MIT Study Finds Chatgpt May Harm Critical Thinking Skills

In early June 2025, researchers at MIT’s Media Lab quietly dropped a bombshell preprint titled “Your Brain on ChatGPT: Accumulation of Cognitive Debt…” (albertoromgar.medium.com, media.mit.edu). Their controlled experiment with 54 college-aged participants — measured via EEG during essay writing — has ignited new debates on generative AI’s impact on cognition and learning. 

---

What the Study Did?

• Three groups, three writing sessions:

  1. Brain-only – wrote without any assistance

  2. Google users – used web search

  3. LLM users – wrote with ChatGPT’s help
     (media.mit.edu, economictimes.indiatimes.com)

• Session 4 involved tool-switching: LLM→Brain-only and Brain-only→LLM paths. (media.mit.edu)

• Methods: EEG measured real-time brain connectivity, essays assessed via NLP, teachers, and an AI judge. (media.mit.edu)

---

Key Findings

1. Diminished neural engagement ✨

   • Brain-only authors displayed strongest EEG connectivity across frontal-parietal networks, tied to creativity and memory.

   • Google users ranked in the middle.

   • ChatGPT users showed the least brain activity, up to a 55% reduction in dDTF connectivity. (media.mit.edu, theregister.com, m.economictimes.com)

2. Cognitive offloading & passivity

   • ChatGPT users increasingly relied on copy-and-paste. By session 3, many simply input the prompt, reviewed output, and were done.

   • They struggled to recall or quote their own writing, reporting minimal sense of ownership. (thetimes.co.uk)

3. Language quality & critique decline

   • Essays from the LLM group were more formulaic, homogeneous, and perceived by evaluators as “soulless.” (thetimes.co.uk)

   • Google users performed better but still lagged behind Brain-only peers in critical engagement.

4. Switching tools matters

   • Participants who began unaided then used ChatGPT exhibited higher connectivity and memory recall — a hybrid “scaffolded” path that seemed beneficial. (edtechinnovationhub.com)

   • But going from LLM‑to‑Brain didn’t immediately restore their engagement — underlining the lingering cognitive debt.

---

Implications: What Does It All Mean?

Benefit	Risk
✅ Efficiency & convenience	⚠️ Cognitive atrophy – diminished critical thinking, memory, deeper learning
✅ Supports scaffolding – when used after thoughtful effort	⚠️ Dependency risk – habit-building reliance when started early
✅ Tool-assisted learning	⚠️ Copy-paste tendency – less personal reflection, ownership, nuance

Lead researcher Nataliya Kosmyna warns that while generative AI offers shortcuts, it carries “cognitive cost” — reducing engagement in memory, planning, and evaluation (media.mit.edu, edtechinnovationhub.com, arxiv.org, time.com). Though not peer-reviewed yet, it raises pressing concerns for classrooms already integrating AI.

---

A Balanced Approach

Experts emphasize moderation and strategy:

• Use ChatGPT after you’ve tried writing independently — then refine or brainstorm with AI.

• Resist letting it generate content wholesale; engage actively with AI outputs.

• Educators should teach prompt strategies, critical analysis of AI answers, and help students maintain ownership of their work.

A Cambridge researcher cautioned:

“AI could foster a kind of ‘laziness’ … but context matters — tutor use boosts, tool-as-crutch diminishes.” (thetimes.co.uk)

---

Final Thoughts

MIT’s study sends a powerful signal: our brains are sensitive to how we engage with AI. The effortless—but passive—support of ChatGPT can dull creativity, memory, and intellectual ownership. But when used thoughtfully, it can enhance learning.

As conversation turns to AI’s role in academia, this research reminds us: balance matters. Tech should amplify thinking, not substitute for it.

For Further Reading

• MIT Media Lab preprint – “Your Brain on ChatGPT: Accumulation of Cognitive Debt…”https://www.media.mit.edu/publications/your-brain-on-chatgpt/

• Time Magazine summary – “ChatGPT May Be Eroding Critical Thinking…” time.com

• Economic Times piece – “Is ChatGPT making us dumb? MIT brain scans…” economictimes.com

• The Register summary – “Brain activity much lower when using AI chatbots…” theregister.com

• CoinTelegraph report – “Need ChatGPT’s help? It could be making you dumber” cointelegraph.com

Self-Studying the MIT Applied Math Curriculum

A Deep Dive into Applied Math – The Self-Taught Way

This summer, I’ve committed to an intellectually ambitious (and deeply personal) project: independently completing the MIT Applied Math Curriculum. I’ll be auditing classes (in person and online), connecting with professors and mentors, and tackling the OpenCourseWare (OCW) assignments and exams wherever available.

