# Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task

**Authors:** Nataliya Kosmyna, Eugene Hauptmann, Ye Tong Yuan, Jessica Situ, Xian-Hao Liao, Ashly Vivian Beresnitzky, Iris Braunstein, Pattie Maes  
**arXiv:** [2506.08872](https://arxiv.org/abs/2506.08872)  
**Date:** 2025-06-10  
**Summary page:** https://www.summarizepaper.com/en/paper/2506.08872v1/

## Key points

- Study investigated effects of Large Language Models (LLMs) in essay writing tasks on neural and behavioral levels
- Participants divided into LLM, Search Engine, and Brain-only groups for sessions 1-3
- EEG used to measure cognitive load during essay writing; NLP analyzed essays
- Significant differences in brain connectivity among groups: Brain-only strong networks, Search Engine moderate engagement, LLM weak connectivity
- LLM-to-Brain participants showed reduced alpha and beta connectivity in session 4
- Brain-to-LLM users exhibited higher memory recall and activation of specific brain areas similar to Search Engine users
- LLM group had lowest self-reported ownership of essays and struggled with quoting own work accurately
- Over four months, LLM users consistently underperformed at neural, linguistic, and behavioral levels compared to other groups
- Study raises concerns about cognitive costs of relying on LLMs for educational tasks; emphasizes need for further research into AI's role in education

## Summary

This study investigated the effects of using Large Language Models (LLMs) in essay writing tasks on neural and behavioral levels. Participants were divided into three groups: LLM, Search Engine, and Brain-only (no tools), with each group completing three sessions under the same condition. A fourth session was conducted where participants from the LLM group switched to the Brain-only group (LLM-to-Brain), and participants from the Brain-only group switched to the LLM condition (Brain-to-LLM). A total of 54 participants took part in Sessions 1-3, with 18 completing session 4. The study utilized electroencephalography (EEG) to measure cognitive load during essay writing and analyzed essays using Natural Language Processing (NLP). Essays were also scored by human teachers and an AI judge. The results showed that within-group homogeneity was observed in Named Entity Recognition (NERs), n-gram patterns, and topic ontology. EEG analysis revealed significant differences in brain connectivity among the groups: Brain-only participants exhibited strong and distributed networks, Search Engine users showed moderate engagement, while LLM users displayed weak connectivity. In session 4, LLM-to-Brain participants exhibited reduced alpha and beta connectivity, indicating under-engagement. On the other hand, Brain-to-LLM users showed higher memory recall and activation of specific brain areas similar to Search Engine users. Self-reported ownership of essays was lowest in the LLM group and highest in the Brain-only group. Additionally, LLM users struggled with accurately quoting their own work. Over a four-month period, LLM users consistently underperformed at neural, linguistic, and behavioral levels compared to other groups. These findings suggest potential cognitive costs associated with relying on LLMs for educational tasks. The study raises concerns about the long-term implications of LLM usage in learning contexts and emphasizes the need for further research into AI's role in education. For more detailed information on the experimental design, participant protocols, prompts used during each session, as well as additional data summaries including specific EEG dDTF values, readers are encouraged to visit https://www.brainonllm.com/.