Google built an AI that just dominated a Go champion

[dkpope]

The DeepMind division of Google has something to celebrate. It’s AI just beat a top ranked Go player in five straight matches. Computers have been beating humans in chess and other games for a long time, but Go is different. According to the Google blog post, the possible moves in the game outnumber the atoms in the universe – that's “more than a googol times larger than chess.”

The system behind the Go dominance is AlphaGo. DeepMind built the system and then filled its neural network with 30 million moves from professional players. The next step was for AlphaGo to learn to create its own strategies. How? By playing itself (the only worthy opponent) and using reinforcement learning (a kind of trial and error process.) Not surprisingly, that took a ton of processing power so Google used the cloud extensively (a nice way for them to plug another part of their company.) There’s no mention if they played the Rocky soundtrack while AlphaGo was in training.

Next, Google invited reigning three-time European Go champion Fan Hui to their London office to play behind-closed-doors. AlphaGo lived up to its name and won 5 games to 0. Google tried to not be too braggy and said the victories were "just one rung on the ladder to solving artificial intelligence." The computer’s next match is in March against world champion Lee Sedol.

Permalink to story.

https://www.techspot.com/news/63641-google-built-ai-dominated-go-champion.html

 

[Squid Surprise]

Would be interesting to see how it does against other champions... I vaguely recall Kasparov being a little miffed when he lost to Deep Blue, saying that the computer had been trained to "beat him" but wasn't really that good a chess player... A few years later, and noone can beat the computer...

When a computer program exists that can beat me 7 straight matches in cribbage, I'll be impressed...

 

[QuaZulu]

When a computer program exists that can beat me 7 straight matches in cribbage, I'll be impressed...

I wouldn't be surprised at all if...the computer was dealing. Have you ever been beaten in 3 hands before?

 

[Squid Surprise]

I wouldn't be surprised at all if...the computer was dealing. Have you ever been beaten in 3 hands before?

Can lose any given match... It IS cards... But never 7 in a row - I am ranked #1 in the world for a reason after all...

 

[MilwaukeeMike]

I get it that it's an achievement, but I wish they'd devote all those dollars and time and computing to a problem that would be of some benefit when they're done.
Go solve rush hour traffic, or the growing ineffectiveness of antibiotics, or balance the US budget.

Nope... they build the smartest computer ever created to one thing...play Go. A game many of us had never heard of, but hey, it has a million, billion gazillion combinations of moves... Yay google. :/

 

[Seraphim401]

I get it that it's an achievement, but I wish they'd devote all those dollars and time and computing to a problem that would be of some benefit when they're done.
Go solve rush hour traffic, or the growing ineffectiveness of antibiotics, or balance the US budget.

Nope... they build the smartest computer ever created to one thing...play Go. A game many of us had never heard of, but hey, it has a million, billion gazillion combinations of moves... Yay google. :/

Be careful what you wish for.

 

[Camikazi]

I get it that it's an achievement, but I wish they'd devote all those dollars and time and computing to a problem that would be of some benefit when they're done.
Go solve rush hour traffic, or the growing ineffectiveness of antibiotics, or balance the US budget.

Nope... they build the smartest computer ever created to one thing...play Go. A game many of us had never heard of, but hey, it has a million, billion gazillion combinations of moves... Yay google. :/

Yeah, a computer figuring out how to play and win a game that has a massively massive amount of moves won't be a good stepping stone to being able to figure out things that has large amounts of variables... did you miss something there? Computers like this have to start somewhere and most will start with games that won't hurt anyone and are sufficiently complex enough to prove that they are working. Hell, Watson is one of the smartest computer around now and it is making food recipes that people can search for.

 

[Squid Surprise]

Bah.... NOW it's playing chess and go... Tomorrow, it's called Skynet and launching nukes at Russia...

And trying to kill Sarah Connor

 

[MilwaukeeMike]

Yeah, a computer figuring out how to play and win a game that has a massively massive amount of moves won't be a good stepping stone to being able to figure out things that has large amounts of variables... did you miss something there? Computers like this have to start somewhere and most will start with games that won't hurt anyone and are sufficiently complex enough to prove that they are working. Hell, Watson is one of the smartest computer around now and it is making food recipes that people can search for.

Watson is being used to do complex evaluations of companies financials (and a thousand other variables) to determine if they're worth investing in, not only making food recipes. They need a problem with a million variables, why not try for climate science or something actually useful. There's nothing wrong with having your stepping stone be an achievement in itself.