AlphaGo

So it seems that the new Go AI Overlord is actually more smarter than humans when it comes to the game of Go. Which seems to be quite an achievement, because it's a complex game.

Chess computers apparently rely on just crawling the entire tree of possible moves rapidly to find the most optimal move, and basically waiting for the human player to make some mistake.

But this Go bot apparently actually outsmarts the best human players in terms of strategy, coming up with innovative & unusual tactics.. which is cool...

It comes up with moves which the commentators do not understand at first...

So within the limited universe of the game of Go, this is actually the first time a computer program is more intelligent than a human.

It's now leading 3-0 in a challenge against one of the best human players:

(skip to 15:00)

yes. cool stuff!

I don't know what the implications will be for AI research in general, but it beating a human in such a deep game as Go is damn cool. A computer beating a chess master really didn't impress me much, at the time it was impressive, but basically it can be simply by raw computing advantage on the very strict and limited rules of chess. Go is far more open-ended game when it comes to strategy and tactics. Aren't there more options in a Go game than there are atoms in the observable universe?

The way I see it, our own brain develops the way it does in largest part because of the steady stream of experiences fed into it's labyrinthine circuits. Every day is an exercise in how to survive, live, gain, do, and Go. Perhaps the neural networks that researchers are creating now have hit really close to home. Maybe this is how our brain works?

My intuition tells me that the brain can be conditioned like our muscles. Repeated exercise causes vigorous growth of new cells and structure, neuroplasticity.. By continuing to mimic our own neural networks in technologies i don't see why we won't continue to see some really insane AI algorithms unfold. The brain stores information somehow, it constantly adapts to new things, always in flux.

It also reminds me of mycelium. Fungal organisms are surprisingly advanced.. I think they achieve this because of the "networked" nature of mycelium. It started to sink in that maybe the mycelium in the ground isn't all that different from our own brain. It's just going about things differently, and has a really different method of intracellular communication. Not conscious, but encoded into the genome something really special.

http://www.wired.com/2016/03/sadness-beauty-watching-googles-ai-play-go/

This article was a great read on the subject. Most intriguing to me is how AI can approach things from a different perspective; e.g., the strange move that sent Sedol into a tizzy in game two.

  On 3/12/2016 at 12:59 PM, azatoth said:

I don't know what the implications will be for AI research in general, but it beating a human in such a deep game as Go is damn cool. A computer beating a chess master really didn't impress me much, at the time it was impressive, but basically it can be simply by raw computing advantage on the very strict and limited rules of chess. Go is far more open-ended game when it comes to strategy and tactics. Aren't there more options in a Go game than there are atoms in the observable universe?

Can you explain how Go is deep? Not quite sure on the rules ( I think it's something to do with getting 4 or 5 in a row?) but on the face of it, it looks much more simple than chess

no, you're thinking of this which is a different game:

the definitive Go intro:

  On 3/12/2016 at 7:56 PM, Danny O Flannagin said:

Can you explain how Go is deep? Not quite sure on the rules ( I think it's something to do with getting 4 or 5 in a row?) but on the face of it, it looks much more simple than chess

anyone any good at this?

I'm losing all the games to a program which is not this AI.

Go seems like a game you need to have played for years to be reasonably good at it.

Lee Sedol won the fourth game. AlphaGo resigned, apparently it did some mistakes earlier in the game. I think losing will probably improve it even more than winning.

  On 3/13/2016 at 9:47 AM, azatoth said:

Lee Sedol won the fourth game. AlphaGo resigned, apparently it did some mistakes earlier in the game. I think losing will probably improve it even more than winning.

Or perhaps Lee got a better handle on how to play Alpha.

Like was mentioned in the article posted earlier, where the european champion got involved in the development of Alpha and learnt himself from doing so as well. Kinda like playing tennis where you're used to playing righthanded players and all of a sudden are confronted with a lefty.

  On 3/13/2016 at 4:03 AM, phling said:

anyone any good at this?

I'm losing all the games to a program which is not this AI.

^ ya... I basically lost all motivation after a few days of not winning a single game against basic computer opponent lol

--------

so there is a new thing going on which is mysterious. A player named "Master" obliterating world top players in online games. Highly likely yet another AI, but it's just playing and winning and nobody knows who/what it is:

https://www.reddit.com/r/baduk/comments/5llntr/foxy_games_of_masterp/

http://china.org.cn/china/2017-01/04/content_40035096.htm