Researchers at the Hebrew University of Jerusalem revealed a startling discovery: the human brain understands spoken language through a tiered process that almost perfectly mirrors the internal architecture of advanced AI systems like GPT-2 and Llama 2. Led by Dr. Ariel Goldstein in collaboration with Google Research and Princeton University, the study recorded brain activity from participants listening to a 30-minute podcast. The findings, published in Nature Communications, show that as a person listens to a story, their brain builds meaning in a structured, step-by-step sequence.
Scientists Are Surprised by How Closely the Brain Resembles AI https://t.co/A9yvwK8n1n pic.twitter.com/kWAF5n8yM8
— HealthIT Policy (@HITpol) January 27, 2026
The research team found that “early” brain responses to a word align with the initial, shallow layers of an AI model, which handle basic features. As the word is integrated into a sentence, the activity moves to “deeper” language regions like Broca’s area, matching the deeper layers of AI where complex context and tone are synthesized. This discovery challenges decades of traditional linguistic theory, which argued that the brain relies on rigid, symbolic rules. Instead, the brain appears to use a more flexible, statistical approach, one that Large Language Models have independently converged upon through machine learning.
You May Like To Read: Diplomatic Thaw: Why Starmer’s Beijing Visit Signals a Global Shift
To accelerate future breakthroughs, the team has released the full dataset of neural recordings to the public. This open resource provides a new benchmark for scientists to test how the human mind physically constructs meaning. While the brain and AI are built differently, this “shared roadmap” suggests that there may be a universal mathematical logic to how intelligence, whether biological or artificial, extracts meaning from a stream of sound.
Check out our latest video:
