<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Mehul G on Medium]]></title>
        <description><![CDATA[Stories by Mehul G on Medium]]></description>
        <link>https://medium.com/@mehul99001?source=rss-17be598ba9b8------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*VHrZZV6fC3jFRlqvCOnIjw.jpeg</url>
            <title>Stories by Mehul G on Medium</title>
            <link>https://medium.com/@mehul99001?source=rss-17be598ba9b8------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 21:59:00 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@mehul99001/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[If You’re Using AI to Write, You’re Literally Losing Your Brain.]]></title>
            <link>https://medium.com/@mehul99001/if-youre-using-ai-to-write-you-re-literally-losing-your-brain-59fd64705db5?source=rss-17be598ba9b8------2</link>
            <guid isPermaLink="false">https://medium.com/p/59fd64705db5</guid>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[brain]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[vibe-coding]]></category>
            <dc:creator><![CDATA[Mehul G]]></dc:creator>
            <pubDate>Fri, 22 May 2026 01:35:15 GMT</pubDate>
            <atom:updated>2026-05-22T18:17:54.104Z</atom:updated>
            <content:encoded><![CDATA[<h4>Scientists strapped EEG electrodes onto 54 people using ChatGPT. What they found should terrify every student, writer, and knowledge worker alive.</h4><p><em>5 min read · Science · Productivity · Artificial Intelligence</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xOIIanv0uacoQiWi-zihmQ.png" /><figcaption>Representational Image (Gemini AI)</figcaption></figure><blockquote><em>“Participants in the LLM-assisted group mostly failed to provide a simple quote from their own essays when asked about them during follow-up sessions. They had generated the text, but they had not internalized the knowledge.”</em> — MIT Cognitive Atrophy Study, 2025 (under review)</blockquote><p>Let me ask you something uncomfortable.</p><p><strong>When was the last time you wrote something, truly wrote it, without asking AI to clean it up, restructure it, or give you a starting point?</strong></p><p>If you had to think about that for more than three seconds, keep reading.</p><h3>The Study Nobody Is Talking About</h3><p>A team of researchers at MIT ran an experiment that should be front-page news.</p><p>They recruited 54 participants and divided them into three groups over four months, across four separate writing sessions each:</p><ul><li><strong>Group 1:</strong> Brain-only. No tools. Just thinking.</li><li><strong>Group 2:</strong> Web search (Google).</li><li><strong>Group 3:</strong> LLM-assisted (ChatGPT).</li></ul><p>Every session, participants wrote essays. Then the researchers <strong>strapped EEG electrodes to participants’ heads</strong> and measured actual neural activity while they wrote.</p><p>The results weren’t just surprising. They were alarming.</p><h3>Finding #1: Your Brain on ChatGPT Looks Like Your Brain on Autopilot</h3><p>Here’s what the EEG data showed across all three groups:</p><p><strong>Brain connectivity scaled down with the level of external support.</strong></p><blockquote>Brain-only group → Strongest, widest neural networks<br>Search Engine group → Intermediate engagement<br>LLM-assisted group → Weakest overall neural coupling</blockquote><p>The LLM group showed <strong>up to 55% reduced neural connectivity</strong> compared to the Brain-only group, specifically in the low-frequency networks responsible for deep semantic processing, memory formation, and critical thought.</p><p>In simple words: <strong>using ChatGPT to write switched the brain from “creating” mode to “coordinating” mode.</strong></p><p>Instead of lighting up the regions associated with semantic integration and creative ideation (left parietal, right temporal, anterior frontal), LLM users were primarily running their motor coordination and procedural networks. They were basically <em>copy-editing a machine.</em></p><p>The <strong>Search Engine group</strong> was interesting too. They showed increased activity in the visual cortex, because they were actively scanning, evaluating, and selecting information. That’s cognitively demanding. That’s learning.</p><p><strong>LLM users skipped that part entirely.</strong></p><h3>Finding #2: LLM Users Couldn’t Remember What They “Wrote</h3><p>This is the finding that stopped me cold.</p><p>After each writing session, researchers asked participants one simple question: <strong>“Can you quote something from your own essay?”</strong></p><p>The results, broken down by group in Session 1:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/722/1*UNKaoHDjYyBWK1LNX3bC3w.png" /></figure><p><strong>Zero correct quotes from the LLM group in Session 1.</strong> Not a single person could accurately recall something they had just written with AI help, even minutes later.</p><p>By Session 3, the LLM group had improved slightly. But 6 out of 18 participants <em>still</em> couldn’t correctly quote their own work.</p><p>The Brain-only and Search Engine groups? By Session 2, near-perfect recall. By Session 3, <strong>100% quoting ability.</strong></p><p>The EEG data explained exactly why this happened. The theta and alpha brain waves which are the ones responsible for episodic memory consolidation and semantic encoding were dramatically suppressed in LLM users. When you hand ideation to an external system, your brain doesn’t bother encoding the result. It was never <em>yours</em> to remember.</p><blockquote>They didn’t write it. They approved it. There’s a difference their brain knows even when they don’t.</blockquote><h3>Finding #3: You Start to Lose Ownership of Your Own Thinking</h3><p>This one is subtler, but arguably the most disturbing long-term.</p><p>When researchers asked participants <em>“Is this your work?”</em>, here’s what happened:</p><ul><li><strong>Brain-only group:</strong> 16 out of 18 claimed full ownership in Session 1. Rising to 17/18 by Session 3.</li><li><strong>LLM group:</strong> Fragmented and conflicted. Some claimed full ownership. Others explicitly denied it. Many split credit: “maybe 60% mine, 40% the AI’s.”</li><li>The neural correlate? Reduced activity in the anterior frontal regions — specifically the areas responsible for self-evaluation and metacognitive monitoring. The regions that tell you <em>“this is my idea, I thought this, I own this.”