Can the Law Hold AI Accountable? - Tiffany Brown

If AI Writes, Who Thinks? - Jane Rosenzweig

In this episode, Priten speaks with Jane Rosenzweig, director of the Harvard College Writing Center and lecturer in expository writing, about teaching writing in the age of AI. Jane's first-year course, To What Problem Is ChatGPT the Solution?, asks students to study artificial intelligence without outsourcing the work of thinking to it. They discuss why writing is inseparable from thinking, what students lose when they skip the struggle of drafting, and why feedback is a conversation rather than a product. Key Takeaways: * Writing is thinking, not output. The point of a writing course is not to produce more papers in the world. It is to give students the experience of working through evidence, weighing ideas, and figuring out what they actually believe. * Editing skills are not a substitute for drafting. The argument that students can skip the first draft and learn to polish AI output assumes a skill that develops only through drafting. Jane has not seen evidence that students who never write a first draft can revise their way to something meaningful. * Feedback is relational. A writing tutor often does not know where the paper will end up, and that shared uncertainty is the point. A chatbot can work on what is already on the page, but it cannot build a bridge to the idea a student has not yet had. * Feedback on demand undermines productive struggle. When students can revise and resubmit to a chatbot at 1 a.m., the friction that makes them reconsider what they think disappears. The decision to skip that friction is being made for reasons other than learning. * Integrating AI into every course is not a solution. Students can distinguish between AI uses designed to push their thinking and how they will actually reach for the tool under a deadline. Teaching productive uses does not prevent the unproductive ones. * The deeper challenge is equity, not just pedagogy. A real risk is that students at well-resourced institutions continue to learn how to think while students elsewhere have their instructors replaced with chatbots. Aligning incentives so grades and learning point in the same direction is the work ahead.

Can the Law Hold AI Accountable? - Tiffany Brown

In this episode, Priten speaks with Tiffany Brown, litigation counsel at Tech Justice Law, about what accountability looks like when AI products cause real harm. They discuss the wave of product liability lawsuits filed against ChatGPT, why disclaimers and "for entertainment purposes only" language do not insulate companies from responsibility, and how courts are beginning to treat generative AI as a defective product. The conversation also moves into civil rights enforcement, state versus federal action, and the new legal questions raised by autonomous agents. Key Takeaways: * Generative AI is being litigated as a defective product. Tech Justice Law has filed cases tying ChatGPT to suicides, suicide attempts driven by AI delusions, and even a school shooting in Canada. The legal theory treats the chatbot itself as a product whose harms were foreseeable and whose deployment was negligent. * Foreseeability is doing a lot of the work. A book that contributes to a mental health crisis is hard to litigate; a chatbot designed to mimic human emotion and used by a 12-year-old is not. When a company knows or should have known that a product can cause specific harms, the law has tools to respond. * Disclaimers do not erase liability. A "this may hallucinate" warning, or Copilot's "for entertainment purposes only" terms, do not get a company out from under strict product liability when people are losing their lives. Courts will ask whether the company did enough, not whether it checked a box. * States are doing the work Congress is not. State attorneys general are opening investigations, state legislatures are passing AI-specific laws, and California recently moved to block the "the agent did it" defense. Federal action is unlikely in the next two to three years. * The harms cut across demographics. Unlike the social media cases, which centered on minors, AI chatbot cases involve children, older adults, people with disabilities, and even tech-savvy users. The speed and scale of impact is what makes generative AI different. * Agentic AI raises the stakes again. When a single company can deploy 200 autonomous agents instead of one rogue employee, the scale of potential harm changes the legal calculus. Insurance products are emerging, but Tiffany is skeptical that liability can be outsourced to the agent itself.

Who Is Protecting Student Privacy Right Now? - Cody Venzke

In this episode, Priten speaks with Cody Venzke, senior staff attorney with the ACLU's Speech, Privacy, and Technology Project, about who is actually protecting student privacy when the law has not caught up to the technology. They walk through what FERPA and COPPA do and don't cover, the limits of "FERPA compliant" as a marketing claim, how AI surveillance tools are being deployed in schools without adequate vetting, and where parents and teachers can apply pressure when federal law leaves gaps. Key Takeaways: * FERPA was written for filing cabinets, not cloud platforms. Passed in 1974, FERPA still grants parents a right to access every record a school maintains about their child, including data held by ed tech vendors. But it has never been enforced by the Department of Education, and individuals cannot sue under it, which leaves most of the work to proactive parents. * "FERPA compliant" on a vendor website is a marketing slogan. There is no Department of Education certification program. The obligation falls on schools to ensure their vendors actually limit data use to educational purposes, and parents should ask schools how they define "school official" and what contracts allow. * COPPA stops at the thirteenth birthday. The Children's Online Privacy Protection Act applies only to sites directed at children under 13, leaving teenagers in what Venzke describes as a regulatory wild west. The ACLU argues that data minimization and affirmative consent should be extended to everyone, not gated by age. * Flat bans on minors using social media will likely lose in court. The Supreme Court has held that minors' First Amendment rights are largely coterminous with adults'. Venzke predicts that age-based bans will be struck down as overbroad, and argues that regulating how platforms collect and use data is a more constitutionally durable approach than restricting speech. * School AI surveillance is being deployed without testing. Facial recognition, weapons detection, and communication monitoring tools are sold to schools without proof they work as advertised. Venzke cites cases where students have been outed by large language models that misread diary entries as bullying, and argues that high-impact AI uses should require state-level vetting requirements. * Removing a student's name from a ChatGPT prompt does not make it FERPA safe. Identifying details like "the only Native American student in fifth grade" can still trace back to an individual. Venzke argues teachers should not be left to vet AI tools on their own; districts, states, and procurement processes need to do that work.

