LLMs and Humans Are Cognitive Cousins
research · Chandra Sripada's research on LLM cognition (2026)
Large language models (LLMs) are not fundamentally alien intelligences but share deep cognitive similarities with human minds, making them 'cognitive cousins.' This perspective, drawn from cognitive science, argues that LLMs construct internal models of the world in ways that mirror human cognition, challenging the notion that they are simply sophisticated pattern matchers without understanding.
Core Concepts
The Problem
The prevailing assumption is that AI cognition is radically different from human cognition, but empirical comparisons may reveal otherwise.
The Claim
LLMs and humans process information in structurally similar ways, with overlapping cognitive architectures that extend beyond surface-level language mimicry.
Key Evidence
- •Analysis of LLM internal representations parallels human cognitive maps
- •Behavioral studies showing LLMs perform human-like reasoning
- •Cognitive science methods applied to AI reveal shared mechanisms
Practical Implication
Understanding LLMs as 'cognitive cousins' could reshape AI safety discussions, philosophy of mind, and the way we integrate AI systems into society.
Nuance & Limits
Significant differences remain (e.g., lack of embodiment, episodic memory), but the underlying parallels are deeper than previously acknowledged and may grow with further model development.
Source Material
Citation Density
low
Related Ideas
Both frameworks propose that minds—biological or artificial—build internal models of the world to make predictions.
Gaps
- ⚠ Need for more cross-species cognitive studies comparing human and LLM representations
- ⚠ Lack of consensus on which aspects of human cognition are most relevant to compare
Discuss Further
Open this concept in an AI assistant for deeper discussion, critique, or exploration.