</em></li><li><strong>Over-reliance on AI was literally disrupting the sense of cognitive agency.</strong></li><li>And here’s the part that should make you pause: the study also found that human teachers who were not told which essays were written with AI could routinely spot the LLM essays just from their <strong>conventional structure and homogenized points.</strong></li><li>The AI judge they also built? It couldn’t do the same, even after extensive fine-tuning.</li><li>Your authentic voice is more detectable and more valuable than any algorithm currently understands.</li></ul><h4>The Controversy Worth Having</h4><p>Here’s where I’ll say something that might make you close this tab:</p><p><strong>I don’t think AI writing tools are making most people better writers. I think they’re making people better at <em>appearing</em> to write.</strong></p><p>That’s a meaningful distinction that the productivity crowd doesn’t want to hear.</p><p>The MIT study shows this clearly: the LLM group produced essays that were structurally cleaner and packed with named entities, citations, and polished language. By surface metrics, impressive. But the people who wrote those essays couldn’t remember them, didn’t own them, and showed significantly weaker neural engagement throughout the task.</p><p>Meanwhile, the Brain-only group, grinding through the friction of independent thought showed the deepest learning outcomes, the strongest memory retention, and the firmest sense of intellectual ownership.</p><p><strong>Friction isn’t a bug. It’s the feature.</strong></p><p>The struggle of not knowing what to write next, of reaching for an idea that isn’t quite there yet, of revising a sentence six times that’s where your brain is actually building something. That’s where <em>you</em> are being built.</p><h3>What Happens When You Go Cold Turkey</h3><p>The researchers added a clever fourth session. They swapped the groups.</p><p>LLM users were told to write without any tools. Brain-only users were given LLM access for the first time.</p><p><strong>What happened to the LLM group when AI was taken away:</strong></p><ul><li>78% failed to quote anything from their new essay</li><li>Only 11% produced a correct quote</li><li>EEG showed they lacked the fronto-parietal synchronization needed for deep semantic encoding</li></ul><p>Compare that to the Brain-only group given LLM access for the first time:</p><ul><li>Showed <em>increased</em> brain connectivity across all frequency bands</li><li>Performed with stronger top-down regulation and episodic memory access</li><li>Used the AI strategically, like a tool not as a crutch</li></ul><p><strong>The difference? The Brain-only group had built the cognitive infrastructure first.</strong> They had the internal architecture to use AI as an amplifier. The LLM-dependent group had never built that infrastructure, and without the tool, they were cognitively exposed.</p><p>The researchers even have a name for what the LLM group was accumulating: <strong>cognitive debt.</strong></p><blockquote>“Cognitive debt defers mental effort in the short term but results in long-term costs: diminished critical inquiry, increased vulnerability to manipulation, decreased creativity.”</blockquote><h3>What You Should Actually Do (Practical Steps)</h3><p>This is not an article telling you to throw AI in the bin. Used intentionally, AI is extraordinary. But the key word is <em>intentionally</em>.</p><p><strong>→ Write the first draft alone, always.</strong> Before opening any AI tool, spend at least 20 minutes writing your raw ideas unassisted. This activates the exact neural networks the MIT study identified as critical for memory encoding and ownership. It doesn’t have to be good. It has to be <em>yours</em>.</p><p><strong>→ Use AI as an editor, not a ghostwriter.</strong> There’s a massive cognitive difference between “ChatGPT, write me an essay on X” and “Here’s my draft — what am I missing?” The first outsources thinking. The second augments it.</p><p><strong>→ Test your ownership regularly.</strong> After writing anything with AI assistance, close the document and try to summarize in your own words, out loud, what you argued and why. If you can’t, you have a cognitive debt problem worth addressing.</p><p><strong>→ Keep a no-AI writing practice.</strong> Even 15 minutes a day of writing without any AI assistance keeps those frontal and temporal networks active. Think of it like going to the gym while also taking the elevator at work you’re at least maintaining the baseline.</p><p><strong>→ Notice the friction and sit in it.</strong> The moment you reach for AI because writing feels hard is exactly the moment you should resist it longest. That friction is your brain doing its most important work.</p><h3>The Bottom Line</h3><p>54 people. Four months. EEG electrodes. The data doesn’t care about your productivity metrics.</p><p><strong>Using AI as a replacement for your own cognitive effort doesn’t make you smarter. It makes you more dependent, less memorable, and neurologically less engaged with your own ideas.</strong></p><p>The researchers themselves concluded that longitudinal studies are urgently needed before LLMs can be declared “net positive” for the human brain. That’s a striking thing for scientists to say. It means we are running a massive, uncontrolled experiment on human cognition right now — and most of us signed up without reading the terms.</p><p>You can use AI. You should use AI. But use it the way the Brain-to-LLM group did in Session 4: from a position of cognitive strength, with your own thinking built first, using the machine as a force multiplier rather than a replacement.</p><p><strong>Your brain is the asset. Don’t let it atrophy in the name of productivity.</strong></p><p>This article draws on the preprint “<a href="https://www.media.mit.edu/publications/your-brain-on-chatgpt/">Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task</a>” (Nataliya et al., 2025, under review at MIT), the Doshi &amp; Hauser study “<a href="https://pubmed.ncbi.nlm.nih.gov/38996021/">Generative AI enhances individual creativity but reduces the collective diversity of novel content</a>” (Science Advances, 2024).</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=59fd64705db5" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>