{"slug": "why-ai-might-not-change-education-much-at-all-with-justin-reich", "title": "Why AI Might Not Change Education Much At All (with Justin Reich)", "summary": "MIT researcher Justin Reich, director of the Teaching Systems Lab, said AI will not transform K-12 education, based on interviews with 120 teachers and students across the U.S. Reich argues that new technologies, from filmstrips to MOOCs, have historically extended old habits and produced modest gains, often benefiting affluent schools most.", "body_md": "# Why AI Might Not Change Education Much At All (with Justin Reich)\n\nAI will revolutionize education! No, it will destroy it! Which is it? If history is any guide, the impact will be limited in either direction. In this episode, MIT education researcher [Justin Reich](https://tsl.mit.edu/team/justin-reich/) walks through a century of hype cycles — from filmstrips and radio to MOOCs and smartphones — to show what actually changed in classrooms, and what didn’t. Drawing on[ interviews with 120 teachers and students](https://www.teachlabpodcast.com/the-homework-machine-ep-1-buckle-up-here-it-comes/) across the U.S., he explains why new tools tend to extend old habits, why gains from new technology are usually modest, and why they tend to benefit affluent schools most. (Part 2/2 in our back-to-school series.)\n\n## Further Listening and Reading:\n\n[The Homework Machine](https://www.teachlabpodcast.com/the-homework-machine-ep-1-buckle-up-here-it-comes/) (Audio)\n\n[Failure to Disrupt: Why Technology Alone Can’t Transform Education](https://www.amazon.com/dp/0674089049?lv=shuf&channelId=500&plpRedirect=mhFallback)\n\n## Episode Transcript\n\n*This is a computer-generated transcript. While our team has reviewed it, there may be errors.*\n\n**Jess Love:** Welcome to Life, Automated, the show where we explore how to live, work, and make decisions in a world increasingly shaped by machines. I’ll admit it, I’m the kind of person who hears something about the incredible capabilities of AI and then immediately tries to project all the ways those capabilities might transform the world.\n\nThat’s actually one of the goals of this podcast, to try to determine how many of these mental leaps are grounded in evidence. And sometimes I talk to someone and I’m like, “Yeah, this is a really big deal.” Today’s conversation, not that. I’ll be talking to Justin Reich, a researcher and director of the Teaching Systems Lab at MIT, about AI’s likely impact on K-12 education.\n\nThis is not my first conversation about AI and education. I recently spoke with one of my colleagues here at Kellogg, Sébastien Martin, who has entirely reimagined many aspects of how he teaches his MBA students in ways I find pretty inspiring. But he’s admittedly an edge case, like if you accidentally ended up having Beethoven as your piano instructor and he gave you a warped sense of what piano lessons typically look like.\n\nJustin Reich, on the other hand, does have a sense of what lessons typically look like, not those designed by tech-savvy professors at top-ranked business schools, but in the rest of the world where budgets are limited, professional development is scarce, and time and time again, technology has entered the classroom only to fail to live up to its promise.\n\nJustin recently interviewed 120 K-12 teachers and students across America about AI for his podcast, “The Homework Machine.” His verdict? AI might modestly help some students learn some things. AI certainly is causing a lot of headaches. But somewhat surprisingly, given we’re talking about a technology that can pretty convincingly mimic human intelligence, he does not view it as transformational in the least.\n\nI’m Jess Love, and this is Life, Automated, a project from Kellogg’s Ryan Institute on Complexity, distributed by KQED. And here is my very grounded conversation with Justin.\n\nYou have been described as a healthy skeptic when it comes to bringing technology into classrooms. Do you agree with that characterization?\n\n**Justin Reich:** Yeah, healthy skeptic. I was recently at DeepMind in London where one of the Google engineers said that I was the token skeptic. I don’t know. I’m not ideologically skeptical. Like, I’ve used technology, computers in my teaching for more than 20 years now, and I still do it. What I aspire to be is evidence-based.\n\nTo me, maybe that’s skeptical, but it’s not skeptical as like, “I don’t know if I trust this technology stuff.” It’s more skeptical in the sense of like, what does a century of evidence tell us, and based on that century, what would we expect to happen next?\n\n**Jess Love: **Well, it seems like you’re kind of in a position to interview yourself, so I’m just gonna ask you the question you asked yourself. What does a century of experience with technology tell us about this current moment?