Authors Francesco Cicala, Blake Richards, and colleagues investigate the co-evolution of self-replication and problem-solving in a digital primordial soup. They initialize random 32-byte Z80 assembly programs where self-replication must emerge through random mutations and pairwise interactions. A task-based validation step links behaviors: correctly evaluating a polynomial raises interaction probability.

Experiments yield four primary findings. First, self-replication and mathematical problem-solving co-evolve from randomness. Second, computational pressure accelerates compact, robust reproductive architectures preserving memory for tasks. Third, metabolic constraints increase conditional halting—programs terminate early during validation but bypass halt during interaction for block-copy replication. Fourth, spatial task niches create an emergent learning curriculum, using simple solutions as stepping stones toward complex polynomials.

Results demonstrate an interactive feedback loop: environmental task demands shape self-replication architecture, while spontaneous replication alters functional problem-solving trajectories.