AI tutors in schools: The hidden cost of silent classrooms

An AI tutor helped students get right answers but not grasp core concepts, highlighting how AI in schools can silence productive struggle and deeper learning.

In this episode:

• An observation of seventh-grade math students showed AI tutors in schools can help students get right answers without truly understanding core concepts like fractions, raising concerns about AI for deeper learning.
• Shael Polakow-Suransky, president of Bank Street College of Education, argues that AI can strip away 'productive struggle,' a crucial element for students to build their own knowledge, emphasizing the human-centered aspect of the AI in education debate.
• Integrating AI into classrooms could deepen social isolation among teens, mirroring concerns raised by Jonathan Haidt about excessive screen time and the need for more student AI interaction.
• The New York Board of Regents' "portrait of a graduate" framework emphasizes critical thinking, communication, and creative problem-solving, underscoring the need for teacher AI tools that support complex, project-based learning.
• Science teacher Brendan Harney discovered students prefer a real teacher for complex problems, using AI to help students probe assumptions *before* human interaction, illustrating a balanced approach to teacher AI tools.

Chapters:

• 00:00 — Cold open & welcome
• 00:45 — The silent classroom: AI tutors helping, but not teaching, fractions
• 01:45 — The cost of silence: Why productive struggle is essential for deeper learning
• 02:45 — AI tutors in schools: Undermining relationships and the Bank Street approach
• 03:45 — Social implications: Jonathan Haidt's warnings on isolation and student AI interaction
• 04:45 — Systemic issues: How standardized testing influences AI deployment and equity
• 05:45 — A path forward: Designing AI for deeper learning and authentic assessment
• 06:45 — Teacher AI tools: Brendan Harney's strategy for human-in-the-loop AI
• 07:45 — The choice: Amplify teachers or replace them with AI tutors in schools

What are the hidden costs of using AI tutors in schools?
The hidden costs include sacrificing 'productive struggle' essential for deep understanding, reducing vital human interaction, and potentially widening educational equity gaps by providing isolated screen time instead of rich, collaborative learning experiences.

How can AI in education support deeper learning without replacing teachers?
AI can support deeper learning by handling logistical tasks, organizing student drafts, and gathering feedback, which frees teachers to focus on critical capacities like ethical debate, complex problem-solving, and fostering genuine student connections.

What is the primary concern about student AI interaction in the classroom?
The primary concern is that over-reliance on one-to-one AI tutors can lead to social isolation, disrupting the relationships and collaborative interactions that are fundamental to how children learn and develop, and which AI cannot replicate.

Featuring: Dan Fitzpatrick, Shael Polakow-Suransky, Bank Street College of Education, Mary Helen Immordino-Yang, Jonathan Haidt, Fannie Lou Hamer Freedom High School, New York Performance Standards Consortium, New York Board of Regents.

Follow AI in Education with Dan Fitzpatrick for more on AI in education.

