Internal World Models as Imagination Networks in Cognitive Agents
Published 6 Nov 2025 · arXiv · Saurabh Ranjan
Overview
The paper examines the computational role of imagination in cognitive agents, focusing on internal world models (IWMs) in humans and large language models (LLMs). It uses psychological network analysis to compare imagination networks.
Key Insights
- Imagination Networks: Human imagination networks show strong correlations between centrality measures like expected influence and closeness.
- Evidence: Based on vividness ratings from questionnaires.
- Verifiable: Yes, through the study's methodology.
- LLM Networks: LLMs exhibit a lack of clustering and lower correlations between centrality measures.
- Evidence: Observed under different prompts and memory conditions.
- Verifiable: Yes, through the study's methodology.
- Comparison: The study highlights a lack of similarity between human and LLM IWMs.
- Evidence: Comparative analysis of network structures.
- Verifiable: Yes, through the study's methodology.
BFSI Relevance
- Why Relevant: Understanding imagination in AI can enhance decision-making models in financial services.
- Primary Sector: Financial Services
- Subsectors: Asset Management, Risk Assessment
- Actionable Implications:
- Explore AI models that incorporate human-like imagination for better predictive analytics.
- Develop training programs for AI systems to improve decision-making processes.
researcher peer-reviewed-paper cross-bfsi technology-and-data global