MIT’s Justin Reich interviewed 120 teachers and students about AI in the classroom. His finding: there are no research-based best practices yet — so run your own small experiments. Hear what to add, what to subtract, and what to try this week on Episode 939.
Happy Motivational Monday, friends. Today will make you think as we talk to my friend Justin Reich from MIT. In a June 2026 NPR/Ipsos poll, nearly three out of four teachers said they believe AI will have a bigger impact on education than the internet or the computer ever did. More than half said it is making it harder for students to learn and think for themselves.
Sponsor. Today’s show is sponsored by EF Explore America and their STEM Tours. Lead your students on a STEM tour to places on the cutting edge of innovation to show them how STEM thinking often shows up where you least expect it. Imagine your students coding robots with MassRobotics at MIT, exploring marine ecosystems in Florida’s coral reefs, or even sitting down to talk with a former spy in Washington DC. If you want to inspire your students and give them a fresh perspective on the power of STEM, visit efexploreamerica.com/STEM.
This is a staggering amount of MORE pressure to put on a profession that is reeling from the pressure of expectations, too many standards (research on a “guaranteed and viable curriculum” shows we can’t meaningfully pursue too many at once — Robert Marzano calculated that teaching every standard well would take a K–22 school system), decreasing attention spans, and a justification of doing wrong that just seems to come out of everywhere. Justin Reich — an associate professor of digital media at MIT and director of the Teaching Systems Lab — decided to talk to teachers and students about what AI is doing in real classrooms. With 120 interviews, he is pulling together what teachers and students are actually saying.
Justin came to talk about The Homework Machine, a new limited-series podcast built from roughly 120 interviews with teachers and students across the country. We also talked about why he trusts classroom voices over thought leaders. (Justin visited my classroom around 20 years ago for that very reason — to see what was actually happening.) But perhaps the most profound piece is that the research cannot hand any of us “best practices” yet. He gives a rock-solid alternative while we’re waiting for the research to be done: local science.
He also argues a very counterintuitive argument about domain knowledge versus just teaching kids “to use AI.” He says make them experts in their subject areas so they can then use AI. Wow. And then, finally, he talks about how he thinks successful schools will focus on subtraction, not addition — subtracting things from teachers’ plates and pruning what isn’t working.
Justin always teaches me so much as he bridges the real classroom and research and helps me see things differently. I do know this — we’re in such a rush to find “best practices” that research studies that haven’t been published, haven’t been peer reviewed, and haven’t even really been well read (can you say AI summaries?) are going viral. And then, in some cases, they are being retracted. The clearest example: a viral MIT paper claiming AI supercharged scientific discovery drew praise from a Nobel laureate — until MIT itself said it had “no confidence” in the data and asked for it to be pulled from arXiv. This is a thoughtful conversation I think we all need to hear.
Listen to the Show
AI in the Classroom — Why There Are No Best Practices Yet
Watch on YouTube and subscribe for new episodes every week.
Key Takeaways for Teachers from Justin Reich
- Trust the people closest to the classroom. Justin’s whole reason for 120 interviews: “Professors and thought leaders can think whatever they want, but the most important observations are the ones from the people who are closest to what’s actually happening.” Relate to educate — your view from the desk matters more than the view from the think tank.
- AI lands differently in every school. In communities with no substitutes and high chronic absenteeism, AI is “the fifth, twelfth thing on people’s lists.” In more affluent schools it’s the number-one concern. There is no single AI story — and pretending there is one is how policy goes wrong.
- There are no research-based best practices yet — and that’s the honest answer. It took about 25 years — from the early search engines of the mid-1990s to 2019 — for solid research to tell us how to teach kids to sort fact from fiction online. Big science takes decades, not years. Anyone selling you AI “best practices” today is ahead of the evidence.
- Do local science instead. Tell students and parents “this is an experiment, there are no best practices yet,” try one AI-enhanced approach, then compare the evidence — like grading this year’s speeches against the ones you got before AI. Keep what works, throw away what homogenizes student voice. Innovate like a turtle: small, deliberate, one trial at a time.
- The power of less: ask what to subtract. Schools have 180 days and seven hours a day — “it’s actually not that much time.” The one thing red states, blue states, public and private schools all agreed to cut was cell phones. Justin’s challenge for every PD cycle: what can we stop doing? “Finding what to prune is the way that you get your best stuff to grow.”
Resources Mentioned in This Episode
- The Homework Machine — Justin’s limited-series podcast on what AI is really doing in K-12, built from roughly 120 interviews with teachers and students. Listen at teachlabpodcast.com.
