Complex Roads to Evolvable Agents: Computational Evolution, Development and Ecology of Artificial Agents

Roughly 540 million years ago, in the Cambrian explosion, nearly every major animal body plan appeared in a geological instant; a single cichlid lineage fills a lake with hundreds of species while its neighbours stand still; a foraging trick spreads through a population in a single season. Evolutionary novelty on this scale is not explained by selection alone: the space of possible variants is itself structured, biased towards some forms and away from others. The Extended Evolutionary Synthesis locates this structure in development, the biased and evolvable process standing between what is inherited and what selection sees, and holds it to be not a passive relay but a cause of evolution in its own right, coupled to ecology and behaviour in loops of reciprocal causation. This thesis aligns with that view and builds on it, asking how development, ecology and behaviour together generate, or obstruct, evolutionary complexity.
To study these couplings this thesis builds artificial worlds, in the constructive tradition of Artificial Life: worlds tractable enough to manipulate and replay, yet rich enough to exhibit the complexity at stake. Five computational systems climb the levels of biological organisation, from proto-life to society. Flow-Lenia, a mass-conserving continuous cellular automaton, shows that a single physical constraint bootstraps life-like creatures and self-organising ecosystems in which evolutionary dynamics spontaneously arise and can be measured. From this medium, where the individual is not even defined, the thesis turns to established individuals and to their most powerful developmental engine, the nervous system, the richest generator of adaptive behaviour evolution has produced. Lifelong Neural Developmental Programs unify pre-experience morphogenesis and lifelong plasticity in one process, and show that structural plasticity improves adaptation under change; where such programmes fail for want of ecological richness, the chapters that follow build that richness back. A model of sympatric adaptive radiation demonstrates that ecological competition alone can drive a radiation, and that development is causative: biases channel variation, evolvability is selected under ecological opportunity, and variation and emergent ecological order reciprocally shape each other. Eco-Evo-Devox goes further: motivated by the changing information landscapes of a Cambrian-like radiation, it is a modular Artificial Life platform centred on a gene-regulation-inspired developmental model that grows a brain within a body, making ecology, development and embodiment jointly manipulable for the study of early neural evolution. Already, behaviour is shown reshaping the environment into new selective pressures that drive further adaptation. Finally, the same logic extends beyond genes to culture, taking a side in a long-standing tension between externalist accounts of transmission as replication and internalist ones emphasising reconstruction and generation through individuals’ own biases. A model of cultural evolution operationalises the latter perspective in fully generative form, agents themselves cast as biased developmental causes of change, and shows that their biases generate both increased evolvability and biased drift.
Across these systems one claim recurs: development is a cause of evolutionary pattern in its own right, its biases setting which forms lie within reach and so shaping what evolution produces. Together they motivate a broader conjecture: that coupling an evolvable developmental organisation to a rich, self-constructed ecology, each remaking the other, is a road towards open-endedness, a horizon these models project rather than reach.