Transcript

SPEAKER_00 0:00

If this episode makes you think, please let us know in the comments and support us by subscribing and leaving a review. Thank you. Today we are exploring the critical role of human interaction in learning, particularly in the context of emerging AI tools, drawing insights from an illuminating guest essay in the New York Times. It's titled The High Cost of Silent Classrooms, published on june eighth, twenty twenty six and written by Shile Polakov Saransky, who is the president of the Bank Street College of Education. What immediately jumped out at me and what I think acts as our cold open hook today is the stark image Shale paints right at the beginning of his piece. He recounts visiting a seventh grade math classroom in the Bronx, where twenty students sat bent over laptops, working in silence with an AI tutor on story problems about fractions. Now, on the surface, he notes, the classroom looked like it was working. Students were engaged and eventually most of them got to the right answers. But here's where it gets interesting, and frankly a bit unsettling. When he looked closer, Shales saw that many of those students were actually quite lost. They didn't grasp the core concept of fractions. Each time a student made a mistake, the AI tutor would offer another step, but it never actually identified why the understanding had broken down, and the teacher, her dashboard while showing who was stuck, couldn't tell her what the underlying gap was either. This really resonates with one of my core philosophies, that AI is about enhancement, not replacement. What the article suggests is that in this scenario, the AI was effectively replacing a crucial human function, not enhancing it. What Shil Polakov Saransky argues is that the core intellectual work of teaching is precisely this, noticing why a child's understanding breaks down, and they're knowing what to do about it. That might mean pausing for a quick mini lesson or bringing out some physical fraction tiles for a student who needs to really see the math. In that Bronx classroom that essential work had been handed over to a tool that simply couldn't do it. There was no lively discussion, no one turning to a classmate to ask, wait, how did you get that? Just individual kids silent in front of a screen clicking away. The cost of this silence, he argues, is both cognitive and social. When artificial intelligence tries to anticipate every single step before a student even recognizes a hurdle, it strips away what I often call the productive struggle that learning truly depends on. Students absolutely need to wrestle with confusion to build their own understanding. The neuroscientist Mary Helen Immordino Young and her colleagues have shown that this kind of deep learning, the kind that genuinely sticks, happens when students connect what they're learning to bigger ideas and to their own lives. If we replace that vital dialogue and productive struggle with isolated screen time, we risk disrupting the very neural circuits that allow students to truly build knowledge. This is a powerful warning about the dangers of outsourcing your thinking. While the AI was handling the doing of providing the next step, it wasn't fostering the thinking that students needed to develop conceptual understanding. The real value is not just in what the machine produces but in how the student responds and critically understands. The article makes the case that this trend of one-to-one AI tutors is backed by significant investment, and while early research might show some gains in procedural skills, efficiency isn't the same as true understanding. And this brings me to a crucial point about AI tutors in schools and the AI in education debate. When we isolate a student with an AI tutor, we cut them off from the relationships that fundamentally drive learning. Bank Street College of Education, where Shayel Polakov Saransky is president, trains teachers using a developmental interaction approach, which rightly recognizes that children learn best within the context of trusting relationships. A clearer explanation from a bot is rarely enough when a student is really struggling. A child needs to hear another student explain it, to argue back, to figure out where they genuinely disagree. And crucially they need a teacher who knows them, who can tell if a student is lost or bored, or can read their hesitation and know whether it's a language barrier or the normal fumbling right before a breakthrough. Machines can compute, they cannot wonder, they cannot care, they cannot judge a child's nuanced emotional state. The essay then really dives into the social implications, suggesting that doubling down on isolation is truly dangerous. Jonathan Haidt, as Sheil notes, has warned us about a collapse in teen mental health, driven by a rewiring of childhood that's replaced play and community with excessive screen time. If schools widely embrace one-to-one AI tutoring as the norm, it could deepen that crisis, exchanging the in-person interactions children desperately need for yet another screen. This is a vital part of the AI and education debate that we need to be clear-eyed about. Now this isn't just a critique of AI itself, but a powerful commentary on the system in which AI is being deployed. The article suggests that for a generation, American schools have been shaped by standardized tests that measure a very narrow band of skills. Because of the high stakes involved, teachers often end up teaching to the test, narrowing the curriculum and reducing time for things like projects, argument and real problem solving. In this context, the AI tutor drilling concepts a seventh grader doesn't understand isn't an aberration, it's a logical extension of that system. This brings up a concern about equity. The article warns