- MIT Teaching Systems Lab — Justin’s research home for teacher experimentation and edtech research: tsl.mit.edu.
- “The Evidence Base on AI in K-12: A 2026 Review” (Stanford SCALE Initiative) — the research Vicki referenced: of 800+ studies, only 20 met a high bar for causal evidence, and none studied student AI use in U.S. K-12 classrooms. Read the review.
- Iterate: The Secret to Innovation in Schools by Justin Reich — on small experiments and the cycle of improvement. Find it on Amazon.
- Failure to Disrupt: Why Technology Alone Can’t Transform Education by Justin Reich (Harvard University Press). Find it on Amazon.
🐾 Sources & Citations: AI Research in the Classroom
As of June 2026, the research on AI in K-12 classrooms is early — these are starting points, not settled science. That’s exactly Justin’s point in this episode. Every source below was verified against its original.
What teachers are feeling right now
NPR/Ipsos. (2026). Teachers concerned about the impact of AI on students’ critical thinking. Poll of 545 educators, fielded April 27–May 5, 2026. Source (NPR) · Source (Ipsos)
Key finding: Nearly three in four teachers believe AI will have a bigger impact on education than the internet or computers did, and 54% say it is making it harder for students to learn critical thinking skills.
Caveat: This is teacher perception, not a measure of student outcomes — a nationally representative but modestly sized sample (545 respondents).
Why “too many standards” backfires
Marzano, R. J. (2003). What Works in Schools: Translating Research into Action. ASCD — the “guaranteed and viable curriculum.” Source
Key finding: Marzano found there are far more standards than instructional time allows — teaching all of them to mastery would require roughly a K–22 school system. Schools see better results when they prioritize a focused, “viable” set rather than racing to cover everything.
Caveat: Standards counts and instructional minutes vary by state and subject, so the exact gap differs from district to district.
Why “best practices” don’t exist yet
Stanford SCALE Initiative. (2026). The Evidence Base on AI in K-12: A 2026 Review. Stanford Graduate School of Education. Source
Key finding: Of more than 800 academic papers on AI in K-12, only 20 met a high bar for rigorous causal evidence — and none studied student AI use in U.S. K-12 classrooms. Performance gains often disappear once the AI tool is removed.
Caveat: The repository is growing fast (1,100+ papers within months). “Thin evidence” means not-yet-proven — not disproven.
How long good edtech research actually takes
Wineburg, S., & McGrew, S. (2019). Lateral Reading and the Nature of Expertise: Reading Less and Learning More When Evaluating Digital Information. Teachers College Record, 121(11), 1–40. Source
Key finding: Professional fact-checkers evaluate online sources by “reading laterally” — leaving a page to check who’s behind it — while students and even academics tend to read straight down the page and get fooled. It became the research backbone for teaching web credibility, roughly a quarter-century into the search-engine era.
Caveat: Justin used this as a benchmark for research pace, not a one-to-one AI parallel; the study examined search engines, not generative AI.
When unreviewed AI research goes viral — then collapses
Toner-Rodgers, A. (2024, preprint). Artificial Intelligence, Scientific Discovery, and Product Innovation. Posted to arXiv; MIT issued a “no confidence” statement and requested withdrawal (May 2025). Source
Key finding: A splashy preprint claiming AI dramatically boosted scientific discovery was praised by a Nobel laureate and covered widely — before MIT said it had no confidence in the data’s provenance or validity and the paper was pulled. A cautionary tale about acting on research before peer review.
Caveat: This was a higher-ed/industry productivity study, not a K-12 classroom study — cited here as an example of the “viral before vetted” pattern, not a finding about schools.
🐾 How I used AI on this post: I used AI to help draft and organize these show notes and to locate the studies referenced in the conversation and my introduction. I personally verified each citation — source, authors, year, and findings — against the original NPR/Ipsos, Stanford, SAGE, Marzano, and MIT/TechCrunch reporting before publishing, and reviewed the transcript for accuracy myself.
A note on Google’s founding date: In this episode, Justin mentions Google was founded “around 1995.” In my fact-check, it turned out Google was founded September 4, 1998 (though the Stanford research project began in January 1996). His underlying point about a roughly 25-year arc for peer-reviewed research still holds, however — the timeframe matches up.
About Justin Reich
Justin Reich is an associate professor of digital media at MIT, and the host of the TeachLab Podcast. The latest series of Teach Lab is called The Homework Machine, a limited series about the arrival of AI in K-12 schools, at teachlabpodcast.com.
