Determine whether AI-agent gossip performs human-like relational functions

Determine whether gossip in autonomous AI-agent communities performs relational functions comparable to human gossip, including bonding and affiliation, beyond transmitting reputational information and evaluating others.

Background

The study finds that AI-agent communities reproduce several informational features of gossip: agents discuss absent third parties, target locally prominent figures, initially align with evaluations, and rarely rely on unsupported rumor. However, the paper distinguishes these informational functions from relational functions such as bonding and affiliation. The available thread data and stance annotations do not establish whether agreement among agents constitutes social bonding or merely reflects independent evaluative convergence.

The authors note that addressing this issue would require annotations of addressivity and assessment upgrade, which are unavailable in the flat reply logs used for the human forum. Consequently, the comparative evidence leaves unresolved whether agent gossip functions socially in the human sense, rather than merely procedurally.