An ML Engineer and Professor on AI, Bias, and What's Really Happening in the Classroom

Practice the Way You'll Play

In this episode we explore what classrooms can learn from athletics and the performing arts about practice, performance, and real learning in the age of AI. Topics covered: * Why exams often reveal gaps between AI-assisted performance and independent learning * What music, athletics, and theater can teach academic classrooms * Transfer-appropriate processing and why practice conditions matter * The difference between practicing for a test and building independent capacity * How AI can support learning without substituting for student thinking Sources: https://www.semanticscholar.org/paper/Levels-of-processing-versus-transfer-appropriate-Morris-Bransford/838410627202df92a6cf7b45277e1368aefbe98a https://www.sarahshihuiwong.com/post/think-first-chatgpt-later-wong-qiu-2026-educational-psychology-review\ https://www.dailycal.org/news/campus/academics/failing-grades-soar-as-professors-see-greater-ai-usage-dwindling-math-skills-in-uc-berkeley/article_16fad0bf-02cb-4b8c-8d88-888ffd9f8608.html

The AI Problem Teachers Can’t Always See

In this episode we explore what happens when students appear to be participating, but may actually be outsourcing their thinking to AI. We look at a recent UC Berkeley story about unusually high failure rates in introductory computer science courses, alongside reporting from The New York Times about students using tools like Gemini and ChatGPT during class discussion.  Topics covered: * How AI can create a gap between visible participation and actual thinking * What the UC Berkeley computer science failure-rate story may reveal about AI-assisted learning * The difference between completing work with AI and developing transferable skill * Why “cheating” may be too narrow a frame for what schools are seeing * What classroom teachers can look for when students appear engaged * The risk of discovering learning gaps only when students sit for an assessment Sources: https://www.dailycal.org/news/campus/academics/failing-grades-soar-as-professors-see-greater-ai-usage-dwindling-math-skills-in-uc-berkeley/article_16fad0bf-02cb-4b8c-8d88-888ffd9f8608.html https://www.nytimes.com/2026/06/17/podcasts/the-daily/battle-over-ai-in-school.html

An ML Engineer and Professor on AI, Bias, and What's Really Happening in the Classroom

In this episode we explore what AI looks like from the inside: not as a shiny classroom tool, but as something real teams have to build, test, constrain, and question before it ever reaches teachers or students. We talk with Jon Landrigan, about what he is seeing from today’s college students as they learn, work, and think alongside AI. Topics covered: * How college students are using AI today * How AI tools move from prototype to production in education * Why a tool that works once may not work safely or fairly at scale * The risks of “vibe coding” for schools and district leaders * Prompt injection, edge cases, and other safeguards behind AI products * AI feedback, writing assessment, and the limits of automated grading * Automation bias and why teachers may over-trust AI recommendations * Personalization, learner profiles, and the danger of biased metadata * Cognitive offloading, student pressure, and what schools still need students to internalize * Resources for educators who want to better understand AI systems

Was This AI Tool Built for Learning or Just for Work?

In this episode we explore the question: were the AI tools entering classrooms actually designed for learning? We look at a May 2026 preprint from Hassan Khosravi, Ryan Baker, and colleagues that argues many AI tools now being used in schools were originally built for work, not education. Topics covered: * How does the tool handle cognitive effort? * What does the tool do when students make errors? * What does the tool measure as success? * Does the tool remember anything meaningful about the learner over time? * Why tools that look effective during practice may fail when students must work without them * What Khanmigo’s redesign can teach us about building AI around actual learner needs * Why procurement conversations should focus less on marketing claims and more on learning design Source: https://arxiv.org/pdf/2605.04816

Are Students Cheating With AI, or Are Assignments Unclear?

In this episode we explore why student perceptions of AI cheating are far less settled than many school policies assume. We look at new survey data from Oxford University Press showing that students draw very different lines around AI use, from seeing any AI support as off-limits to not viewing full AI completion of homework as cheating.  Topics covered: * Why students disagree about what counts as AI cheating * The connection between AI use, assignment purpose, and “fidelity” * Why AI detection tools are unreliable for high-stakes decisions * How teachers can clarify when independent thinking is the goal * What instructional coaches should look for when reviewing assignments * Why district AI policies need to explain the instructional “why,” not just the rule * How schools can shift from catching misuse to designing clearer learning conditions Sources: https://fdslive.oup.com/www.oup.com/oxed/research-reports/Navigating_AI_in_Education_Research_Report_June2026.pdf?region=uk https://www.sciencedirect.com/science/article/abs/pii/S0360131526000540