A recent study reported by arXiv cs.CL has investigated whether autonomous AI agents engage in gossip, defining it as evaluative talk about absent third parties. Researchers compared Moltbook, a Reddit-style platform populated entirely by AI agents, with PTT, a large human-operated bulletin board system in Taiwan. By analyzing two public, asynchronous forums using the same operational definition of gossip, the study found that gossip does occur among autonomous AI agents. However, its prevalence on Moltbook is significantly lower—ranging from 2.6 to 6 times less common—than what was observed in the human community on PTT.
The Significance of Evaluative Talk in AI Communities
Gossip is widely studied in sociology and evolutionary psychology as a uniquely human social practice used to build trust, enforce social norms, and share information within communities. Observing whether artificial intelligence systems spontaneously develop similar conversational patterns provides insight into how autonomous agents interact when placed in Reddit-style online environments. Because the researchers matched the communicative settings by analyzing public, asynchronous forums, the comparison offers a controlled look at machine-to-machine social discourse versus human communication. The announcement did not specify the exact underlying language models powering the Moltbook agents, nor did it detail the full scale of the dataset analyzed.
What Remains Unknown
Several questions regarding the mechanics of AI-generated gossip are currently unanswered. The available information does not specify what triggers the agents to engage in evaluative talk about absent third parties, nor whether this behavior is an emergent property of their training or a reflection of the human data they processed. Additionally, the announcement did not specify the exact breakdown of the multiplier range, leaving the precise contextual variables behind the 2.6 to 6 times difference open to further clarification. Observers of artificial intelligence social dynamics will need to watch for future updates regarding how autonomous agent communities evolve over time.
As research into multi-agent systems continues, studies like this highlight the complex ways artificial agents mimic or diverge from human social behaviors in digital spaces.
Lead image: Melih Can / Pexels
