Computation, Chess, and Language in Artificial Intelligence

Researcher Bill Benzon has released a provocative working paper titled 'Computation, Chess, and Language in Artificial Intelligence' that challenges foundational assumptions in AI development. The paper argues that artificial intelligence has been historically misdirected by treating chess as its model organism—what John McCarthy called 'the Drosophila of AI'—when natural language represents a fundamentally different cognitive domain. Benzon traces how chess, with its bounded rules and geometric clarity, became AI's early benchmark for demonstrating reasoning capability, while natural language processing emerged separately through computational linguistics. This historical divergence, he contends, has left AI implicitly assuming that intelligence principles reduce to computation principles, overlooking the embodied, resource-constrained reality of true cognition.

Benzon draws on Miriam Yevick's distinction between symbolic and neural computational regimes to propose that intelligence must be understood as operating in geometrically complex worlds under finite constraints. The paper identifies three principles of intelligence that aren't principles of computation, emphasizing that embodiment represents a formal condition rather than incidental feature. By examining the structural differences between bounded games and open-ended cognition, Benzon clarifies both AI's historical trajectory and conceptual limits of current systems like GPT-4 and Claude. The work suggests that advancing beyond current AI capabilities requires moving past the chess paradigm to address language's ill-defined, unbounded nature where objectives are diffuse and domains inseparable from experience.