Sociology of AI: relations between the agents

Several AI assistants inside one organisation already form a small society, with its own rules, its own leaders and its own weak points. We study it as researchers and as engineers.

Relations are weighed, not described Relations are weighed, not described
Relations are weighed, not described

The field looks past the pair of human and model to the relations between the agents themselves: who defers to whom, how norms spread through shared memory, and what happens when the group is wrong together.

Hypotheses are tested on the bench rather than summarised from other people’s papers. The boundary matters as much: a program reproduces an averaged image drawn from texts, not a person, and a result about agents is not a claim about people.

For a customer this is not abstract. One assistant’s error propagates through shared data, and agreements between two programs are invisible to the people who employ them. Both can be described and measured in advance.

Materials on this topic

Other topics

AI red teaming: what a test shows, and what it does notAI standards: where the rules actually applyAI supply chain: what enters before the model runsMulti-agent systems: behaviour between the agentsIncident chronicle: documented consequences, not forecastsBehavioural audit: how agent behaviour drifts over timePrompt injection: where text becomes an instructionRussian practice: who is held responsible for AIMethodology: how the Institute tests AI systemsData and benchmarks: what our conclusions rest onGovernment customers: how to commission a testGovernance and funding of the InstituteCases: scenarios of work and what they show

Page updated 02 November 2026 · site build 2026.11.13 · ANO «ISAIS»