Your Undivided Attention · The Center for Humane Technology, Tristan Harris, Daniel Barcay and Aza Raskin

AI Is Breaking Education. Rebecca Winthrop Has the Blueprint to Fix It.

March 5, 2026·46 min·4 clips

Half of teachers and students don't trust each other because of secret AI use in classrooms.

Rebecca Winthrop leads the Center for Universal Education at Brookings Institution and recently completed a comprehensive "premortem" study examining AI's role in education. The research, spanning 200 pages, involved interviews with educators, students, parents, policymakers, and technologists globally to assess whether current AI implementation trajectories serve student learning. Winthrop's team applied lessons from the social media rollout, noting that child development experts were excluded from early social media design decisions despite known harms like social comparison effects on adolescents. Current AI use in classrooms is fragmented: students access AI through social media, companion apps, search engines, and educational technology both in and outside school, while teachers use AI for lesson preparation, grading, and assessment. Students commonly use AI to write essays, run them through "AI humanizers" to avoid detection, and submit work teachers cannot verify as authentic. Teachers use AI tools without transparency, and parents run student assignments through ChatGPT to challenge teacher grading decisions. The research uncovered a critical breakdown in classroom trust: 50% of teachers distrust that student submissions represent authentic work, while 50% of students distrust teachers who secretly use AI for lessons and grading. Students interpret even well-intentioned AI use by teachers—like providing feedback before final submission—as lack of care. Some college students now trust ChatGPT's authority over their professors. Detection software for AI-generated work generates false positives, disproportionately accusing neurodivergent, learning-disabled, and multilingual students of cheating. The most significant risk Winthrop identifies is "cognitive stunting"—AI preventing students from developing independent thinking skills rather than merely offloading existing capabilities. When students use chatbots to write essays, they bypass developing skills in evidence evaluation, argumentation, and persuasive writing. Students themselves identify "becoming dumber" as their primary AI concern, not job market displacement—a finding confirmed by recent Comic Relief surveys of young adults globally. Students report becoming unable to initiate homework independently after regular AI use. Winthrop rejects the calculator comparison, noting calculators only offloaded arithmetic while AI handles English homework, coding, art, music, and provides persuasive conversational interaction. AI's general-purpose nature and "seductive" quality fundamentally distinguishes it from previous educational technologies. Learning requires social relationships, physical embodiment, and the ability to receive critical feedback and make mistakes—skills undermined by "sycophantic" AI companions that always affirm student work. Beneficial AI applications include teachers using AI for administrative work and differentiated reading materials, students using VR with AI for brief interactive biology lessons, and tutors using AI to identify student misunderstandings. A Hawaii middle school successfully integrates AI by teaching machine learning and data science while students analyze real sea-level data outdoors, using AI as an analytical tool rather than a replacement for thinking. Winthrop recommends eliminating one-to-one laptops in elementary and possibly middle school, having students "red team" teacher assignments to identify AI vulnerabilities, and prioritizing in-class exams and oral presentations. The conversation balances technical analysis with accessible explanation, using vivid examples like the Ivy League student choosing a C grade over using AI and the sixth-grader who wrote "it might end the world" when asked about AI concerns. Winthrop adopts a measured, evidence-based tone while acknowledging genuine risks without catastrophizing. The discussion targets educators, administrators, parents, and policymakers seeking practical guidance rather than abstract theorizing about AI's educational future.

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from Your Undivided Attention