\n\n**Justin Reich: **The first things that happen when teachers get access to new technologies is they use them to extend existing practices. So they do whatever they were doing before but with the new technology. So we used to write our notes on chalkboards, and then we wrote our notes on whiteboards, and then we put them on acetate sheets, and then they got projected onto SMART Boards, into LCD projectors. It’s the same notes. We just reproduce them in all these different formats. If you give teachers enough time and support and coaching, they will invent new kinds of practices. Unfortunately, the second pattern that we see is that, where there are benefits to new technologies, they tend to disproportionately benefit the affluent.\n\nThey benefit people with the financial, social, and technical capital to take advantage of new innovations. So for instance, in 2012, we invented these giant, these massive open online courses, and the main thing that we found is that they were pretty good for helping people earn their second master’s degree.\n\nSo if you were already educated, already affluent person, there are these great new opportunities for you, and it turned out that they really weren’t very good at helping onboard new kinds of students into the higher education system.\n\nThe third pattern that we see over and over again is that technologies are only as powerful as the communities that guide their use. So we’re constantly hoping that we can invent the software or these machines. You just sort of download things onto a bunch of people’s computers, and all of a sudden, learning gets better. And that essentially never happens. What can happen, where you can see improvements in learning, is when teachers have time to experiment and to try new things and to collaborate with their colleagues.\n\nPrincipals come up with new disciplinary standards. Students learn new routines. Families learn new ways of helping people. When whole communities have a chance to make improvements, you know, across the curriculum and lesson planning with technologies, that’s when we can sometimes see some benefits. And, we should typically expect that benefits are modest, because the benefits of anything that we do in educational settings are typically modest.\n\nIf you want to make education better, what you’re usually doing is, like, putting your shoulder to the wheel for a long time and being like, “Oh, it’s one percent better this year. That’s great. Let’s see if we can make it one percent better again next year.” Which is, of course, not at all what techno-utopians want to have happen or describing what happened, where there’s this giant disjunction with the past, and everything is better afterwards.\n\nI mean, it’s kind of fun to think that way. There’s not a lot of historical evidence for it, and, I mean, I think the reason why I critique that approach the most is that it tends to be where you see people wasting a lot of time and money.\n\n**Jess Love: **So let’s do a couple of examples here. So can you give us a couple of examples of times when techno-utopians came in and said, “This is it. This is going to change everything,” and then what actually happened?\n\n**Justin Reich: **Well, you got to start with the OG, Thomas Edison. Wow, so my man, a century ago, Thomas Edison, went in front of Congress, I think it was in 1913, and said, “In 10 years, textbooks will be gone, that they’ll be completely replaced by film strips, and this is gonna be a good thing as computers have been introduced” – I mean, radio went through this phase.\n\nThere’s a Larry Cuban has a great book called Teachers and Machines with a photograph in it, of a big, like, an armour-sized radio set. And it says, “With radio, the underprivileged school becomes a privileged one.” And so the idea that, like, we’re gonna have the best experts in the world broadcast radio lectures, radio lessons into homes all across the country, and it’s gonna be totally transformative of how students learn. Massive open online courses are probably the one that most recently went through higher education.\n\nAnd, I don’t know, Sebastian Thrun, who was a Google employee, a founder of Udacity, said that, “In 10 years, there will be fifty universities left, and Udacity might be one of them.” And as it turns out, today, sitting here in 2026, there are more than fifty universities that are left. People were really enthusiastic about the web, online courses.\n\nThere was a book called Disrupting Class, which Clay Christensen wrote. He’s the developer of the theory of disruptive innovation. In 2009, he said that in 10 years, by 2019, half of all secondary school courses would be mediated online, that they would cost a third as much to deliver, and they would have better outcomes.\n\nAnd my hunch is, if any of your listeners wander to their local public high school, they will not find that half of the classes are delivered online, that they will not find that the costs of running educational institutions have gone down by sixty-six percent, and they will not find, that the educational outcomes of the online learning experiences are substantially better than the ones that are being mediated by teachers. So there’s a pile of them.\n\n**Jess Love:** Now, this does not mean that these previous technologies haven’t changed anything about the classroom experience, or that they didn’t feel, in small ways, kind of magical. Here’s a story Justin likes to tell, one from before he became a researcher, back when he was a teacher.