Suck It, Humans! Go Gone.
The toughest game, the one that emphasized human intuition, seems to be gone. Google's AI machine, AlphaGo crushed the best human again today, and it wasn't even close. Lee Se-dol, the best human player in the world, and plausibly the best human player ever, went down again and this time nobody could find significant mistakes in his play. From the New Scientist:
In the minutes after today’s match ended, Jackson said AlphaGo’s flawless play left him in shock. “Things that looked questionable in hindsight turned out to be correct. That’s its hallmark.”
At yesterday’s post-game press conference, Lee looked shell-shocked. Today he seemed resigned. One reporter asked what AlphaGo’s weaknesses are? Lee laughed: “Well, I guess I lost the game because I wasn’t able to find out.”
This is a huge milestone for artificial intelligence. The methods and techniques which conquered the best humans in chess had proven hopelessly impotent against go. One of the most shocking things is the speed with which AlphaGo seems to be getting better. The program that beat the European champion a few months ago was clearly much weaker than the one that faced Lee on Tuesday, and the one that beat him today seems much stronger than the that one.
The most important component of AlphaGo's artificial intelligence program is its deep neural networks, an old idea based on the brain given extraordinary power by new computer hardware. Far more than the computer algorithms that vanquished chess, the Deep neural net is generalizable to a vast host of tasks. If you have been paying attention, you might have noticed that the voice recognition software at your local pharmacy etc. has taken a gigantic leap in the last year or so. Siri and Amazon Echo are more examples. Almost certainly deep neural networks have done the work there as well.
Because the neural networks learn a lot like the way humans do, only a million times faster, many human intellectual tasks are likely to open up to them soon.
Be afraid. Be very afraid.