A new pre-print study by MIT Media Lab researcher Nataliya Kosmyna and her team suggests that using ChatGPT to write essays may reduce brain activity, undermine memory, and diminish creativity—especially when relied on heavily or early in the writing process. The findings have sparked caution around AI use in education and cognitive development.
What the Study Did
- Participants: 54 volunteers (ages 18–39), divided into three groups:
- Brain-only – wrote essays unaided
- Google Search – used web search to research
- ChatGPT – used LLM assistance for writing
- Method: Over four sessions, participants wrote SAT-style essays while wearing EEG headsets to record brain activity.
- In session 4, researchers swapped methods: ChatGPT users wrote without AI, and vice versa.
🔬 Key Findings
- Decline in EEG-measured brain connectivity
ChatGPT users showed the weakest neural activity—up to 55% lower connectivity than brain-only users, with Google users in between. - Lower memory recall & essay ownership
Over 83% of former ChatGPT users couldn’t recall their own recent writing. They reported feeling less “ownership” of what they wrote. - Reduced novelty and originality
Essays generated with AI were judged to be formulaic and “soulless” by educators, while independent writers produced more original, creative work - Order of usage matters
Brain-to-LLM (write then refine with AI) users maintained strong neural engagement. In contrast, starting with AI led to persistent low engagement, even after removing AI. - Cognitive debt warning
The authors warn of accumulating “cognitive debt”—ease and efficiency at the cost of learning, critical thinking, and memory.
⚠️ Why It Matters
The study raises urgent questions about AI’s classroom role and how dependency on ChatGPT might hinder internal thinking, memory retention, and creativity. Lead researcher Nataliya Kosmyna shared the findings pre-peer review to spur early dialogue on these risks.
Limitations & Balance
- Early-stage research: Still awaiting peer review; small, region-limited sample (Boston-area students).
- Not all AI use is harmful: Properly timed AI tools—used after independent thinking—could enhance learning.
- Growing body of mixed evidence: Other MIT studies note risks like emotional dependency on AI tools.
🔎 What’s Next?
- Researchers are now exploring how AI affects programming skills and other cognitive domains. Early results suggest even stronger negative effects.
- Educators and policymakers are encouraged to teach balanced AI use—educating students to first think independently, then use AI to refine their ideas.time.com
✅ Final Take
The MIT study serves as an important wake-up call: while ChatGPT offers efficiency and convenience, overuse—especially early in learning—may impose hidden cognitive costs. For students, writers, and lifelong learners, the message is clear: let AI assist, not replace, your own thinking.
