Self-Governing Multi-Agent Systems Course

This lecture presents motivating examples of collective action and cooperative survival, from actual examples such ad hoc networks and SmartGrids to experimental testbeds, presents the foundations of rule-based self-organisation in norm-governed multi-agent systems. It introduces key concepts like institutions, institutionalised power, and social construction, and also defines a methodology for formalising theories of social science in computational logic.

---

This lecture shows how Nicholas Rescher's philosophical theory of distributive justice, based on the idea of legitimate claims, can be formalised as an executable specification. A group of autonomous agents can use this specification in a self-governing electronic institution, to resolve the problem of 'fairness' in resource allocation. Critically, the agents can introspect of the outcome of the allocation and vote to change the priorities assigned to the agreed legitimate claims.

This lecture shows how political scientist Josiah Ober's insights from classical Athenian democracy, in particular its knowledge management processes which enabled Athens to out-perform other city states, can be used to formalise a process of interactional justice. This is concerned with mapping each agent's individual subjective perception of its treatment by an institution into a collective objective evaluation, as an input to rule re-organisation or reformation.

This lecture shows how key principles of Andrzej Nowak et. al's psychological theory of social influence (RTSI: the Regulatory Theory of Social Influence) can be implemented over a multi-agent system's social network. In the context of distributed information processing, this allows for the emergence of expertise to provide cognitive efficiency as well as more accurate truth-tracking, and balance out four pairs of conflicting systemic drivers of rule adaptation.

This lecture shows how ideas from cybernetics and psycho-acoustics can be combined in a multi-agent system which has both a regulator and a regulated unit, and achieve the requisite pathways of attention and expression between the two components. In this hybrid system of learning and reasoning, the agents in the regulated unit reason between different `voices', while the regulator uses a standard Q-learning algorithm to pay attention to the regulated units and respond accordingly. 

This lecture shows how the process of thorybos used in classical Athenian democracy, as described by political scientist Mirko Canevaro, can strike a balance between consensus formation and majority decision-making. This process is formalised in an extended Q-learning algorithm called Theta-Learning. It validates majority decisions if the underlying principle is consensus by ensuring that dissenting voices and compromises are used as conceptual resources in subsequent decisions.

This lecture addresses the problem of dealing with 'unknown unknowns', for which all existing social arrangements are unsuitable. This presents two challenges for self-governance: firstly, where do new social arrangements come from, with respect to the cybernetics law of requisite complexity? and secondly, how do agents imagine new social arrangements to be different, and preferable, to current arrangements, without ever having experienced them? The proposed solution lies in innovation, including knowledge assimilation (from new agents), knowledge transfer  (from other communities), and hybrid approaches involving ML agents and human users.

This lecture concludes the course by extending the discussion from cyber-physical systems to socio-technical systems, consisting of interacting human and computational intelligence. Two of the subjects covered in this lecture are socially-guided machine learning, where human and ML-agent work together in defining policy or social arrangements, and ethical platformisation, where intelligent agents provide services to people and communities while respecting their values.