Cultural Evolution

Cultural evolution names the cumulative, transmissible change in the shared knowledge, practices, and artifacts of a population across generations. Originally developed in anthropology and evolutionary biology (Boyd & Richerson 1985; Cavalli-Sforza & Feldman 1981), the theory treats culture as a second inheritance system alongside genes: ideas, techniques, and tools vary between individuals, are transmitted between them through social learning, and undergo selection-like processes as bearers adopt or discard them. Cultural evolution explains the ratchet-like accumulation of complexity seen in human technology — each generation inherits, modifies, and passes on artifacts that no individual could have invented alone.

In AI systems, cultural evolution describes the analogous process in populations of agents. Cultural evolution in populations of Large Language Models (Baranov et al., 2024) demonstrates that LLM agents playing iterated games develop and transmit game-theoretic strategies across populations. In the SwarmWorld benchmark (Pal, Wang, Buehler, 2026), hundreds of LLM agents accumulate executable artifacts, controllers, and physical structures across simulated generations through collaborative construction and executable inheritance. Earlier work in cultural transmission learning (2022) and computational models of cumulative culture (2016) shows the same dynamic outside LLMs.

The mechanism requires social learning, faithful transmission, and incremental modification. Failure modes include high-fidelity copying without variation (no accumulation), low-fidelity copying with no selection (no retention), and cultural transmission bottlenecks that erase progress. Cultural evolution is adjacent to evolutionary-algorithm and neuroevolution (which use genetic, not cultural, inheritance) and to self-evolution (which modifies a single model's scaffold, not a population's shared culture).