Beyond Adversarial: The Case for Game AI as Storytelling

The paper argues that game AI should move beyond the single goal of winning and instead aim to craft satisfying player experiences. Its central thesis is that a game's trajectory, the sequence of states that unfolds during play, can be treated as a narrative that is evaluated, shaped, and optimized. Traditional techniques such as minimax search, Markov decision processes, and state-space search are reviewed alongside storytelling approaches including simulation-based and deliberative narrative generation, interactive narrative, and drama management. The authors contend that reasoning about trajectory structure, rather than only win or loss outcomes, offers a more engaging path for game AI.

A key idea is that an AI may deliberately make "intelligent mistakes" to shape how a player experiences the game. To support this, the authors propose narrative heuristics, including thought flow, plot homing, dramatic arc, and suspense, as criteria for judging trajectories. They use chess to illustrate the approach, citing the Falkbeer Counter-Gambit, in which a deliberate material concession creates a later advantage, and poker, where a dealing AI could control the reveal of cards to produce narrative features. The poker example carries the caveat that such manipulation must avoid appearing to be cheating.

The analysis also acknowledges significant technical challenges. Searching trajectory space is computationally harder than searching state space, and scaling requires robust heuristic functions. Hidden information, as in poker, further complicates design, since the AI must shape narrative outcomes without compromising the fairness players expect. The authors treat their case studies as illustrative rather than exhaustive.

Ultimately, the paper concludes that framing gameplay as narrative allows designers to apply AI storytelling methods to genres not traditionally considered narrative. Its contribution lies less in a finished system than in a conceptual reorientation of game AI toward the quality of the experience it produces.