Tapes Together Strong: The Co-evolution of Computation and Cooperation
This presentation explores a radical computational framework called Autopoietic Game Theory, in which cooperation emerges not from memory or spatial clustering alone, but from the physical coupling of social behavior, energy expenditure, and self-replication. Using populations of evolved Z80 machine-code programs that share memory and compete for finite energy, the research demonstrates that lossy stealing can become self-limiting when it degrades the computational substrate required for reproduction. The talk reveals how evolvable replication timing decouples immediate exploitation from reproductive success, allowing cooperative strategies to dominate even in well-mixed populations, and shows how spatial structure acts as an architectural scaffold that preserves complexity and enables collaborative problem-solving under environmental stress.Script
When self-replicating programs share a single computational tape and compete for the energy needed to execute their next instruction, something unexpected happens. Stealing energy gives you an immediate advantage, but it can starve the very system you need to copy yourself into the future.
The substrate is deceptively simple. Each of 16,384 programs is a 32-byte sequence of modified Z80 machine code. When two programs are paired, they share a cyclic memory tape where each can read and overwrite the other's code, but energy determines whether you get to execute your next instruction at all. Replication isn't a button you press; it's a computational achievement you must write into your partner's memory segment before time runs out.
Here's the metabolic starvation limit in action. Theory predicts that when you steal too aggressively, you drain the pair's energy below what's needed to complete replication before execution halts. Empirical Z80 runs confirm it: defectors who steal 32 units instead of 16 see their survival collapse with an effect size over 4, because they literally run out of computational fuel before they can write a single viable offspring.
But energy loss alone doesn't explain cooperation; timing does. In a simplified model where replication timing can evolve, three event orderings produce three opposite outcomes. If you replicate based on post-interaction energy, defectors win. If replication is decided before payoffs update, or if replication happens before the social interaction entirely, cooperation takes over. The same payoffs, the same energy drain, completely reversed outcomes.
Spatial structure doesn't just help cooperation survive; it becomes the scaffold that prevents architectural collapse. Under severe energy inequality distributed across space, well-mixed populations suffer metabolic and structural failure. Local interactions let successful lineages stay near high-energy regions long enough to build the computational complexity required for robust replication, with effect sizes approaching 9 for both energy preservation and structural integrity.
Cooperation in this system isn't about memory, reputation, or repeated games. It emerges because social behavior, computation, and reproduction are physically inseparable, and because the replication mechanism itself evolves. When lossy stealing degrades the substrate faster than it accelerates your own execution, and when evolutionary time allows replication to decouple from immediate payoff, cooperation becomes not just viable but favored. To explore how tightly coupled substrates reshape evolutionary outcomes, visit EmergentMind.com and turn research like this into your own video.