\n\n**Justin Reich:** So when I went to go get my first teaching job, the department head was interviewing me the summer before and he said, “Can you teach world history?”And I said, “No, but I promise that if you hire me by September, I’ll figure out how to teach world history.” And he said, “Well, maybe.” And he said, “All right, one more thing, you’re gonna be teaching in this kind of trial classroom where there’s a cart of laptops in the corner. There are these blue and orange clamshell MacBooks”, this sort of iconic form factor.\n\nAnd he said, you know, “And we’ve used ninth grade world history as sort of a testing bed to,” this was in 2003, “to see how these new computers could affect teaching and learning.” I said, “You can put a cart of bananas in the back corner of the classroom and I’ll teach with them. I just really need this job.”\n\nAnd he went ahead and hired me, and it was really fun teaching in that classroom. It was a moment where the world’s government and archives and museums were rapidly digitizing primary sources. And so as a history teacher, with those computers, I could really do some things that were quite different from my own high school education where, you know, maybe I had a book of primary source documents with 20 documents in it or something like that.\n\nNow I can, you know, you just sort of imagine like, oh, what was, you know, I wanna teach my students about the Harlem Renaissance. Oh, the Smithsonian has 20,000 song sheets from the Harlem Renaissance. They can each study their own document, which maybe nobody has looked at, in the last hundred years or something like that.\n\n**Jess Love: **Now, Justin says this was great. His students got a taste of what real historians do, find their own sources and documents and interpret them. But it also came with some hidden costs, in many ways, much greater than those of the computers themselves.\n\n**Justin Reich: **I was working in a private school, and I also recognized that the kinds of resources that were required to keep those computers running, to keep them charged, to keep our networks running, to find the productive things that students were doing and to highlight them, to find the malicious things that students were doing and stop them, was an enormous amount of resources.\n\nAnd so the sort of incredible possibility of what students and me as a teacher could do with new computers was always balanced against the challenges and realities of turning those new affordances into everyday routines of learning that really helped students.\n\n**Jess Love: **So fun, genuinely interesting, but not necessarily a game changer, and very resource intensive. I could see where this was going, and I wanted to know, is ChatGPT really just computers in classrooms all over again?\n\nYeah. Well, what is different, if anything, about generative AI?\n\n**Justin Reich: **Well, I think we will by and large see those same kinds of patterns, I mean, people get really enthusiastic about the new technology that are in front of them to the point of dismissing the magic of previous technologies. There’s sort of an argument that emerges, which is kind of like… I’m calling it the web was met. Like you hear people say, like, “Well, you know, this AI thing is just totally different. I mean, we-~~ ~~like, what could the web have possibly done?” I was like, “My guy, we took a handheld supercomputer, and we put it in the pocket of every child 13 years older in the networked world. We connected them to basically the world’s corpus of information, to every person they know, to every expert you can possibly imagine, and the effects on education range from not that much to maybe actually not that good.” You know, to the point where schools across the country are banning those mobile devices from people’s classrooms.\n\n**Jess Love: **Justin says that like previous technologies, there will be some things that AI is really good at. They just won’t be, in his words, “transformative.”\n\n**Justin Reich:** They seem to be pretty good at translation. Like, maybe that will become less expensive, and that will be sort of helpful, but we actually… It’s gonna find that it’s not transformative to schools because, like, just translating materials is not, like, the only thing you need to unlock educating, you know, students that come from all over the world and speak all kinds of different languages.\n\nI’m kind of enthusiastic about writing feedback, maybe. You know, one of the things we know is that, like, you need a lot of feedback to improve at things, and the machines seem to be able to generate reasonable writing feedback, but then you get other kinds of reports from classrooms that are like, “Yeah, my students really just want feedback from me, the human being teacher in the room,” because it turns out that most of what motivates us to learn is our social relationships with one another, and it doesn’t seem like social relationships with chatbots is a very promising direction for humanity to go.\n\n**Jess Love:** Well, there is at least one kind of disruption that absolutely is happening in schools right now. So this very deep intel comes from my husband. He is a Chicago public school teacher. He teaches at high school.\n\n**Justin Reich: **Excellent!