Predictability, Risk and Online Management in a Complex System of Adaptive Agents

Abstract

We discuss the feasibility of predicting, managing and subsequently manipulating, the future evolution of a Complex Adaptive System. Our archetypal system mimics a population of adaptive, interacting objects, such as those arising in the domains of human health and biology (e.g. cells), financial markets (e.g. traders), and mechanical systems (e.g. robots). We show that short-term prediction yields corridors along which the model system will, with high probability, evolve. We show how the widths and average direction of these corridors varies in time as the system passes through regions, or pockets, of enhanced predictability and/or risk. We then show how small amounts of 'population engineering' can be undertaken in order to steer the system away from any undesired regimes which have been predicted. Despite the system's many degrees of freedom and inherent stochasticity, this dynamical, 'soft' control over future risk requires only minimal knowledge about the underlying composition of the constituent multi-agent population.