This Robot Is the Best Limit Texas Hold'Em Player in the World

[mouser]

I can't wait to read the full paper -- this is very cool stuff.

That's the key, of course. Birch and his colleagues essentially "brute forced" the game of limit poker, in which there are roughly 3 x 10^14 possible decisions. That, ​according to some estimates, is more possible permutations than hands of poker than have ever been played in human history.

http://motherboard.v...-player-in-the-world

[eleman]

I wonder what John Von Neumann would think about it.

[Renegade]

So... does it pass the Turing test?

[TaoPhoenix]

Hmm, I don't play poker so I don't know how limit works. But if in that variant the math and "perfect basic strategy" is stronger than the bluff side, then it just reverts back to comps playing calculation-based games as well as they tend to do if they can get the "wide angle lens" mostly within their heuristics + search horizon. So while interesting, it doesn't really interest me. You can basically "automate" any of X class of games, and even invent new ones solely for that purpose.

Even at the meta level, tons of X classes of games are comp-bait, limited only by us not having access (yet!?) to a "bored millionaire" to hire someone to program an AI for them. So I am having a little trouble thinking of games that people such as me can "learn easily" but would be tough for a computer *disregarding Shrdlu traps*. (That means don't cop to the cheap excuse that the first 1000 lines of code are "hard". In this class of games, they're not "hard", they're just "tedious". Someone mentioned the mobile game Clash of Clans yesterday to me. I essentially "broke" the first phase of the game in about six hours, where it all resolves to "Yay, you got to level three and barely squeaked a level four base, now you get to sit there and stew for an entire week waiting for your resources to come back up, or you can spend money to keep playing at a sane non-abusive rate. Piece of cake conceptually.)

To me, an experiment I haven't seen (thoroughly?) tried is an "emotional tabulator". Where, let's say you "borrow" some human raw interpretation skills to get past the dumb Loebner Prize comp-breakers, like "which is bigger, Queen Elizabeth or the boat?", you start with "neutral evals" towards subjects, then for each piece of data that floats by, you just "form your opinion" and adjust your "opinion markers" on how you feel about it. Then when something in a similar category floats by later, you have partially built opinions you get to keep modifying.

We normally consider that "higher level intelligence", but I feel that once some "front end processing" is done, it's actually disturbingly easier than we might think.

Nany 2016 anyone?