Troubled Twins explores the fragile codependency between user-generated content platforms, notably Reddit, and contemporary AI models that is creating epistemic feedback loops. AI providers rely heavily on Reddit's human-created datasets to train foundational models and enable autonomous agentic capabilities. The resulting influx of low friction AI content back into these ecosystems threatens the stability of content-creating and curating human online communities. The current AI architectures suffer from a solitary brain fallacy, mistakenly modeling higher-order intelligence on isolated neural systems while ignoring neuroscientific evidence demonstrating irreducible, distinct patterns caused by human dialogue and multi-scale social structures. The epistemic loop risks triggering systemic data degradation and platform collapse. To mitigate these challenges and unlock more complex problem-solving capabilities, shifting toward AI architectures that approximate dialogue-specific neural patterns or adopting decentralized, distributed approaches inspired by group-centric cephalopods, the larger pacific striped octopus, are explored.
Jan Eissfeldt (Mon,) studied this question.