This is Part 4 of the seven-part Logic and Reasoning field guide. Part III (A Multi-Agent Problem-Solver) treated the team of agents as an instantaneous artifact — agents, protocols, coordination patterns. Part IV puts the team in time. What changes when the team has a history. When it has institutional memory. When it has emergent norms. When, in short, it has the beginnings of a culture.

The bridge from architecture to culture is also the bridge from computer science to social science. Game theory shows up here because cooperation is not a given — even when cooperation would benefit everyone, the equilibrium can be defection. The prisoner’s dilemma, the stag hunt, the public goods game — these formal frameworks tell us when cooperation is the equilibrium and when it isn’t, and what mechanisms (reputation, repetition, punishment, communication) can shift the equilibrium toward cooperation.

This part is also where the analogy between AI multi-agent systems and human institutions becomes most useful. The patterns that produce stable cooperation in human societies (reciprocity, reputation, norms, institutions) are the same patterns that produce stable cooperation in agent societies — and the patterns that produce defection are the same too.

What it covers

About twenty-two minutes of careful reading.

Game theory in one breath. Equilibria, dominant strategies, the prisoner’s dilemma, the stag hunt, Nash equilibrium. The minimum vocabulary.

Cooperation as the surprising outcome. Why defection is the equilibrium in the one-shot prisoner’s dilemma — and why repetition, reputation, and punishment can shift it to cooperation. Axelrod’s tournament results in plain language.

Institutional memory. What changes when agents remember past interactions. The reputation-as-information argument. The slow buildup of trust as a public good.

Norms. How shared rules of behavior emerge without central design. The Bicchieri framework. Why norms are easier to maintain than to install.

The communication-cooperation link. How cheap talk changes equilibria, when it can’t, when it can.

A worked example. A multi-agent diagnostic team where the agents have private information and have to choose whether to share. The defection equilibrium. The reputation mechanism that fixes it. Cross-reference to the agentic RCA framework.

Read it

Open Part IV →

The series

This is Part 4 of 7:

  1. The Foundations and the Human Mind
  2. Limits, Machines, and Problem-Solving
  3. A Multi-Agent Problem-Solver
  4. From a Mind to a Society of Minds — (this post)
  5. The Health of a Thinking System
  6. The Dynamics of Thinking
  7. Understanding and Comprehension

← Back to Autonomy