\n\n**Jess Love: **And so I asked him, I was like, “All right, I’m gonna talk to this ed tech expert.” Like, “Give me the lowdown. Like, what is happening in your high school?” And so he had a number of things to say, and so I’m gonna share these with you. I think we’ll do it, like, one at a time, and you can tell me if you are in any way surprised by this.\n\nSo, to prevent students from using these chatbots to just do entire homework assignments, there’s been a big shift toward in-class assignments done on paper, which does seem to help with that cheating problem, but it introduces another challenge, which is that you’re then not spending that class time actually doing instruction.\n\n**Justin Reich:** Yeah. Happening all over the place, happening in universities, happening in lots of different contexts. If, like definitely, one potential thing to be sad about is that, like, if you believed it could be possible that five years ago you could send students home to write stuff and be reasonably likely that they would write stuff, and then you could use class for the time of being together and engaging with one another.\n\nAnd now it sounds like your husband, like many other teachers, believes, “If I want to read something that my students have actually written, I pretty much have to put them in a room and watch them write it themselves.”\n\nYeah, I think it’s quite possible that there are millions of fewer minutes of homework being assigned than in previous years. And if you believe that homework gives students practice that’s valuable for their learning, then that could be a massive drawdown in the amount of learning time that students are doing.\n\n**Jess Love: **So he and his colleagues have not entirely given up on the idea of homework, but what they’ve done is try to use technology to fight technology.\n\nSo he and his colleagues pay out of pocket, mind you, for a Google Doc extension that shows them a detailed history of the revisions made to a document. So the idea is you can actually see a video of an essay being constructed in real time, kind of sped up. And the downside is that it is still possible for students to cheat.\n\n**Justin Reich: **They just, they ask ChatGPT to write the essay, and then they’re literally sitting there, like, typing the essay from ChatGPT into Google Docs at, like, roughly a pace that they think it would look like you know, a 17-year-old comes up with thoughts in a unique manner and things like that. It’s kind of an interesting learning experience for the students, but obviously they’re not learning the thing that they’re supposed to be doing.\n\nYeah, like, the best possible use of your husband’s expertise is not surveilling students’ writing. I think we could find that as we ramp up the level of surveillance in schools, that that’s really not good for a democratic society, that you know, that a certain amount of privacy, a certain amount of freedom from surveillance is actually necessary for the Republic to continue as a Republic.\n\nIn the context that he’s in, what he’s doing is sensible, but if you aggregate that context across a lot of classrooms, you’re like, “Oh, that could be really bad, actually.”\n\n**Jess Love: **Yeah. And I’ll share one more with you. So the last thing he mentioned is building, he calls it “layers of resistance.” So for an essay, this might mean breaking down an assignment into a bunch of different pieces. Then students do each piece separately, and then once they’ve already done that work, they then combine it into a longer piece. So, you are making it easier to do the eventual assignment yourself since you’ve had to do earlier parts yourself before.\n\nBut you’re also making it harder to cheat because that would be, you know, just a lot more difficult to cheat at each of those steps and then combine them into cheating. And again, he says it’s fairly effective, and that’s kinda like his favorite strategy right now. And I asked him, I said, “Well, is this having the effect of requiring your students to kind of use training wheels to think longer than they would otherwise?”\n\n**Justin Reich:** So training wheels to think is oftentimes better than we imagine it is. There’s a group of educational researchers that are interested in this set of ideas called cognitive load theory, and an idea that we often have is that people become experts by behaving like experts. And they sort of argue, “No, no, no when people develop expertise, it looks quite different than being an expert.”\n\nBut I mean, you do, I think, Jess, have a good intuition there, which is like, “man, at some point, the kids, like, before they leave high school, probably just need to be able to write the essay.” Like, that it seems like that would be a pretty good thing, that we would want young people to be able to independently generate an argument in prose.\n\nAt least for the last, like, 30 years, we’ve thought that’s a pretty good idea to do, you know, with computers in particular. And boy, is generative AI making it hard for teachers to assign that task that we think is pretty good.\n\n**Jess Love: **But after going into these specifics, like, wouldn’t you agree that this is pretty transformative in terms of what students are actually doing during the school day?