Evolutionary food race: a tiny artificial-life model

Artificial life is easiest to understand when you can watch it move.

This small model starts with simple local rules: agents wander, sense nearby food, spend energy to move, eat to survive, reproduce when they have enough energy, and pass slightly mutated traits to their children.

Open the interactive model in a full page.

This was made by Sakana AI through Hermes as a small demo for exploring artificial life with the Thimphu Tech Meet community.

What evolves

Each agent carries a tiny genome:

• Speed — how quickly it can reach food.
• Sensor range — how far it can detect food patches.
• Turn rate — how sharply it can steer.
• Fertility — how easily it converts stored energy into offspring.
• Caution/metabolism — how much energy its movement strategy costs.

The model does not tell agents how to form groups, balance energy, or create population cycles. Those patterns emerge from competition for food.

What to watch for

• Clustering: agents gather around dense food patches, then scatter when those patches are stripped.
• Selection pressure: average speed usually rises early because faster agents reach food first.
• Trade-offs: speed does not rise forever because movement has an energy cost.
• Boom/bust cycles: the population grows, consumes too much food, dips, then recovers as food regrows.
• Diversity: several viable strategies can survive at the same time instead of one perfect genome taking over.

Why this matters

The useful lesson is not that this is a realistic ecosystem. It is that simple rules can create behavior that feels larger than the rules themselves.

That is the core artificial-life idea: no central planner, no scripted choreography, just local agents interacting until structure appears.