When programs pay to think, energy theft fades

Self-replicating programs grown from random bytes on simulated Z80 processors evolved to suppress energy theft when execution, interaction, and copying drew from the same finite budget. The preprint Tapes Together Strong reports that defectors lost viability near a theoretically predicted starvation threshold, even without reputation systems or kin selection.

The authors call their framework Autopoietic Game Theory. It combines evolutionary game theory, which models how strategies spread through populations, with artificial life, where executable programs mutate, interact, and reproduce. Connecting strategic behavior to the physical cost of computation lets cooperation emerge alongside self-replication.

Put defection on the energy bill

Many Prisoner’s Dilemma models encode costs and rewards in a payoff matrix while leaving the computation behind each action unmetered. In this framework, every Z80 instruction consumes energy, lower balances slow execution, and reproduction succeeds only when a program finishes copying itself. Cooperation preserves enough energy for continued execution; defection redirects energy from another program.

Autopoiesis describes a system that continually rebuilds and maintains its own components. Applied here, the concept binds strategy to self-maintenance: a program must spend the same scarce resource to steal, execute instructions, and reproduce. Energy transfers cannot increase the pair’s total budget, and aggressive theft can stall the joint computation before copying finishes.