A swarm of 200 particles that constantly push against one another and align their velocities is learning to predict chaos. A single driver particle receives an external signal resembling the unpredictable behavior of weather, and the rest have to guess where it will move next. Such a swarm is a physical computer, and its ability to predict depends on how it handles heat.
It turned out that what matters is not how much heat the swarm produces on its own, but how strongly the external signal changes that flow. When the driver is connected, the heat flow increases thousands of times over. It is this increase, and not the heat itself, that is linked to prediction accuracy. The best result is achieved in a regime where, without the signal, almost no heat is released, while with the signal it rises sharply. Such a swarm predicts chaos with a correlation of 0.97, that is, almost perfectly.
This discovery helps quickly find suitable regimes for physical computers without wasting time on lengthy training. A swarm of particles that once seemed like just a chaotic jumble turns out to be a delicate instrument: its entropy works like a compass pointing to computational capabilities.
The reservoir's maximum performance is achieved in a regime where heat flow into the environment is minimal in the absence of a signal, but increases thousands of times when the signal is applied.
The idea of using a swarm of particles for computation echoes the concept of 'computing matter' from Stanisław Lem's novel The Invincible, where a collective of microscopic machines processes information.