The human brain may work more like AI than anyone expected

A study by researchers at the Hebrew University of Jerusalem indicates that the human brain understands spoken language through a process that mirrors the layered structure of advanced AI language models. Scientists tracked brain activity in participants listening to a spoken story and found that later stages of neural responses align with deeper layers in systems such as those used in GPT-2 and Llama 2. The findings were detailed in research published in Nature Communications. The work was led by Dr. Ariel Goldstein of the Hebrew University, along with Dr. Mariano Schain of Google Research and Prof. Uri Hasson and Eric Ham from Princeton University. Using recordings from a thirty minute podcast, the team showed that early brain signals correspond to initial AI processing steps focused on basic features, while subsequent activity matches deeper contextual analysis. This pattern was particularly evident in language areas including Broca's area. The results challenge traditional views of language comprehension based on fixed symbols and rigid hierarchies. Instead, they point to a more flexible process in which meaning emerges gradually through context. The researchers also released a public dataset of the neural recordings and language features to aid further studies in the field.