibilities. This is just for 6 moves, and Go has FAR more many moves than a chess game. Further, chess is fairly easily optimized, you can place a value on each piece and play strategies to protect your high value pieces, just recognizing a high value piece in Go is difficult because each piece can be strategically important.

Google's approach is different than the approach used by chess computers, they don't brute force as many turns out as possible, with Go it would take stupid amounts of computation power, Google is using an artificial neural network that is very loosely based off the neurons in the brain it uses pattern matching, and it's own experiences to learn from millions and millions of matches it runs against itself to teach itself Go.

In October it beat a 2-Dan, Lee Se-dol (todays opponent) is a 9-Dan (highest Go rank).
The fact it has taken a single game is fairly significant and was unexpected.
 
While I can understand your wanting them keep in mind that one of them doesn't mind killing humans. But she does drop her panties on demand. 😵
 
mnewsham said:
(this first part is more for others who don't know why Go is more difficult)
In Chess there is a branching factor of 35, so to look 6 moves ahead that's 35^6 (1,838,265,625), a super computer doesn't have much trouble with this, or even 20-30 moves. However, Go has a branching factor of around 250, that means 6 moves ahead is 250^6 (244,140,625,000,000) or roughly 130,000 times as many possibilities. This is just for 6 moves, and Go has FAR more many moves than a chess game. Further, chess is fairly easily optimized, you can place a value on each piece and play strategies to protect your high value pieces, just recognizing a high value piece in Go is difficult because each piece can be strategically important.

Google's approach is different than the approach used by chess computers, they don't brute force as many turns out as possible, with Go it would take stupid amounts of computation power, Google is using an artificial neural network that is very loosely based off the neurons in the brain it uses pattern matching, and it's own experiences to learn from millions and millions of matches it runs against itself to teach itself Go.

In October it beat a 2-Dan, Lee Se-dol (todays opponent) is a 9-Dan (highest Go rank).
The fact it has taken a single game is fairly significant and was unexpected.
Click to expand...

This. It was a pretty incredible feat.
 
Thread merge?



Anubis said:
no because people have no idea how computers work

beating humans at things isn't all that difficult.

Chess was brute forced - they simply tossed enough computing power at it to analyze all possible moves and pick the best one, it also could look up all boards with the exact layout and such at the same time

Go is more complex due to the sheer amount of moves available so it cant be brute forced, its more about pattern recognition. its near impossible to brute force as it would take billions of years. but with simply game theory /alpha/beta decision trees it can come to the "best outcome"

a computer can simply simulate more possible outcomes then a human . even if the best Go player can be 30 moves ahead the computer can be more

its not "learning" its simply building a database and using a decision tree,
Click to expand...
I don't think we're much different.
- The decision tree and database are enormous and very complex.
- We can't plug in a debugger to determine the state of a human brain at any instant, so it's something of a mysterious black box.

It doesn't mean it's something that can never be paralleled or exceeded. It just means it's complicated and poorly understood.
 
Last edited:
Miles Vappa said:
So what do you think? Are we going to have the singularity and robot overlords in 30 years?
Click to expand...

Overly ambitious. Maybe 50 years, probably 100+. Too bad I'll be dead by then, I for one would welcome our new robot overlords. Humans are too stupid, time to get rid of us and let the Earth move in a new direction.
 
I think the more pressing concern would be AIs that can take over white collar jobs. We're going to have to figure out what our society will look like in the next fifty years.
 
They played the second match either today or yesterday, and AlphaGO won again, so it's 2 for 5.

According to the Human opponent (paraphrasing again); he was "Shocked, as AlphaGO played a nearly perfect game"

So it's not looking good for us humans who want to be the planetary GO champions.

XavierMace said:
I think the bigger news is that there's a world championship of Go.
Click to expand...

Go has been around for over 2500 years, so no surprise there is a Go championship (not that you weren't being sarcastic, but I have to be a dick).
 
Last edited:
Miles Vappa said:
Go has been around for over 2500 years, so no surprise there is a Go championship
Click to expand...

Wasn't Lee Se-dol just beaten 8-2 a couple of days ago by another human? It's not like Lee Se-dol is actually considered the world champion (he isn't, he IS the highest professional rank however)

Still a very high ranked Go player, though I would like to see a match between the worlds BEST and AlphaGo, I don't know if there is an agreed upon singular best Go player however.
 
mnewsham said:
Wasn't Lee Se-dol just beaten 8-2 a couple of days ago by another human? It's not like Lee Se-dol is actually considered the world champion (he isn't, he IS the highest professional rank however)

Still a very high ranked Go player, though I would like to see a match between the worlds BEST and AlphaGo, I don't know if there is an agreed upon singular best Go player however.
Click to expand...

I don't know if that'd be matter. AlphaGo AI is on another plane of skill.

Se-Dol just made a statement yesterday that a new Go player has surfaced that is on another plane of skill from us humans.

And not for one moment he felt he was ahead. He was outplayed in every possible way.

Let's theoretically say there's someone who's 10-20% better than Se-Dol (which probably doesn't exist). Even that isn't even close.
 
Last edited:
Zeze said:
I don't know if that'd be matter. AlphaGo AI is on another plane of skill.

Se-Dol just made a statement yesterday that a new Go player has surfaced that is on another plane of skill from us humans.

And not for one moment he felt he was ahead. He was outplayed in every possible way.

Let's theoretically say there's someone who's 10-20% better than Se-Dol (which probably doesn't exist). Even that isn't even close.
Click to expand...

Oh I doubt it would matter much either, I would still enjoy seeing it though.
 
GagHalfrunt said:
Overly ambitious. Maybe 50 years, probably 100+. Too bad I'll be dead by then, I for one would welcome our new robot overlords. Humans are too stupid, time to get rid of us and let the Earth move in a new direction.
Click to expand...

what do you mean? we've done some pretty awesome things like create airplanes, foie gras, and Stevie Ray Vaughn.

:colbert:
 
So did they program this computer on how to play GO or did they just have it watch and learn? There's a huge difference.

Edit: Sounds like it was programmed with at least a basic set of rules and continually gets better. So a bit of both but still not completely what I would call AI.
 
Last edited:
MixMasterTang said:
So did they program this computer on how to play GO or did they just have it watch and learn? There's a huge difference.
Click to expand...

I forgot the exact numbers, so just making them up now. But the magnitude is similar.

AlphaGo has 2 main parent programs. 1 is data driven- it has a history of hundreds of millions of top kill Go matches. Also it plays 30K-50K games a month.

Second one is machine learning. It learns and gets better at it for every match it plays out of 300 millions it's already played.
 
What's astonishing is that yesterday's match, during early game, AlphaGo made one move that seemed to make no sense.

All the top tier (9-dan) analysts were baffled. It almost seemed like a bad manner or a mistake.

Couple hours later, that strange 'out of position' stone came into play.

Amazing calculation.
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top