In this post, I’ll outline the motivations behind this journey, share a bit about my academic and research background, and lay out the specific plan I’m following. If you're mainly interested in the course list, feel free to skip down to The Plan.

 

Why Take on the MIT Applied Math Curriculum in 2025?

Currently, I’m a dual Master’s student at Harvard and Georgia Tech, specializing in Machine Learning and Computational Biology. I also serve as a research assistant across several labs, where I work at the intersection of Applied Math, Theoretical Neuroscience, and Deep Learning. My previous experience includes co-founding an ML startup. This fall, I’m preparing to apply to PhD programs focused on AI and mathematical biology.

My journey here hasn’t been linear. I originally studied Biology and Cognitive Science at UC San Diego, and my early internships leaned toward Product Management. Along the way, I taught myself to code—initially for web and mobile development—and gradually became captivated by the parallels between information processing in computers and biology. That fascination, particularly with neural networks, led me into graduate programs in CS and Biology, where I began conducting ML research and independently learning the mathematical foundations necessary to interpret cutting-edge papers.

 

As I progressed, something unexpected happened—I fell in love with math itself. Although foundational topics like Linear Algebra, Probability, and Vector Calculus are often sufficient for ML, I’ve noticed that the most innovative thinkers around me draw upon a much broader and deeper mathematical toolkit. They’re able to tap into concepts from PDEs, Dynamical Systems, Numerical Methods, and Optimization Theory, and apply those ideas in novel ways.

After speaking with mentors, including professors and graduate students from Harvard and MIT, the consensus was clear: if I want to push the boundaries of my research and sharpen my mathematical intuition, pursuing a rigorous curriculum like MIT’s Applied Math track is an excellent foundation.

I've been fortunate to gain access to audit select courses, both online and in person, and this self-directed program is my way of organizing, committing to, and sharing that experience. Hopefully, others interested in ML, bioinformatics, or computational theory might find this roadmap helpful too.

The Plan: Courses I’ll Be Studying

MIT’s Applied Math curriculum is thoughtfully designed, blending mathematical theory with real-world applications. I’ve selected a sample path based on the 2024–2025 course listings, emphasizing courses available via MIT OCW, edX, and other open resources, supplemented with similar material at Harvard, Georgia Tech, and nearby institutions.

 

✅ Core Courses (Foundation Building)

• 18.03: Differential Equations

• 18.06: Linear Algebra

• 18.100B: Real Analysis

• 18.330: Introduction to Numerical Analysis

• 18.600: Probability and Random Variables

• 18.700: Linear Algebra (Theoretical)

🧠 Restricted Electives (Applied Focus)

• 18.303: Linear Partial Differential Equations

• 18.330: Numerical Methods for PDEs

• 18.404: Theory of Computation

• 6.041A/B: Probabilistic Systems Analysis

• 6.036: Introduction to Machine Learning (MIT AI curriculum)

• 18.650: Statistics for Applications

   

🔄 Electives and Supplementary Topics

• Optimization Theory (18.335 or equivalent)

• Stochastic Processes

• Information Theory

• Dynamical Systems

• Mathematical Biology (as applicable to my research)

What Comes Next?

I don’t plan to rigidly confine this journey to a summer timeline. While I’ll be actively involved in summer research and class auditing, this project will evolve organically throughout the year. My goal isn’t just to tick off course boxes, but to deeply internalize the material—to make the math intuitive, applicable, and generative for future work.

I’ll be posting updates, study notes, problem solutions, and reflections here periodically. Ideally, this process will lead to:

• More elegant and powerful approaches to ML modeling

• Cross-disciplinary research ideas

• New ways to think about problems in neuroscience and biology

And who knows—once I’ve finished Applied Math, maybe I’ll give the Pure Math curriculum a shot next!

 

Follow Along If you're curious, want to join the ride, or have resources to suggest—drop a comment or connect. I’m excited to see where this self-studying journey leads!