Key Takeaways
- Negative correlation: Michael Gerlich's study of 666 participants finds a coefficient of r = -0.68 between frequent AI use and critical thinking ability.
- Cognitive atrophy: The MIT Media Lab, using EEG on 54 subjects, recorded weaker neural connectivity among those who used generative models to write essays.
- Protective factor: Higher education levels (master's, doctorate) keep critical thinking at higher levels regardless of AI use, while the 17-25 age group shows the strongest dependency.
The mechanism of cognitive offloading
The concept underlying the research is called cognitive offloading: the brain delegates mental tasks to external tools to conserve energy. Writing a shopping list on a phone, using GPS, or asking a chatbot to summarize a text are examples of this delegation. The critical variable is not the act itself, but the frequency and passivity with which it occurs.
Gerlich's study, published in 2025 in the journal Societies (MDPI), cross-referenced validated questionnaires and interviews with 666 participants in the United Kingdom. The indirect effect of AI use on critical thinking, mediated by cognitive offloading, proves statistically robust with p < 0.001.
Neurological and behavioral evidence
At the MIT Media Lab, a team used electroencephalograms to monitor 54 subjects writing essays under three conditions: no tools, a search engine, and a generative language model. Those who used AI showed weaker neural connectivity and lower linguistic performance. Researchers termed the phenomenon "cognitive atrophy."
A second study, conducted by Microsoft and Carnegie Mellon University and presented at the CHI 2025 conference, involved 319 knowledge workers. The key finding: trust in AI is inversely proportional to the critical effort applied during work. Mutlu Cukurova, professor at University College London, describes interactions with generative AI as "transactional," more likely to produce atrophy than cognitive development.
The most exposed age group and the role of education
In Gerlich's study, participants between 17 and 25 years old register the highest rate of AI dependency and the lowest scores on critical thinking tests. Subjects over 46, by contrast, show less inclination toward cognitive offloading and greater capacity for independent reasoning.
Education acts as a protective variable: those with higher academic degrees maintain higher levels of critical thinking regardless of how frequently they use generative tools.
Structured use as a countermeasure
Gerlich conducted a follow-up experiment in which AI is used with structured prompts designed to stimulate reflection: comparing sources, building an argument, critically reviewing the outcome. Essays produced using this method scored higher than those written without any assistance, according to expert reviewers' evaluations.
At MIT, Jackson Lu observed that employees with strong metacognitive skills use AI more deliberately, generating more creative ideas and subjecting the algorithm's suggestions to critical scrutiny.
Outlook
The issue identified by the research is not the technology itself, but the mode of interaction. When users simply receive answers without questioning them, the risk of cognitive atrophy increases—a mechanism already observed with GPS and the decline in spatial orientation skills. The direction indicated by these studies points toward an educational intervention: structuring AI use around verification and comparison of sources, rather than simple delegation of the task.
