4

There is a new poker computer that is claimed to be unbeatable.

http://www.theguardian.com/science/2015/jan/08/poker-program-cepheus-unbeatable

I beat this computer on my first try today but my friend says that my 100 rounds against the computer was too small of a sample size to prove that the computer's strategy is not optimal. How big should my sample size be?

Edit: This is heads up texas hold'em poker (1 vs 1)

John
  • 41
  • 2
    Well, no algorithm for playing a game with a substantial luck element can be truly unbeatable. – KSmarts Jan 20 '15 at 21:19
  • Still, I want to further disprove that its strategy is "optimal". – John Jan 20 '15 at 21:24
  • I'm no poker expert, but I find the whole "perfect poker program" claim to be kind of weird. Not because of the luck element (which averages out on the long run) but simply because poker is a psychological game. How can a computer even play actual poker? – Guri Harari Jan 20 '15 at 21:31
  • @GuriHarari: After all, it is also a combinatorial and probabilistic game. – String Jan 20 '15 at 21:53
  • I suppose you also read this link about the limitations of Cepheus Maybe the model/learning algorithm used for Cepheus can be improved to make it adapt better to different playing styles. Perhaps each player should login to identify themselves too ... – String Jan 20 '15 at 22:03
  • Also, strictly speaking, you can't prove it isn't optimal by testing, since you could just be really lucky. You could, however, give a confidence interval for the program's success rate. – KSmarts Jan 21 '15 at 22:59

0 Answers0