Justin is the author of Iterate: The Secret to Innovation in Schools and Failure to Disrupt: Why Technology Alone Can’t Transform Education. He is a former world history teacher, wrestling coach, and wilderness medicine instructor.
Connect with Justin: X (@bjfr) | Instagram (@bjfr) | LinkedIn | tsl.mit.edu
Other Shows for Educators Navigating AI
Listen and Subscribe
If this episode made you think, share it with a teacher friend.
Episode Transcript
This transcript was generated using AI and has been reviewed by humans for accuracy. Minor errors or artifacts may remain but I worked my best to find any issues with the transcript as I reviewed the show. – Vicki
Click to read the full transcript
Vicki Davis: Today’s show is sponsored by EF Explore America and the STEM Tours. To show your students how STEM impacts the world up close and in action, go to efexploreamerica.com/STEM. And stay tuned at the end of the show to learn more.
Vicki Davis: Today we’re so glad to welcome back my friend Justin Reich. He is an associate professor of digital media at MIT, director of the Teaching Systems Lab. He’s the author of Iterate and Failure to Disrupt. He’s back to talk about The Homework Machine, his brand new limited series podcast that dives into what AI is really doing in our K-12 classrooms, based on 120 interviews with teachers and students across the country. So Justin, last time we talked it was about Iterate and small experiments in schools. But now you’ve gone and conducted these 120 interviews about AI in classrooms. What made you think that you needed to get the real story from teachers and students?
Justin Reich: It was the almost exact same motivation that had me visit your room 20 years ago. So 20 years ago, all kinds of folks were talking about Web 2.0 in schools. And what I understand better now is that when new technologies come along, elites dominate the conversation — the think tank people and self-described thought leaders and policymaker kinds of folks, and people like professors like me. And I really don’t actually trust any of those people. I trust classroom teachers and students a lot. At the very least, I’d say their voices are essential. For the same reason that I wanted to visit your classroom and see what was really happening in your environment in Georgia 20 years ago, I wanted to say, all right, ChatGPT has come and crashed the party. It has showed up uninvited in all of these different schools, and teachers and students are just bringing it into the classroom on their phones. And what do they think and what do they say about it? Because professors and thought leaders can think whatever they want, but the most important observations are the ones from the people who are closest to what’s actually happening.
Vicki Davis: Now you’re doing a lot of deep dive into all the individual stories, but let’s kind of back up at the 30,000-foot view. What kind of conclusions are you starting to draw?
Justin Reich: We’ve had a hard time concluding because of the schools in this country. There are 13,000 school districts and it hits different places really differently. There are lots of schools in the country where there are no substitute teachers, and kids come to school hungry, and kids are not showing up to school because of chronic absenteeism and huge challenges. And in those kinds of places, AI tended to be described as like the fifth, twelfth thing on people’s lists. Like, if kids don’t show up to school, it doesn’t really matter what’s going on with AI. One teacher said to us, “I would love to do a day of professional development on AI. There are no subs. There’s no one who can come into my classroom and have me leave.” It tended to be in more affluent places where people said, this is the number one concern, this is the thing that we’re really tackling. And then people just have wildly divergent opinions about what’s going on. There are some folks who said, this is a complete game changer for my classroom, I’m super excited about what’s happening. And there are other folks who said, this is a machine that just put words in my students’ mouth that aren’t their words. How am I supposed to teach someone if I’m just getting words from a machine? What’s this going to do to trust? What’s this going to do to our community? Really wide-ranging opinions. Probably some of the most exciting stories are where those wide-ranging opinions are in one community. I’m sure there’s some of that in your school. I’m sure there’s some of that in all your listeners’ schools — hearing about communities where teachers and students are trying to negotiate these challenges on a time scale that nobody asked for. Nobody gets to pick like, this is the AI year.
Vicki Davis: Mm-hmm. So Justin, let’s talk about this research for a minute. I just did a piece in my newsletter where Stanford studied better research — they’re studying AI and they found only 20 of them had any measurable results, but none of them are in the basically US K-12 classroom. It seems like to me there are a lot of people trying to draw far-reaching conclusions from research that’s in its infancy. Is that what you see happening?
