Blackjack-Playing Agents in an Advanced Al Course

摘要

Blackjack is a multi-player card game in which each player makes a sequence of decisions based on a partially-observed game state that evolves under uncertainty. In its casino variant, Blackjack is typically played as a set of separate contests, each involving a player and the dealer, whose decisions are determined according to a published, fixed policy. We have developed and used a sequence of assignments for an advanced artificial intelligence course in which students determine an optimal strategy for Blackjack, modeled as a Markov Decision Process. The course culminates in a Blackjack tournament in which agents employ policies students have determined through their analyses.