AI Does Something Horrifying To Human Thinking

This video argues that the viral panic over students losing reading ability misses a deeper concern: outsourcing thinking to AI measurably weakens human cognition. Citing MIT brain-imaging studies, the speaker shows that ChatGPT users exhibited reduced neural connectivity, often couldn't recall their own essays, and suffered "cognitive debt" that persisted even after the tool was removed. Overreliance on AI—cognitive offloading—is linked to diminished critical thinking across adults and children, with the greatest risk for novices and developing minds who may never build independent reasoning. The video also highlights real-world interventions showing that removing tools can rebuild skills, and warns that AI flattens thinking into homogenized patterns, prompting global regulatory moves like phone bans in schools. Rather than banning AI or allowing unrestricted use, the speaker advocates purpose-built educational AI, such as a Socratic-method tutor that challenges students to argue and edit output, thereby strengthening their thinking. The conclusion: AI can degrade or improve cognition depending on intentional design, and society must act deliberately before harm accumulates.

• ▶ 0:00 The viral claim that "college students are rapidly losing the ability to read" triggers predictable reactions (AI panic, moral panic dismissal, sarcasm), but all of them skip what actually matters: closely reading the research.
• ▶ 0:32 The evidence base includes MIT brain imaging, Carnegie Mellon surveys, large-scale cognitive assessments, and counterpoint studies showing AI can improve learning—revealing a science far more nuanced, interesting, and important than any headline.
• ▶ 0:54 This is not just about students: the real story is what happens to human cognition when we outsource thinking to machines, with implications reaching well beyond the classroom.
• ▶ 2:34 MIT's Your Brain on ChatGPT study used EEG on 54 adults writing SAT essays, comparing ChatGPT, Google search, and no-tool groups.
• ▶ 3:01 ChatGPT users showed the weakest neural connectivity and 83% couldn't quote a line from their own essay — total cognitive amnesia.
• ▶ 4:11 Key finding: when AI users later wrote without AI, their brain activity did not return to normal — cognitive debt persisted even after the tool was removed.
• ▶ 4:31 A 2025 peer-reviewed study with 319 knowledge workers found that the more workers trusted AI's ability, the less critical thinking they applied—warning of long-term overreliance and diminished independent problem-solving.
• ▶ 5:20 A 666-participant study found frequent AI use negatively correlated with critical thinking, mediated by cognitive offloading—letting AI do the thinking—which showed an even stronger link to decline.
• ▶ 5:54 A 2026 preprint with software developers showed full AI delegation produced working code but 17% worse conceptual quiz scores, and the risk is greatest for novices with no existing knowledge to evaluate AI output.
• ▶ 7:20 The problem isn't just students: research on adults (MIT, Carnegie Mellon, cognitive offloading) shows AI reliance is a universal "use it or lose it" phenomenon.
• ▶ 8:25 AI didn't single-handedly destroy critical thinking—NAEP and collegiate data show it was already fragile—but that makes the risk more urgent, not less, since an easy escape hatch worsens an existing weakness.
• ▶ 10:00 Real-world evidence (the Minneapolis teacher's tech-free class, Wartburg phone-ban study) shows removing tools can rebuild cognitive skills and confidence, yet the section warns the goal isn't permanent removal but rebuilding the muscle.
• ▶ 10:53 Children face a unique risk: unlike adults who lose existing cognitive abilities by offloading thinking to AI, children may never build those capacities at all, since AI replaces the development of reasoning entirely.
• ▶ 11:14 AI is flattening classrooms into cognitive homogenization—students process information through the same model, making their reasoning, defaults, and even language sound identical, risking a generation shaped by one AI's biases.
• ▶ 12:49 Global regulatory pressure is rising—the UK banned social media for under-16s, 20 U.S. states ban phones in schools, and several European countries are following—because AI captures thinking itself, not just attention, while students themselves report growing concern (70% worry AI erodes critical thinking).
• ▶ 14:17 The speaker rejects both banning AI and unrestricted access, instead advocating for purpose-built educational AI applications rather than general-purpose chatbots.
• ▶ 16:06 A key example is a Socratic-method educational AI that questions and challenges students—not handing over answers—so they do the thinking themselves.
• ▶ 17:26 Educational AI must adapt to different learning styles and force critical engagement, because research shows that arguing with and editing AI output improves learning outcomes.
• ▶ 17:55 The tech and research base to make AI improve thinking already exists; what's missing is priority and aligned incentives, not capability.
• ▶ 19:17 AI is measurably changing how brains engage with information: indiscriminate use carries real cognitive costs, especially for developing minds, but critical and active use can make people better thinkers.
• ▶ 20:40 The social media analogy warns that potential without intentionality leads to harm; society is earlier on the same curve with AI and must decide whether to act now or wait for damage to accumulate.

• ▶ 0:00 Introduction: Headlines vs. the Evidence (0:00 - 2:23) - - The viral claim is too simple; the speaker sets up the studies and roadmap.
• ▶ 2:23 MIT Study, Caveats, and Cognitive Debt (2:23 - 4:31) - - The famous preprint finds persistent cognitive debt after AI use, though it has notable caveats.
• ▶ 4:31 Broader Studies on AI and Critical Thinking (4:31 - 7:20) - - Larger studies confirm the negative correlation, while a pro-AI study was retracted and critical engagement helps.
• ▶ 7:20 Interpretations and the Human Side (7:20 - 10:46) - - This affects everyone, using AI to cope is understandable, but cognitive atrophy remains a real risk.
• ▶ 10:46 Children, Society, and Global Concern (10:46 - 14:17) - - Children are uniquely at risk, class discussions flatten, and bans/student worries show rising unease.
• ▶ 14:17 A Better Path: Purpose-Built Educational AI (14:17 - 17:55) - - Not bans or unrestricted use; specialized AI should keep students actively thinking and adapt to diverse learners.
• ▶ 17:55 Incentives, Research, and the Real Takeaway (17:55 - 21:48) - - Missing incentives and rigorous longitudinal studies are urgent; the evidence shows both costs and benefits.