\n\n**Justin Reich: **I mean, definitely not transformative in the sense of, “Boy, this is great.”\n\n**Jess Love: **Justin really wouldn’t take the bait here. He was adamant. No, we don’t have any great evidence that how teachers and students actually spend their days, think lectures, group work, has really changed that much over the past couple of years, or for that matter, the past few decades, even if take-home essays are basically off the table now.\n\nBut I kept at him. That’s after the break.\n\nIt does seem like one really big difference right now is that there’s a lot of discussions about how AI is or isn’t going to change the kind of future and work that we’re preparing students for. I’m curious if you’re seeing, you know, is that impacting students’ motivation to learn, which would obviously have a very big impact in the classrooms.\n\nAnd I guess it also could start to change the question of what school should be for.\n\n**Justin Reich: **Yes. New technologies are a great catalyst to provoke conversations about what schools are for. In the last few decades, we’ve been particularly interested in the question, like, how do you prepare individuals for work in the labor market?\n\nAlthough if you go through the history of schooling, you know, in the United States, we have public schools as a bulwark of our democracy. When Thomas Jefferson wrote about public schooling in the notes of the State of Virginia, and proposed the system of public schooling, it would be so that our nascent democracy would continue to exist and have citizens who are prepared to take on their roles as citizens.\n\nBut, you know, citizenship is gonna change with generative AI, too, and so we should be thinking about some of those kinds of changes. My hunch is if you went to all of the K-12 schools in the United States, you would not see huge changes in student motivation because like, students are not that great at thinking about their long-term futures.\n\nLike, students primarily, like, they do not care that much about the subjects that we teach, for the most part. They care a ton about their teacher and their peers.\n\n**Jess Love: **Well, whether the students feel it or not, I guess my question would be to you, do you think that schools should be rethinking what education is for in this moment?\n\n**Justin Reich:** I don’t know – I think we should start by saying we don’t know. Not only do we not know, but historically, when we’ve made some of these guesses in the past, we’ve been wrong. So you could look at things like, you know, the sort of computer science industry telling people that it’s enormously important to learn to code in order to get good jobs, and now there’s a possibility that computer programming won’t actually be a very viable field in the near future.\n\nAnd, and these things go back histor- You know, in the 19th century, there were a group of educators who passionately believed that you really had to teach sentence diagramming – that if you didn’t teach sentence diagramming, like, Western civilization would fall. And we’ve mostly stopped teaching sentence diagramming, and maybe Western civilization will fall apart, but it’s probably not gonna be for the lack of sentence diagramming.\n\n**Jess Love: **Cursive. The great cursive debate.\n\n**Justin Reich: **The great cursive debate continues. Here, here, here are two stories you could tell about AI. One story you could tell about AI is that there is a lot to learn to figure out how to use AI, and that students should begin the process of learning that as soon as possible, that there should be a set of scaffold experiences, that we should change our curriculum, so that as people get older and older, there are more and more tasks that they do in partnership with AI, ’cause partnering with AI is hard to learn how to do, and if they do it with the supervision of teachers, they’ll be better.\n\nA second story that you could tell is that getting generative AI to spit stuff out is actually super easy, that there really is not that much to learn, and that what really differentiates people who are proficient and less proficient with using generative AI is whether or not they can evaluate output. Since the output is highly uneven, what you really need are people who can say, “Oh, this is a good idea, this is a good practice, and this one is not.”\n\nIt could be that there’s actually very little general expertise that you can develop to distinguish good output from bad output. What you probably primarily need is domain knowledge. Like, if you ask ChatGPT a question about plumbing, there’s nothing about AI which is gonna tell you whether or not it gave you a good plumbing answer.\n\nWhat you need to know about is plumbing. If that was the case, if domain expertise was sort of the key differentiator in people skills in using AI, then that would be pretty good news for schools and universities, ’cause the main thing they’ve done for however many hundreds of years is try to help people develop domain expertise.\n\n**Jess Love:** But I don’t- as a civilization, science does not know the answer to those two stories. Science cannot tell you today which of those two stories is correct.