Home / Articles

AI is a tool, not a teacher

A study done by researchers at MIT raises red flags about the cognitive cost of using Artificial Intelligence (AI) for educational pursuits.

ByBaylee Woodley, PhD
An open laptop with a vibrant blue screen faces the viewer. In the middle of the screen is a picture of a pink brain. The laptop with a picture of a brain sits against a golden yellow background.

In a world where endless streams of information (or misinformation) are at everyones’ fingertips, what are the implications of a tool that makes people passive and less critical consumers?  

A study done by researchers at MIT raises red flags about the cognitive cost of using Artificial Intelligence (AI) for educational pursuits.1 The researchers who conducted the study are careful to avoid condemning all uses of AI. They suggest that AI can be a useful tool for experts, but their warning is clear – using AI in educational pursuits has a cognitive cost. 

What happened during the study? 

This study was a small one. Fifty-four participants were divided into three groups and each group wrote three essays. The first group wrote all three essays using only their brains. The second group could use search engines, but they could not use AI. The third group used AI.2

Eighteen participants also wrote a fourth essay. For this essay, participants who had been in the brain-only and search engine groups could now use AI. Participants who had been in the AI group could now only use their brains.

The essays were based on SAT topics and were evaluated by human educators as well as a specially-designed AI judge. The cognitive load of all participants during their essay-writing sessions was measured using EEG (electroencephalography). Participants were also interviewed after each session.

What did the study show?

The EEG measurements showed that cognitive activity decreased as the use of external tools increased. 

The AI-using group showed the lowest levels of brain connectivity. They exhibited reduced recall and often struggled to quote essays that they had written only minutes before. They also showed reduced cognitive connections and under-engagement. By the time they were writing their third essay, participants in the AI-only group were mostly using copy-paste.

Participants who used only their brains and search engines showed greater recall and higher levels of brain activity. For the brain-only group, their cognitive connectivity increased throughout the first three sessions and peaked in the third session. 

The study found that some of the greatest disparity related to cognitive connectivity occurred in alpha band connectivity. Alpha bands are often associated with creative pursuits that require recall, brainstorming, and the combination of ideas without external aids. Delegating these creative pursuits to external tools changes cognitive function. The differences in total alpha band connectivity between groups were significant: 79 connections in the brain-only group and 42 connections in the AI-using group.

None of the EEG results indicated greater connectivity in any areas of the brain for those using AI. 

The human educators evaluating the essays were also able to identify those written using AI models because of their conventionality and the similarity of their points.

When the brain-only group was allowed to use AI for the fourth essay, they did not achieve the peaks of connectivity seen during their third session. That being said, they did show greater connectivity than that exhibited by the AI-using group during their first three sessions. The authors of the study use these results to suggest that those in the original AI group had reduced use of their cognitive connections, while those in the original brain-only group had greater connectivity to draw on when introduced to this new tool.

For brains that are already wired for creative, critical thought it was easier to engage meaningfully with AI as a tool. 

What does it mean?

There are potential benefits to integrating AI into education. The authors of the MIT study cite other research that shows AI models can facilitate adaptive learning as well as effective learning strategies like repetition and spaced learning.3

Alongside these potential benefits, there are risks to AI use. The results of this study show that it can reduce cognitive connections, lead to decreased effort and engagement, and impact memory. The authors of this study expand on these risks by citing related work that shows that AI can lessen critical thinking, reduce decision-making skills, discourage problem-solving, decrease cognitive development and long-term retention, and encourage passive information consumption.4

The researchers also point to other problems that come with using AI. Ironically, since it is decreased by AI use, critical thinking is an important skill to have when working with AI models. The sources used by AI models are often lost or misinterpreted, which increases the risk of misinformation. The content they produce is also based on their training data, which can be biased or inaccurate.5

The cost of using AI for easy, short-term solutions is cognitive development and creativity. It can be a great tool, but it is not a teacher. 

Sources

Bai, L., Liu, X., & Su, J. (2023). ChatGPT: The cognitive effects on learning and memory. Brain-X, 1(3). https://doi.org/10.1002/brx2.30.

Cacicio, S., & Riggs, R. (2023). ChatGPT: Leveraging AI to support personalized teaching and learning. Adult Literacy Education: The International Journal of Literacy, Language, and Numeracy, 5(2). https://doi.org/10.35847/SCacicio.RRiggs.5.2.70. 

Kosmyna, N., Hauptmann, E., Yuan, Y. T., Situ, J., Liao, X., Beresnitzky, A. V., Braunstein, I., and Maes, P. (2025). Your brain on chatgpt: Accumulation of cognitive debt when using an AI assistant for essay writing task. arXiv (preprint). https://doi.org/10.48550/arXiv.2506.08872. 

Milana, M., Brandi, U., Hodge, S., & Hoggan-Kloubert, T. (2024). Artificial intelligence (AI), conversational agents, and generative AI: implications for adult education practice and research. International Journal of Lifelong Education, 43(1). https://doi.org/10.1080/02601370.2024.2310448.

Pedró, F., Subosa, M., Rivas, A., & Valverde, P. (2019). Artificial intelligence in education: Challenges and opportunities for sustainable development. UNESCO. https://unesdoc.unesco.org/ark:/48223/pf0000366994.  

Sajja, R., Sermet, Y., Cikmaz, M., Cwiertny, D., & Demir, I. (2024). Artificial Intelligence-Enabled Intelligent Assistant for Personalized and Adaptive Learning in Higher Education. Information (Basel), 15(10). https://doi.org/10.3390/info15100596. 

Shen, Y., Heacock, L., Elias, J., Hentel, K. D., Reig, B., Shih, G., & Moy, L. (2023). ChatGPT and other Large Language Models are double-edged swords. Radiology, 307(2), e230163-e230163. https://doi.org/10.1148/radiol.230163.

Footnotes

  1. The study is by Kosmyna et al. (2025) and it can be read in full online (https://doi.org/10.1080/02601370.2024.2310448).
  2. They used OpenAI’s ChatGPT and no other tools were allowed.
  3. The authors cite Bai et al. (2023) and Sajja et al. (2024).
  4. The authors cite Bai et al. (2023) and Pedró et al. (2019).
  5. The authors cite Milana et al. (2024), Cacicio et al. (2023), and Shen et al. (2023).