Evaluate performance of NLTH AIs under full multi-player engagement

Determine the performance of no-limit Texas hold'em AI agents, including DeepStack, Libratus, and Pluribus, in multi-player games where all players participate in every betting round rather than folding early, thereby assessing their effectiveness in settings with sustained multi-player engagement.

Background

Recent AI systems such as DeepStack, Libratus, and Pluribus have achieved strong results in no-limit Texas hold'em (NLTH), including multi-player settings. However, typical NLTH dynamics often see most players folding after the opening round, effectively reducing many hands to two-player scenarios. This raises concerns about how well these AIs perform when the game maintains robust multi-player engagement throughout all betting rounds.

The authors emphasize that Liar's Poker intrinsically enforces continuous multi-player participation across bidding rounds, motivating its use as a testbed. Nonetheless, the specific question of how NLTH AIs fare when every player remains engaged across rounds is explicitly identified as unresolved in the introduction.