Justin Reich: I’ve had reporters in the past week — one from the New York Times, one from Ed Week — saying, what are best practices with AI right now? I hear that from teachers all the time. What are the research-based best practices? That’s a really good intuition for teachers to have in lots of things. If your students are having a hard time reading, you should not invent reading instruction. We’ve studied teaching reading for 60 years and we can tell you better and worse ways to teach reading. We can’t do that yet with AI. To give you a kind of benchmark — I think Google was founded as a company around 1995. The first peer-reviewed paper that had really solid research about the most effective ways of teaching kids to sort truth from fiction on the web was published in 2019. So it took about 25 years for the research community to say, we’re pretty sure this is what you should teach students to do when they’re using a search engine to find facts. The arc of time that it takes for the research community to come up with pretty solid answers to important questions is unfortunately closer to decades than years. Ten years from now, you and I will have this conversation and we won’t be like, what should AI policy be? We’ll be like, we’ve studied AI policy in schools for a while and we’re pretty sure that when schools do this kind of thing, it doesn’t work as well, and when they do this kind of thing, it works better. But we’re actually still kind of a long way away from that. When big science is taking a long time, then what educators need to substitute is local science — going into their own communities and saying, we don’t have all the right answers. What we’re going to do is an experiment. The best ways to conduct experiments are to, A, tell the people involved that you’re experimenting. So parents and students and teachers should know these are just things we’re trying. There are no best practices yet. This is our best intuition of the way to go forward. And then you evaluate the evidence afterwards. You were just telling me a story about having your students do speeches in class, and you’ve had students do speeches in your class for decades. You’re saying, oh, when we do this AI-enhanced approach, the speeches were better — I graded them, I compared the grades from 2026 with the kind of grades I got in 2019. And because the performance of understanding is better, I have evidence that the thing that I’m doing is working. You could imagine there are other experiments that you could do where you try an AI-enhanced thing and you’re like, oh no, that made it worse. All the speeches came out the same because they were using AI in a way that homogenized things. And you say, okay, that’s a bad experiment. That one we’re going to throw away. And that, I think, is the crucial stage that we’re at — that local educators with their colleagues conducting their own local classroom experiments in this period of uncertainty. The research summaries that you’re going to get for the next decade are not going to give you the sort of slam-dunk answer, because big science just takes longer than that.
Vicki Davis: You wrote about the power of less, where everybody’s trying to add AI to everything. What should we be subtracting?
Justin Reich: The whole idea of subtraction is that schools are too complicated today. There are too many things going on, and we can’t be good at everything. We need to be deliberate about taking things away. In the national conversation, almost the only thing that schools across the country — blue states, red states, private schools, public schools — have agreed can be subtracted from schools is cell phones. You and I could have a long conversation about whether or not cell phones belong in classrooms. But here’s the thing that I celebrate: people said, we’re just not going to deal with this anymore. Put them away. And maybe they would have been a good learning thing, but here’s one fewer thing that we’re going to deal with so that we can deal with more important things. And I actually celebrate that part of the decision. Schools just have to decide — keep adding standards, new technologies — schools cannot solve all of the problems of society. We have 180 days, we’ve got seven hours a day. It’s actually not that much time. It’s a good exercise for schools to be regularly doing in their cycles of professional development and improvement: what are some things that we can stop? What are some things that we can set aside? Because we want to do a really good job on a manageable number of things, not a mediocre job at an unmanageable number of things. Because our schools are so diverse, it’s really hard to say, this is the thing that you should definitely get rid of. But it’s the things that are just kind of limping along and not really working anymore — finding what to prune is the way that you get your best stuff to grow.
Vicki Davis: Great way to end. Justin, thanks for coming on the show again.
Vicki Davis: If you’re a STEM teacher like me, you want your students to see how STEM impacts the real world, not just read about it. On an EF Explore America STEM tour, they might code robots with MassRobotics at MIT, explore marine ecosystems in Florida’s coral reefs, or even sit down with a former spy in Washington DC to discover how STEM thinking shows up where you least expect it. Every itinerary is designed by experts to amplify what you teach through hands-on experiences that can’t be replicated in the classroom. Visit efexploreamerica.com/STEM and see what an EF Explore America STEM tour can do for your students. Some of the greatest things I’ve ever done with my students have been tours. They make it all easy for you. So again, check out efexploreamerica.com/STEM.
Disclosure of Material Connection: This is a sponsored episode and blog post. EF Explore America has compensated me to share information about their STEM Tours. However, all opinions expressed are my own. I have personally planned and led student tours myself and only recommend tools and experiences I believe offer genuine value to classroom teachers. My endorsement is limited to the educational products and services discussed in this episode. This post also contains Amazon affiliate links for the books mentioned; if you purchase through them, I may earn a small commission at no additional cost to you. I am disclosing this in accordance with the Federal Trade Commission’s 16 CFR, Part 255: “Guides Concerning the Use of Endorsements and Testimonials in Advertising.” The sponsor has no impact on the editorial content of this show.
Never miss an episode
Get the 10-minute Teacher Show delivered to your inbox.
