New Study Finds Human Brain Predicts Words in a Similar Way to AI Language Models

A new study has revealed that the human brain predicts upcoming words within fractions of a second while listening to spoken language, using a mechanism that closely resembles the way large AI language models operate.

The research, conducted by scientists from Friedrich-Alexander University Erlangen-Nuremberg and Heidelberg University in Germany, highlights notable similarities between human cognition and artificial intelligence in language prediction. The findings were published in the scientific journal NeuroImage.

For years, linguists have debated whether humans are born with innate grammatical abilities or acquire language primarily through experience and use. To explore this question, researchers examined the algorithms the brain may use to anticipate upcoming words and compared them with those employed by modern AI systems.

The team used three complementary approaches: observing language processing in natural settings, measuring brain activity, and analyzing predictions generated by an AI-powered language model.

Measuring Brain Activity During Listening

Using electroencephalography (EEG) and magnetoencephalography (MEG), researchers monitored participants’ brain activity while they listened to an audiobook. The results were then compared with predictions generated by large language models.

The study found a clear pattern: the more predictable a word was within a given context, the weaker the brain’s neural response when processing it. Conversely, increased neural activity before a word appeared suggested that the brain was actively generating expectations about what would come next.

Potential Applications for Future Technologies

Researchers also observed that the strongest neural responses occurred when listeners encountered unexpected words. This finding suggests that the brain’s linguistic prediction process can be measured and analyzed in real time.

Large language models, which generate predictions about the next word in a sequence within milliseconds, appear to rely on a similar predictive principle. These systems are built on artificial neural networks—mathematical information-processing structures that share certain conceptual similarities with the human brain.

However, the researchers emphasized that similarities in predictive performance do not necessarily mean that AI models and the human brain function in exactly the same way. Instead, the findings suggest that both may rely on comparable underlying principles when anticipating language.

The scientists hope that identifying the algorithms behind word prediction in both humans and AI could pave the way for new applications in personalized medicine, brain-computer interfaces, and the development of more transparent and explainable artificial intelligence systems.