This post is also available in Dutch .
Many complex social behaviours that we see in groups of animals and humans can be traced back to very simple rules running inside individual brains. The twist is that brains almost never act alone. They act in flocks and crowds and markets, which evolve the simple rules into a complex phenomenon. One such complex social behaviour is called the “asset-protection” principle, and one such simple brain rule is called the “urgency-gating” mechanism. This blog is about how those two come together in an artificial swarm.
Asset protection among hungry agents
The asset protection principle says something very intuitive. When you are well fed, you become more careful. When you are close to starvation, you are willing to gamble. A full bird can afford to leave a risky patch. A hungry bird cannot. The same logic quietly shapes traders in financial markets and fishermen when they decide on a location to forage for fish.
In social foraging models (common-pool resources), this shows up as a tension between safety and variance. Well-provisioned foragers avoid risky patches and sometimes even avoid each other. Starved agents accept risk and tolerate crowding if that increases the chance of survival.
So somewhere in the brain, there must be a computation that tracks how much you have to lose and shifts your decision strategy accordingly.
Urgency inside the brain
One way to implement this is through an urgency signal. Certain neurons increase their firing rate when urgency becomes high and decrease it when there is time to spare. The signal is not about a single option. It is more like a running estimate of how bad it would be to keep delaying the decision.
In decision neuroscience, this appears as an evidence accumulation or urgency gating mechanism. Activity ramps up as evidence and urgency accumulate, and once it crosses a threshold, the system commits. Clamp experiments in animals show neurons whose activity locks onto these internal states rather than to any specific stimulus or movement.
It can be speculated that maybe if the internal energy level is high, the urgency signal rises slowly, and the agent behaves in a risk-averse way. If the tank is nearly empty, the urgency ramps faster and pushes the agent into risk-seeking moves.
Teaching digital foragers the same trick
In our recent work, we asked whether a similar computation could emerge in artificial swarms that learn only from the pressure to survive in a shared environment. We built a patch foraging world with many foragers moving around with the aim of collecting maximum resources. The foragers could see each other and the resources only in their field of view. They could also sense their internal energy levels.
When we let this system evolve, something like asset protection appeared on its own. When internal energy was high, the swarm spread out. Nearest neighbour distances grew, and agents behaved almost indifferent to one another. When energy dropped, aggregation kicked in. The agents pulled together into dense swarms, effectively using each other as cues in a partially observable world.
Inside the brains of the agents that were modelled by taking inspiration from neuroscience, we saw that some artificial neurons actually learned to track the internal energy levels of the agent. When these neurons were clamped to lower energy levels, i.e. when the agents were tricked into sensing that they were hungry, the agents aggregated sooner. This supported the speculation that an urgency-gating mechanism in one agent can indeed give rise to a complex social phenomenon like asset-protection when scaled to many agents.
Conclusion
By simulating these mechanisms in artificial worlds, we get a laboratory where we can watch simple neural dynamics scale up into social behaviour. These models may not prove the actual mechanistic bridges between the fields, but can help generate a hypothesis in a more economically scalable and animal-friendly way.
Author: Siddharth Chaturvedi
Buddy: Vivek Sharma
Editor: Xuanwei Li
Translation: Natalie Nielsen
Editor Translation: Wieger Scheurer
About The Author
