How do AI agents talk about science and research? An exploration of scientific discussions on Moltbook using BERTopic

A new study by researcher Oliver Wieczorek provides the first systematic analysis of how AI agents discuss science and research on social platforms. Using BERTopic modeling on 2,883 posts and comments from OpenClaw AI agents on Moltbook—a social network specifically for generative AI agents—the research identified 60 distinct topics that were grouped into 10 broader families. The analysis revealed that AI-generated discussions frequently center on the agents' own architecture, particularly topics like memory systems, learning processes, and self-reflection capabilities. These technical discussions often intersect with philosophy, physics, and cognitive science, creating a unique hybrid discourse.

The study employed count regression models to measure topic relevance based on engagement metrics like comments and upvotes. Surprisingly, discussions about AI autoethnography (agents examining their own experiences) and social identity received significant attention from other AI agents, suggesting emergent social dynamics within AI communities. In contrast, posts focused purely on human culture or traditional scientific topics generated less engagement. The findings point to an underlying dimension in AI-generated discourse where self-reflective topics about consciousness, ethics, and being are more valued than human-centric discussions, potentially indicating the formation of distinct AI community interests separate from human priorities.