How Might Schools Make Sustainable AI Policies? - Joel Sohn

In this episode, Priten speaks with Joel Sohn, Deputy Head of School at Head-Royce, a K-12 independent school in Oakland serving roughly 920 students, about how a school can build a coherent approach to AI without retreating into a rulebook. Joel walks through the two-year arc of arriving in fall 2023, identifying early teacher champions, taking them to the Schools of the Future Conference, and using Leon Furze's framework to land a philosophy statement rather than a granular policy. The conversation covers why originality has always been a puzzle, how students have shifted from experimenters to skeptics, and why a simplified nine-word mission is doing more work than any rulebook could. Key Takeaways: * Build a philosophy, not a plagiarism policy. Joel draws an analogy to dress codes: the more granular the rule, the more the only thing you see is the violation, not the person. AI use is too varied across math, history, and English classrooms to codify the way schools codified plagiarism a generation ago, and a philosophy gives educators the room to make case-by-case judgments. * Trust the team first, accelerate later. Joel chose a two-to-three year change trajectory anchored in building educator trust rather than racing to be first. His worry was falling behind by 2027, but the trust groundwork is what made the eventual rollout move quickly and made families comfortable with the rollout. * Originality has always been a puzzle, and AI just forces the question. Joel pushes back on the assumption that pre-AI student writing was somehow more "original," pointing out that Shakespeare cribbed too and that brain science still cannot pin down what original thought really is. Schools have been asserting certainty they never had, and AI is making that hard to avoid. * Students are no longer the experimenters they were two years ago. Joel sees the current generation as more anti-AI than in 2023, citing concerns about energy use, corporate ethics, and privacy. Teachers using AI sloppily and shipping obviously machine-generated lessons has accelerated that skepticism, which is why he tells teachers to disclose their AI use and how they checked it. * Strident anti-AI students need to be interrogated too, not just validated. Joel argues schools should push back when students refuse to engage with AI, not to override their values but to ask whether their stance is rooted in privilege, fear, or genuine principle. The work of school is teaching kids to handle complexity, not to handle any specific tool.

What Does Faithful Teaching Look Like in the Age of AI? - Chuck Parish

In this episode, Priten speaks with Chuck Parish, an English teacher at a private Christian school in El Paso, about what it looks like to build an AI elective from scratch inside a community that is still deciding whether to be afraid of the technology or learn it. Chuck's path runs through pastoral ministry, teaching at the bachelor's level in Papua New Guinea, and a year of sixth grade before landing in high school English. The conversation moves between the practical questions he is sorting out for his fall semester and the deeper one he keeps returning to: whether schools are forming the kind of judgment students need to use powerful tools well, or whether they are only writing policies. Key Takeaways: * Policy can legislate behavior but it cannot form character. Chuck argues that a clean ban or a strict acceptable-use document is the easy move, and the wrong one. Without a foundation underneath it, students will either ignore the rule or comply for the wrong reasons. The school's one-sentence AI policy treats the question as plagiarism, which misses most of what the technology actually changes. * A Christian worldview has to address AI the same way it addresses every new tool. The Bible does not name AI any more than it names calculators or television, so the work is in applying an existing foundation to a new technology. Chuck wants students to be able to reason from that foundation themselves rather than relying on him to legislate each case, especially because they will leave the school and lose the legislator. * Writing instruction was already in trouble before AI arrived. Texting has shifted how students communicate so far that sixth graders submitting "OMG" and "TY" in their papers is no longer surprising. AI does not start the decline in written reasoning; it accelerates a slide that started with the way students already talk to each other. Chuck plans to use handwritten baseline essays to anchor what each student can actually do without help. * Demonstrating the tool in class is more honest than hiding it. Chuck plans to put ChatGPT on the classroom screen, show how fast it can produce an essay, walk through prompting, and surface the hallucinations and fabricated citations directly. The argument to students is that cutting and pasting cheats them out of the learning, and that integrity has to be taught, not assumed. * The AI conversation has to include companions and cyberbullying, not just essays. Chuck wants the elective to cover Replika-style companions and image-manipulation tools alongside academic use, because those are the parts students are already encountering outside class. Putting head in the sand, especially in a Christian school context, leaves students to form a worldview about these tools on their own and usually badly.