One arc, told in simulations

Method — everything is computed live

Every model on this site runs in your browser, from first principles, the moment you open the page. The math lives in lib/lab/models as pure, framework-free functions — integrators, agents, update rules — each covered by unit tests, with a thin canvas renderer on top. Nothing is pre-rendered, with exactly two exceptions: C. elegans locomotion (the OpenWorm c302 + Sibernetic pipeline needs NEURON and a physics engine) and an IQN agent playing Pong (deep RL training is far too heavy for a static page). Both are recorded runs and say so on their cards.

The four arcs

The collection is weighted by interest, not ROI, and reads as one through-line: how order and intelligence arise at every scale. The connective pieces — a Hopfield net is memory and attractor dynamics; an agent-based market is finance and emergence — are what make it one body of work.

1. Arc I — The Math of EmergenceThe substrate: nonlinear dynamics, chaos, complexity. Strange attractors, reaction–diffusion, sandpile avalanches, the road to chaos.
2. Arc II — MindsIntelligence in brains: neuroscience, cognition, learning. Hodgkin–Huxley ion channels, Hopfield memory, dopamine as TD error, Kuramoto synchrony.
3. Arc III — AgentsIntelligence engineered and collective: AI, reinforcement learning, multi-agent systems. Live RL training, attention visualized, gradient descent made spatial, swarm algorithms.
4. Arc IV — Societies & MarketsIntelligence at population scale: game theory, economics, social choice. Agent-based markets that bubble and crash, the evolution of cooperation, segregation, inequality, voting paradoxes.

The Python ABM lab

Three of the agent-based models — boids, SIR epidemic and Schelling segregation — have a companion research environment: a Mesa-based ABM lab where parameter sweeps are tuned with Optuna. The tuned configurations feed back into this site as the preset buttons above the sliders and the batch-result charts on those pages. The code is open at emergence-lab/python/abm-sim.

Credits & license

The recorded C. elegans run is thanks to the OpenWorm project (c302 + Sibernetic). The Pong agent was trained with deep_rl_zoo (IQN). The lab's simulation specs were distilled from the SimulateItNow catalog — structural ideas only; all prose and code here are original. The site itself is MIT licensed.