The first ever human versus computer no-limit Texas hold ’em competition took place from April 24–May 8, 2015, at Rivers Casino in Pittsburgh, PA, organized by
Carnegie Mellon University Professor Tuomas Sandholm.
Twenty thousand hands of two-player no-limit Texas hold
’em were played between the computer program Claudico
and four of the top human specialists in this variation of poker, Dong Kim, Jason Les, Bjorn Li, and Doug Polk (so 80,000
hands were played in total).
While this was the first human versus machine competition for the no-limit variant of Texas hold ’em, there had
been two prior competitions for the limit variant. Two-player limit Texas hold ’em is the smallest poker variant played
competitively by humans. A breakthrough was achieved last
year as the game was “essentially weakly solved” (an ε-Nash
equilibrium was computed for such small ε to be statistically
indistinguishable from zero in a human lifetime of play) by
researchers at the University of Alberta (Bowling et al. 2015).
In the limit variant all bets are of a fixed size, while in no-limit bets can be of any number of chips up to the amount
Reflections on the First Man
Versus Machine No-Limit Texas
Hold ’em Competition
n The first human versus computer no-limit Texas hold ’em competition took
place from April 24–May 8, 2015, at
Rivers Casino in Pittsburgh, PA. In this
article I present my thoughts on the
competition design, agent architecture,
and lessons learned. Several problematic hands from the competition are highlighted that reveal the most glaring
weaknesses of the agent. The research
underlying the agent is placed within a
broader context in the AI research community, and several avenues for future
study are mapped out.