And this, even more than making moves instantly, is how we know AlphaGo is more than a little bit better than humans. We are now entirely outclassed, not even able to see the point behind its decisions.
Master made moves that seemed foolish but inevitably led to victory
Something makes me uncomfortable about computers working in ways humans can't understand.
Lack of control is a very real fear when talking about powerful tools. Fire needs to be checked, lest we are all burned.
I agree wholeheartedly. What stunned me so much about AlphaGo back in March 2016 wasn't that it beat a human champion— that was a Stun Level of Over 9,000. It was that it beat a human using moves never before seen in the history of the game. That's a Stun Level of Over One Million.
And it's reflected here. It's just as Raklian said back then— we might as well be dealing with an alien. These wonderful people, these champions of an ancient game, are not merely experiencing the shock of being beaten by a computer, but are actually playing an ancient game against what amounts to the alien intelligence who created the game in the first place.
And that's one year. One year of improvement over where it was in October 2015, when it beat the European champion. This machine is so unbelievably powerful that it's already beyond us. The greatest Go players of all time can't understand it, and even when they study its moves, they lack the brainpower to know when to use them. Sure, its area of superintelligence is narrow— AlphaGo can only play Go and would need to be extensively retrained to do anything else— but this should come as a taste of what we should expect in the coming years and decades. AlphaGo hasn't even reached the peak of its potential, as I said above. It's probably nowhere close to the peak. But despite that, it's still superior to all humans, and on a meaningful level.
The best analogy really is that of a fire that's growing faster than we know how to deal with. So fast, in fact, that the best word for it would be "explosion."