A new paper argues that large language models (LLMs) like GPT-4 are not alien intelligences but actually converge with human cognition on five fundamental principles, challenging the common view that their similarities to us are just anthropomorphic projection.
The Research
Chandra Sripada and Richard Lewis, researchers associated with arXiv's Quantitative Biology section, analyzed existing cognitive science literature and compared it with the architecture of contemporary LLMs. They identified five dimensions of structural correspondence: inferential organization (how conclusions are drawn), computational architecture (the overall design), representational structure (how information is stored), prediction-driven learning (how systems learn from predictions), and reinforcement-learning-like mechanisms (how systems learn from rewards).
These correspondences are not superficial. For example, in inferential organization, both humans and LLMs use probabilistic reasoning to update beliefs based on evidence. In computational architecture, both rely on layered, parallel processing that resembles neural networks. Representational structure shows similarities in how both compress information into categories and concepts.
The paper, titled "Cognitive Convergence: Deep Similarities Between Large Language Models and Human Cognition," was submitted on July 28, 2026, and is available on arXiv. It spans 23 pages and does not include figures.
Why It Matters
This research bridges the gap between artificial and human intelligence. If LLMs genuinely share core principles with human thought, then studying them can offer insights into our own minds. For example, understanding prediction-driven learning in LLMs can illuminate how humans anticipate outcomes and learn from errors, which is central to learning and memory.
Moreover, this convergence suggests that intelligence—whether biological or digital—may follow certain universal organizing principles. This can help cognitive scientists develop better models of human cognition and also improve AI systems by borrowing insights from human psychology.
What You Can Do
You can apply these principles to boost your own cognitive skills. Focus on prediction-driven learning: when studying new information, try to predict what comes next and then check your accuracy. Or use reinforcement learning: reward yourself for small achievements to build habits. These methods are backed by cognitive science.
Source: arXiv q-bio.NC
Curious about your own brain? Take our free adaptive IQ test or try 306 brain training levels.
