AI
How AI and Institutional Design Can Learn from Each Other
Researchers argue that large-scale AI systems and human institutions face parallel challenges—and that studying both together can improve how we build intelligent systems and run organizations.
Artificial intelligence began as an effort to replicate human cognition, but the scale of that goal quickly proved unmanageable. The field pivoted instead toward studying narrower competencies and developing frameworks for coordinating multiple specialized systems—treating them as agents that operate within organizational hierarchies.
The Multi-Agent Turn#
The agent-based modelling approach found surprising utility well beyond AI research, becoming a tool for understanding and simulating the behavior of social, economic, and legal systems. Yet contemporary AI development has largely abandoned the conceptual depth developed in multi-agent systems and normative systems research, focusing instead on narrow capabilities like text completion.
Bridging AI and Institutions#
A recent research contribution examines what large-scale distributed AI and institutional structures have in common. Three key insights emerge:
-
Historical perspective: Tracing AI's evolution reveals an "ecological" view of intelligence—one that questions whether compression (the ability to represent complex information compactly) truly constitutes intelligence.
-
Separating concerns: Understanding how internal mechanisms create external agency helps organize both individual and collective behavior, creating clearer boundaries for different kinds of coordination problems.
-
Pluralism as foundation: For intelligence to remain dynamic and adaptive, it must embrace epistemological, normative, and value pluralisms—meaning no single way of knowing, acting, or judging should dominate completely.
Applying AI to Institutional Practice#
The research identifies concrete ways AI techniques can strengthen institutional functioning: modelling and clarifying social roles, simulating how normative interactions play out, identifying when institutional mechanisms fail, and supporting processes for examining and improving institutional rules and procedures.
The work bridges a gap between AI research and organizational science, suggesting that designing better intelligent systems and running better institutions may ultimately involve solving similar problems.
