Using large language models like ChatGPT consistently can decrease brain activity and neural connectivity, research from MIT found.

As LLMs become more commonly used, researchers are curious to see how relying on AI can impact brain function and critical thinking.

To test this, researchers at MIT had study participants engage in a series of essay writing tasks, with some relying purely on their brain, some able to use search engines, and some using OpenAI’s LLM ChatGPT.

In the three rounds of essays, researches noticed that LLM users offered ‘low effort’ outputs in the third round, relying almost entirely in a copy-past answer of the ChatGPT answer.

Using electroencephalography (EEG) to record participant’s brain activity, researchers discovered significantly different patterns in neural connectivity and cognitive strategies between the various groups.

Those consistently using LLMs had the weakest level of neural connectivity, as well as the weakest level of neural network engagement. Those that used their brain exhibited better memory recall, as well as visual processing node activations.

LLM users also struggled to report ownership of their essays, with many unable to quote any of the essays they produced just minutes before.

When LLM users were later asked to rely purely on their brain power to produce another essay, “they showed weaker neural connectivity and under-engagement of alpha and beta networks; and the Brain-to-LLM participants demonstrated higher memory recall, and re‑engagement of widespread occipito-parietal and prefrontal nodes, likely supporting the visual processing.”

Though Brain-only participants had a greater cognitive load, they exhibited higher levels of neural connectively and internal content generation, which correlated with stronger memory, greater semantic accuracy, and firmer ownership of written work. This all contributed to deeper learning outcomes.

Recommended reading

• ‘Reasoning’ AI Emits Up to 50x More Carbon Than Simpler Models
• New Apple Paper Pours Cold Water on AI Reasoning
• Is AI Making Us Dumber? Workers Think So.

The search engine groups exhibited potentially the best balance of effort and outcome, as their ability to search for facts enabled them to answer essays with greater accuracy, though still relying on their own creativity and cognition to develop an essay.

While ChatGPT offered greater efficiency, the LLM groups had overall weaker memory traces, reduced levels of self-monitoring, and worse ownership creditability.

While these are just pre-publication results and the study has yet to be peer reviewed, the study exhibits some of educators worst fears when it comes to an over-reliance on AI.