Scientists find similarities between AI language models and human brain patterns

A new study published in Nature Machine Intelligence reveals specific areas where large language models (LLMs) are developing processing patterns similar to human brain functions.

Key findings; Scientists at Columbia University and the Feinstein Institutes for Medical Research Northwell Health discovered similarities between LLM processing hierarchies and human neural patterns during language processing.

• The research team evaluated 12 open-source LLMs with similar parameter sizes, ranging from 6.7 to 7 billion parameters
• Models included prominent LLMs such as Llama, Llama2, Falcon, Mistral, and others
• Mistral demonstrated the strongest performance in matching human-like processing patterns

Methodology and data collection; The study utilized a unique approach to gather human brain activity data through medical procedures while maintaining consistent testing conditions for both human and AI subjects.

• Eight patients undergoing epilepsy treatment with implanted intracranial electroencephalography (iEEG) electrodes participated in the study
• Participants listened to story passages and conversations while their brain activity was recorded
• LLMs were tested on identical content using reading comprehension and commonsense reasoning tasks
• The research specifically focused on models using a consistent stacked transformer decoder architecture

Technical insights; The study revealed notable parallels between artificial and biological neural processing hierarchies.

• Higher-performing LLMs demonstrated more brain-like layered structures in their processing
• The hierarchical nature of language processing, from basic sounds to complex phrases, appears to drive the convergence between LLMs and human cognition
• The black box nature of AI decision-making remains a challenge in fully understanding these similarities

Scientific implications; The research provides new perspectives on the evolution of artificial intelligence and its relationship to human cognition.

• The study uniquely compares multiple LLMs using a consistent architecture, allowing for more direct comparisons
• Results show that LLMs are developing processing patterns that mirror the brain’s cortical areas responsible for sound and language processing
• This alignment between AI and human processing could inform the development of more cognitively compatible AI systems

Future directions; While the study reveals important convergences between artificial and biological intelligence, questions remain about the full extent and implications of these similarities.

• The research opens new pathways for developing AI systems that better align with human cognitive processes
• Understanding these parallels could lead to more effective human-AI interaction design
• Further research is needed to explore the deeper implications of these convergent processing patterns