A new study from the Technion-Israel Institute of Technology has revealed a key difference between how the human brain and artificial intelligence (AI) models process language. While AI models analyze all the words in a sentence at once, the human brain creates a summary as it reads or listens, helping it make sense of longer texts.
How AI and the Human Brain Process Text
Large language models (LLMs) like ChatGPT and Bard have transformed AI’s ability to generate and understand text. These models are designed to work like the human brain in some ways, but they still operate differently.
The study, published in Nature Communications, explored this difference by examining brain activity while people listened to stories. The research was led by Professor Roi Reichart and Dr. Refael Tikochinski at Technion and co-supervised by Professor Uri Hasson at Princeton University.
Researchers analyzed fMRI scans of 219 participants while they listened to spoken texts. They compared the brain’s activity with predictions made by existing AI models. The results showed that AI could accurately predict brain activity when processing short texts (a few dozen words). However, for longer texts, AI models failed to match how the human brain responded.
Why AI Struggles with Long Texts
The reason behind this difference lies in how the brain and AI process information.
For short texts, both the brain and AI process all words simultaneously. But when faced with longer texts, the brain takes a different approach. Instead of trying to process every single word at once, the brain builds a summary—a kind of “mental storage”—to keep track of the overall meaning. This summary helps us understand new information in the right context.
AI models, on the other hand, do not need to create summaries. They process all previous words at the same time, treating each word equally rather than forming a structured context over time. This fundamental difference explains why AI models struggle to predict brain activity when people listen to long texts.
Mimicking the Brain’s Approach
To test their findings, researchers built an improved AI model that mimicked the brain’s summarization process. Instead of analyzing the full text at once, this model created summaries of past text and used them to interpret upcoming words. This adjustment made AI predictions of brain activity much more accurate.
The study also mapped the brain regions responsible for both short-term and long-term language processing. It identified areas involved in accumulating context, which help us understand ongoing narratives, whether in a conversation, a book, or a lecture.
What This Means for AI and Human Understanding
This research highlights how the human brain has evolved to efficiently handle vast amounts of information by summarizing and storing key details. It also suggests that AI could improve by adopting a similar method, allowing it to process longer texts in a more human-like way.
While AI has become powerful at generating and analyzing text, it still lacks the brain’s ability to create structured meaning over time. By learning from how our brains summarize information, future AI models could become more natural in how they understand and process language.
The research findings can be found in Nature Communications.
Copyright © 2025 Knowridge Science Report. All rights reserved.
