Abstract

We study motile agents that both sense, and then seek future control over, their environment. They do this by a process that we call future state maximisation (FSM): they each choose a move in the present that leads to the largest variety of states that they can realise in (their model of) the future. I will first discuss how this may relate to intelligent behaviour more generally. We then focus on groups of moving, re-orientable agents that individually maximise the variety of a simple visual encoding of the environment accessible to them within some time horizon. Collective motion emerges that exhibits cohesion, co-alignment and collision suppression; none of which are explicitly encoded in the model. Crudely put: maximisation of the “entropy” of future states naturally leads to a low entropy (cohesive, co-aligned) state in the present; time reversal symmetry is only broken weakly, if at all. I show how heuristics mimicking FSM could operate in real time under animal cognition. These offer a philosophically attractive, bottom-up mechanism for the emergence of swarming behaviour in animals. Perhaps, one day, we might also see them encoded into artificial “intelligent” matter, able to sense light, process information and move.