Intelligence, a recipe
Matthew Turner (Warwick University, UK)
We study information-processing (artificial), or “intelligent” (living), agents. These agents seek maximal control of their environment via future state maximisation (FSM), a principle that we argue may relate to intelligent behaviour more generally. Here we study moving, re-oreintable agents that seek to maximise their space of accessible (visual) environments, out to some time horizon. The action of each agent is (re)established by exhaustive enumeration of its future decision tree at each time step – each agent chooses the branch of its tree leading from the present to the richest future state space. Remarkably, cohesive swarm-like motion emerges that is similar to that observed in animal systems, such as bird flocks. We develop heuristics that mimic computationally intensive FSM but that could also operate in real time under animal cognition. Finally, we show that iterative application of these heuristics as the model for the behaviour of others, when determining the dynamics of self under full FSM, can lead to a form of closure for the problem. I will argue that this offers a philosophically attractive, bottom-up mechanism for the emergence of swarming.