NYU study finds human brains predict words like AI language models do — suggesting a shared mechanism between biological and artificial intelligence

NYU researchers published a study comparing how the human brain and large language models (LLMs — AI systems trained to predict and generate text) process language. The research examined whether both use similar prediction mechanisms when reading or listening, finding unexpected parallels in how they anticipate what word comes next.

Unlike previous theories that treated brain prediction and AI language modeling as separate phenomena, this study suggests they may operate on the same underlying principle: both systems build a statistical model of language patterns and use that model to guess upcoming words before they arrive. This shifts how neuroscientists and AI researchers think about the relationship between human cognition and machine learning.

For AI developers, this validates that LLM architectures mirror real cognitive processes — meaning improvements to language models might reveal insights into human learning, and vice versa. For neuroscience and psychology professionals, it provides a concrete computational framework (borrowed from AI) to test and explain how the brain handles language, moving beyond older, vaguer descriptions of 'prediction.'