Offline World Models as Imagination Networks in Cognitive Agents
New research shows AI's 'imagination' is fundamentally different from human cognition.
Deep Dive
A new study analyzing 'imagination networks' in 2,743 humans and six large language models (LLMs) found a stark difference. Human imagination shows robust, consistent internal structure, while LLMs exhibit minimal clustering and weak correlation with human networks, even with conversational memory. This reveals a fundamental disparity in how biological and artificial systems organize internal world models, challenging assumptions about AI's cognitive similarity to humans.
Why It Matters
This suggests current AI lacks a core human-like cognitive architecture, which could limit true reasoning and generalization.