I was originally confused why they published this result instead of heading to online casinos and becoming rich enough to buy small countries, but it seems that it’s a very simplified version of the game with only two players. More interesting, the strategy was reinforcement learning – the computer started with minimal domain knowledge, then played poker against itself a zillion times until it learned everything it needed to know. Everyone who thinks that AI is nothing to worry about, please think very carefully about the implications of a stupid non-generalized algorithm being able to auto-solve a game typically considered a supreme test of strategy and intellect. (Bold mine)
He also says in the comments
The problem is that in forty years, someone will be saying “Yes, conquering Europe used to be seen as a supreme test of strategy and intellect, but now that computers can do it the real test of original thinking is something less tractable. Like Go.”
Lets talk a little rational behavior. If poker is "solved" and you can't win anymore then the best thing to do is not play poker online anymore. Pretty straightforward right? How about for computers? Well if it is solved then playing online should mean you only play against other near perfect computers and you will both lose to the rake- which means computers also shouldn't play online unless they are for sure playing against people.
Lets extend it- if war in Europe is "solved" by a computer logically no human should ever start a war in Europe again. Peace in our time! Ok, so a computer could win a war in Europe, but with a computer on the other side it should be very wary of starting war since war carries an extremely high rake- way higher than an online casino. In fact it is so well established that trading > war as an economic fact that turning war over to computers that had figured it out should lead to peace in our time!
Honestly this analysis is pretty weak sauce since winning and losing are defined by the rules of the game. If you define winning as "no Jews left in Europe" then war would go from no chance to 100% chance really quickly, so you have a framing question still to worry about. Still the optimist can present an incredibly strong case- once computers have solved all games and are pitted against one another the only games that they would choose to participate in are positive sum games.... you know, like trade and technological advancement! Given a choice between a game neither can win or a game you both can win the only immediately obvious reasons to pick the former are things like xenophobia or ego.
I, for one, welcome our new robot overlords!
hail ants
ReplyDelete