Home The Guardian No one can read whats on the cards for AIs future |...

No one can read whats on the cards for AIs future | John Naughton

16
0

AI is currently surpassing us at poker, although not even Google cofounder Sergey Brin can say with any certainty what the next steps for machine learning are

Ten days past, in a Davos still shellshocked by Trumps success, among the former masters of the universe sat down with the interviewer to talk about artificial intelligence (AI) as well as the future. His name is Sergey Brin and he’s among the cofounders of Google. The tone of the dialogue was muted, perhaps because Brin isn’t a master in Trumps universe, but mostly because hes a bright and considerate man although thoughtful. He answered that it was impossible to predict correctly and followed up using a narrative from his own expertise when asked regarding the near future of AI.

Ten years back, Brin was running around the place in the place where they work on jobs that have, Googles X laboratory, at best, a 100-1 odds of achievement. One small job there was called Google Brain, which focused on AI. Brin said, to be totally truthful, I didnt pay any attention to it in any way. Brain was headed with a computer scientist named Jeff Dean who, Brin remembered, would occasionally appear to me and say, Look the computer made an image of a cat! and I’d say, OK, thats really pleasant, Jeff go do your thing, whatever. Fast forward a few years and now Brain likely reaches each and every one of our primary jobs ranging from investigation to advertising everything we do to pictures. This revolution in webs that are deep undoubtedly surprised me even though I had been right in there and has been quite profound. Paper clips could, you know, throw at Jeff.

Fast forward a week with that interview and cut to Pittsburgh, where four top professional poker players are matching their wits against an AI software developed by two Carnegie Mellon university research workers. Theyre playing with a certain type of high-stakes poker called heads up no-limit Texas holdem. This system is known as Libratus, which can be Latin for balanced. There’s, nevertheless, nothing balanced about its operation. Just over halfway through the 20-day match, Libratus was $800,000 up on the people, some of whom were starting to feel depressed. I didnt realise how great it was until now, said one of them, Dong Kim. I felt like I had been playing like it may see my cards against someone who was cheating. Im not accusing it. It was simply that great.

The match doesnt finish until tomorrow (it is possible to follow it via the hashtag #brainsvsai) and so we dont know yet in the event the computer will win. But if it doesnt, another milestone was passed to artificial intelligence on the street, since the Pittsburgh competition has provided evidence for a machine can play a pretty mean game of poker against actual individual specialists. This can be a huge deal, because poker needs intellect and reasoning that have up until now eluded machines. Theres the machine along with no single best move must shift its strategies to make sure that its competitors discover that it’s difficult to think when its bluffing. Because an competitions hand stays concealed from view during play, poker, says a robotics specialist, Will Knight, is essentially not the same as checkers, chess, or Go. In games of imperfect information, it’s extremely complex to determine the perfect strategy given your competition could possibly be choosing to every potential tactic. And no-limit Texas holdem is particularly difficult because any sum could be fundamentally wager by an adversary.

The Pittsburgh competition will certainly take its position in a sequence of critical periods in the street to some sort of AI. Earlier ones have contained IBMs 1997 Deep Blue victory over Garry Kasparov in chess, its Watson machines success in the Jeopardy quiz game in 2011 and DeepMinds AlphaGo conquering the world Go winner in March 2016. Every one of these was, understandably, regarded as a spectacular accomplishment at that time, but the truth is they were baby steps about what looks like being an extended journey, despite all the hoopla about wisdom.

Read more: https://www.theguardian.com/commentisfree/2017/jan/29/no-one-can-read-cards-artificial-intelligence-poker-libratus-sergey-brin