TEMPO.CO, Jakarta - A new study from the Hebrew University of Jerusalem has found striking similarities between how artificial intelligence systems and the human brain process language, particularly in their understanding of large language models.
Researchers explored this parallel by comparing recordings of human brain activity with AI responses while both listened to the same 30-minute narrative.
Led by neuroscientist Ariel Goldstein, the study addressed a fundamental question about how humans understand language. Traditional linguistic theories have often viewed language comprehension as a rigid, hierarchical system of symbolic rules.
Goldstein and his team found a far more dynamic process at work. Their research suggests that the brain does not simply apply fixed grammatical rules. Instead, it builds meaning gradually, layer by layer, in a manner that closely resembles the deep-learning architecture used in modern artificial intelligence.
When Brains and AI Follow the Same Sequence
“Even though these systems are built very differently, both seem to converge on a similar, step-by-step build-up toward understanding,” Goldstein said, as quoted in Earth on Sunday, December 14, 2025.
“What surprised us most was how closely the brain’s temporal unfolding of meaning matches the sequence of transformations inside large language models.”
To test their hypothesis, the researchers conducted an unusually rare experiment. Rather than using standard brain imaging techniques, they analyzed electrocorticography recordings from epilepsy patients undergoing brain surgery.
The neural data were then compared with how advanced AI language models such as GPT-2 and Llama 2 process the same story.
The findings showed a strong correlation between the depth of layers in AI models and delays in brain responses. In both cases, deeper and more meaningful information required longer processing time. This indicates that the human brain, much like AI systems, takes additional time to interpret complex language.
What the Findings Mean for AI and Neuroscience
Goldstein said he was struck by how closely the two systems resembled each other. Despite being built from entirely different materials, biological neurons in the brain and silicon-based algorithms in AI, both appear to rely on similar strategies to understand language.
The implications of the study are wide-ranging. On one hand, it offers new validation for artificial intelligence technology.
Large language models are not merely statistical tools that generate text but also function as increasingly accurate models of how the human mind processes language.
On the other hand, the findings narrow the long-standing divide between computer science and biology. Words that AI models can predict more effectively were also found to trigger stronger and earlier responses in the human brain, suggesting that human cognition itself operates as a powerful predictive system.
Read: IPB Researchers Develop Natural Shampoo from Pandan Leaves
Click here to get the latest news updates from Tempo on Google News