that AI risks becoming a new instrument for educational segregation. In wealthier districts, parents might demand schools centered on rich human interaction, seminar tables, heated debates, messy collaborative projects. Meanwhile, students in poor schools, often black and Latino children, as the piece highlights, could be handed laptops and headphones, learning from machines that might correct their algebra, but will never truly care about their curiosity or foster their creativity. This is a critical point for us to address as educators. We have to ensure that AI serves as an equalizer, not a tool that widens existing gaps, especially for the often invisible middle eighty percent of students who need thoughtful support. Accessibility is not an afterthought but a foundation, but Shayel Polakov Saransky doesn't just raise concerns, he offers a path forward, making it clear that we don't have to choose between haphazardly embracing AI and banning it outright. He says that in a world AI is already reshaping, what students need to learn is changing. Content knowledge will always matter, but it's no longer enough. Students must also become original thinkers who can reframe problems, engage citizens who can grapple with power and democracy, and generous collaborators who can work across real differences. An AI tutor can help a student memorize a formula, but it cannot teach her to debate its ethics with a peer who profoundly disagrees. The distinction is clear, outsource the doing not the thinking. So a guiding principle for our work with AI in schools, he argues, should be whether it truly supports this kind of deeper learning. He gives an example, Keisha, a graduate of Fannie Lou Hamer Freedom High School in the Bronx. For her social studies portfolio, Keisha spent weeks researching how federal housing policy and the GI Bill shaped the racial wealth gap in America. After many drafts, she defended her paper in front of a panel of teachers who questioned her closely. She could point to specific passages in her sources and confidently explain her reasoning. This kind of authentic performance-based assessment, which is common in schools within the New York Performance Standards Consortium, demands depth, care and imagination, making it learning that simply cannot be faked. This is exactly the kind of assessment in the AI era that we should be designing for. The good news is that this isn't just hypothetical. The New York Board of Regents has even approved a new portrait of a graduate framework, signalling a shift away from defining readiness only through standardized exams and towards critical capacities like critical thinking, communication, and creative problem solving. But here's the kicker. Without teacher AI tools that actually make these longer projects and inquiry-based learning workable for teachers, this reform risks collapsing back into something easier to measure. This is where AI used wisely can become a force for good, helping us hold the complexity so we have capacity for creativity. Imagine a tool that organizes a student's drafts across a year, gathers peer feedback, and shows a teacher how an argument tightened over time. That kind of functionality could empower teachers to give this deeper attention to more students. For more insights on how AI is shaped in education, follow me Dan Fitzpatrick, wherever you listen to your podcasts. The piece then provides a fantastic concrete example of a teacher who is already navigating this space effectively. Brendan Harney, a science teacher at the Bronx Lab School, learned the hard way what AI in classrooms should not do. With seventy students each designing their own experiments, he hoped AI could shoulder some of the coaching. He built a tool in imitation of his own voice and tone. An Alexa for the classroom, as he put it, but his students pushed back. They told him they wanted to walk through a hard problem with a real teacher, not a machine. So he rebuilt the tool with a much smaller, more focused job. It now helps students probe their assumptions before they sit down with him to talk about their experiments. This is a masterclass in the human in the loop principle and purpose over technology. When one student proposed testing how caffeine affects memory, the AI helped him think through quantities and measurements. But Mr Harney asked the harder question, whether it was ethical to make another student ingest four hundred milligrams of caffeine for a class experiment. That's judgment, that's ethics. That's what AI cannot do. Teachers like Brendan Harney are doing the vital work of figuring out where AI truly helps and where only a teacher will do. This is the essence of thinking with AI, not just using tools. The seventh graders Shayel Polakov Saransky mentioned at the outset don't need a better AI tutor in the sense of one that just gives smarter prompts. They need a teacher who knows them well and has the capacity to help them work through what confuses them until it truly clicks. The silence of that Bronx classroom is, as the article concludes, a powerful warning. Real learning requires friction, debate, and genuine connection. If we use AI wisely, it could actually give teachers more time to do that incredibly valuable work with their students. The choice between a classroom that merely drills a child and one that genuinely teaches her to think should never be determined by zip code. We can build technology that amplifies what teachers do best, or we can sleepwalk our way into letting it replace them. The decisions we make now about AI tutors in schools and student AI interaction will shape more than how students learn. They will shape who they become. Let's design learning that cannot be faked because it demands depth, care, and imagination. That's all for today. Thanks for listening.