Exploring the evolution of population-level complexity in a visually interpretable artificial life world based on plasmodial slime moulds (Myxogastria)
Learn More
Exploring the evolution of population-level complexity in a visually interpretable artificial life world based on plasmodial slime moulds (Myxogastria)
Learn MoreUnderstanding how collective intelligence emerges from simple individual interactions is a crucial challenge in artificial life research.
We investigate the role of interdependence between replicating agents in promoting the evolution of population-level agency in Darwinian-like artificial life.
Given the vast timescales involved, evolution by natural selection is notoriously difficult to visualise and therefore to teach—particularly to visual learners.
Our software provides an interactive and visually interpretable demonstration of evolution, bringing Darwinism directly to every student's fingertips.
The Myxomatrix is a visually interpretable artificial life simulation inspired by plasmodial slime moulds. The system achieves consistent population-level complexity evolution without need of a predefined 'goal' state that would typically be encoded via a fitness function or convoluted initial conditions. The Myxomatrix autonomously evolves collective behaviours that are genuinely novel with respect to the initial agents.
The world consists of a grid of cells, each of which may contain an agent and a plant. Agents must search for and eat the plants to survive, but cannot move from their cell, only replicate and transfer mass to neighbouring agents. As such, agents must cooperate: replicating and transferring tactfully to traverse the environment as a collective.
Our novel Interdependence Condition ensures exploration of collective behaviours rather than individualistic strategies.
Populations develop perception-decision-action loops that exist only at the group level. These "superagents" exhibit novel capabilities inaccessible to individual agents.
Unlike abstract simulations, the Myxomatrix allows direct observation of evolutionary progression from Blobs to Crawlers to Sweepers and ultimately to Switchers, making evolution tangible.
Advanced populations exhibit altruistic mass transfer between lineages, prioritising group survival over individual genetic advantage.
Indefinite NeuroEvolution of Augmenting Spike-Timing-Dependent Plastic Topologies provides a biologically realistic approach to spiking neural network evolution.
The IC framework may be applicable to evolving rule-sets for multi-agent distributed computation and intelligent network architectures.
Comparative runs confirm that only IC-compliant populations consistently climb the entire evolutionary ladder. Non-compliant runs where individualistic strategies take hold tend to plateau without developing population-level behaviours.
Read MoreBradley Luke Smith, Andrew Barnes
We introduce the Interdependence Condition (IC) and demonstrate its effectiveness in scaffolding the evolution of population-level complexity through the novel Myxomatrix simulation. Our results show reliable emergence of superagency in IC-compliant systems.