Θ-Learning: An Algorithm for the Self-Organisation of Collective Self-Governance

We investigate a formal characterisation of thorybos, a governance mechanism used in classical Athenian deliberative assemblies to help reach consensus through democratic deliberation. The process of thorybos encouraged Athenian citizens to express vocally their positive or negative reactions to policy proposals to judge whether or not consensus had been achieved. We model this process through an algorithm called Θ-learning, which has two phases: an expression phase where agents ‘vocalise’ their preference for specific policies, and a reflective phase where the agents learn how to reach an agreement, over time, while minimising compromises. We describe experiments which show the different outcomes of Θ-learning under different conditions (e.g. variation in objectives and rate of change of population). It is concluded that Θ-learning contributes to ‘good’ governance of self-organising systems by extracting a productive signal out of what might otherwise be regarded as distracting ‘noise’, and by creating a sustainable consensus in the form of a persistent general agreement derived from democratic participation.