Autopoietic Game Theory links cooperation to compute cost

Coupling computational capacity to a shared energy budget can make cooperation self-organize, according to a preprint titled "Tapes Together Strong: The Co-evolution of Computation and Cooperation". The authors — Kunal Jha, Francesco Cicala, Blaise Agüera y Arcas, Blake Aaron Richards, Natasha Jaques, Max Kleiman-Weiner and Eyvind Niklasson — call the framework Autopoietic Game Theory, integrating "social interactions, replication mechanisms, and their associated computational costs" into one co-evolving model.

The demonstrations run on Z80 machine code programs. When resources are limited, the paper argues that parasitic behavior becomes counterproductive because "stealing destroys shared energy, slows execution, and can prevent reliable replication." Evolved programs suppress stealing across multiple Z80 environments, and spatial arrangement adds structural complexity. Two researchers we track posted the link this week.

The framework also absorbs external pressures like math tasks structured as sequential dilemmas when tied to energy budgets. The abstract publishes no head-to-head baselines against non-coupled evolutionary systems, and the transfer story to modern neural agents is left to future work.