Macro-Programming Multi-Agent Systems: A Framework for Artificial Collective Intelligence

This paper elaborates on the opportunity and idea of multi-agent system (MAS) macro-programming, i.e., programming in terms of macroscopic denotations of system goals, structure, or behaviour. Indeed, macro-descriptions with explicit macro-to-micro mapping can be a formidable way to harness the complexity of emergent collective behaviour (for humans) and to provide a structure for guiding optimisation, learning, and generative artificial intelligence (AI) processes (for computers). Despite contributions about meso-level (e.g., organisational), multi-level (e.g., holonic) and declarative (e.g., goal-oriented, normative) paradigms exist in the MAS literature, research is fragmented and the topic arguably overlooked by both the scientific and software engineering viewpoints. Recent survey works on macro-programming spanning areas from sensor networks to swarm robotics suggest that a synthesis is possible, despite the variety of methods, techniques, and abstractions. With reference to early and recent literature both within and outside the MAS community, this paper motivates that multi-scale and especially macroscopic descriptions are possible, useful, and timely—opening up to research opportunities and community debate.