Even while passively listening to an audiobook, your brain is already predicting what word comes next—and it's doing so in milliseconds, using computational logic surprisingly similar to artificial intelligence. Researchers at Friedrich-Alexander-Universität Erlangen-Nürnberg and Heidelberg University have made this discovery by combining natural speech recordings with high-resolution brain measurements, revealing that human cognition and large language models may organize information in fundamentally comparable ways.
The finding matters because it bridges a long-standing gap between neuroscience and artificial intelligence, offering fresh insight into how the human mind processes language and suggesting why AI systems have become so effective at language tasks. For decades, researchers debated whether humans possess innate grammatical knowledge or learn language purely through experience—a question that has intensified as powerful AI language models have emerged, built on the principle that language can be processed by predicting what comes next.
PD Dr. Patrick Krauss and his interdisciplinary team combined three methodologies to test this: they recorded participants listening to a continuous audiobook while simultaneously measuring their brain activity using electroencephalography and magnetoencephalography. They then compared this neural activity directly against the predictive probabilities calculated by large language models, tracking everything with temporal resolution measured in mere milliseconds.
The results were striking. The brain became active before a word even began to sound—a sign of prediction already underway. Crucially, when a word was highly probable in its context, the neural response was weaker. Conversely, unexpected words triggered much stronger neural reactions. "This allowed us to prove that the brain actively predicts language. These predictions can be measured and follow similar patterns to modern language models," Krauss explained.
What surprised the researchers most was not just that the brain and language models made similar predictions, but that they appeared to organize language internally in comparable ways. Language models operate as artificial neural networks—mathematical information processors architecturally inspired by the human brain itself—while biological brains use electrical and chemical signals. Yet despite these mechanical differences, both systems seemed to converge on similar organizational principles.
The convergence doesn't necessarily mean the brain and AI work identically, as PD Dr. Achim Schilling cautioned. Rather, it suggests they follow similar information-processing logic. The real puzzle is why two such different systems would independently arrive at such similar ways of handling language—and where the boundaries of this similarity end.
The implications extend well beyond pure science. Understanding how the brain and language models both represent and predict language could eventually lead to new diagnostic tools for language disorders, personalized therapies tailored to individual brain patterns, and perhaps even more transparent artificial intelligence systems that better reflect human reasoning. For now, the research team is focused on testing whether these principles are robust enough to apply beyond their laboratory findings—work that could reshape how we understand both human cognition and artificial intelligence for years to come.