\n\n**Jess Love: **Yeah, it’s interesting. So I’m guessing you’re not a big fan of some of these moves more recently by school districts, university systems, states, even potentially the federal government, to implement various AI competency or AI literacy requirements.\n\nAnd I will point out some of these seem more, you know, pro-technology, like giving the students the skills they need to succeed with these technologies. Others actually take a bit more of a defensive crouch, like let’s teach students the critical thinking skills so they can discriminate between the good and the bad. But there is nonetheless a lot of overlap, which is that these requirements purport to prepare students to live in a world alongside AI. And I’m curious what you make of those.\n\n**Justin Reich:** Well, one thing that we’ve tried over the last twenty years in the United States is a strategy they might call, like, the ‘tech literacy’ strategy, where every time a new technology comes along, you define a set of skills that correlate with that technology.\n\nYou write some policy documents that say schools should be teaching those things, and then you, like, bake it for a while and watch and see what happens. And, like, if you were to pick a sort of education reform strategy that we could be almost certain does not work, it would be that one.\n\nIt works really well for pundits and policymakers. Like, it’s a great way for policymakers to be like, “Look, we did a thing. We passed a bill which says you have to learn some stuff.” But what you actually have to do to make a difference in schools is you have to translate those policy guidance into curriculum documents.\n\nThere are 3.5 million teachers in the United States. Like your husband, one in every one hundred living Americans has to raise their hand and say, “I will be a teacher this year,” in order for our system to function. To improve the capacity of 3.5 million people is mind-bogglingly complex.\n\nYou could probably tell me the number of minutes or hours that your husband has gotten for, you know, AI-related professional development, and I bet the number is not super high. What I’m sure of is the number is not commensurate to some kind of transformational change. And so, I mean, I’m not opposed to that strategy on any kind of ideological or philosophical… Like, sounds kind of great to me. Just historically, it has not worked at all. Go ask young people – “have people, have young people describe their social media practices to you?”, and you’ll be like, “Oh, that sounds pretty bad and not good for your health, actually.”\n\nBut there has been ten or fifteen years of, like, social media literacy in schools. Like, ask one of your students to, like, save a file to a folder, and watch their head explode. And you’ll be like, “Oh, maybe, like, we’re not that good at teaching digital literacy in schools.” So one is just, like, an efficacy approach – But even if you believe that that, like, general approach would work, you have to sort of ask the question, like, what kinds of things are we gonna stuff into that AI fluency and AI literacy?\n\nLike, what should that be? And I think we really, to this day, don’t know. So I mentioned before that I went to DeepMind the other day in London, and I cornered every engineer I could find, and I said, “Do you know how to train a junior engineer to code with a copilot?” I asked in big groups, in small groups, one-on-one. There was not an engineer or program manager there who told me yes. Every single one of them told me, “We do not know how to do that.”\n\n**Jess Love:** DeepMind, Google’s elite AI laboratory. Justin went to a conference there about AI and education. And when he says that software engineers told him that they don’t know how to train a junior engineer to code with a copilot, what he means is that they may have protocols and practices, but they’re not yet confident that they work.\n\nThis is a major concern in the software industry. Many big companies are adopting coding assistants like Claude Code, GitHub Copilot, Cursor, and Codex. Senior engineers can thrive in this kind of environment. They can prompt the AI with exactly what they’re looking for, and then they can manually check the outputs and write their own code when something goes wrong.\n\nA much more junior engineer can also use these code assistants to generate code that seems like it works, but if it has a bug or a vulnerability, they may not notice or be able to fix it. And worse, they may never get a chance to develop their skills further. So we could end up with a generation of software engineers who don’t understand how the software works, can’t fix it if it goes wrong, and have shaky ideas of what is technologically possible. Not ideal.\n\nCompanies know this could be a problem, so they’re experimenting with protocols for junior engineers, but this is all so new that nobody knows yet if these protocols will work. And they were frank about this with Justin.\n\n**Justin Reich: **If Google, which has billions and billions of dollars at stake to answer this question, does not know how to teach a junior engineer how to code with a copilot, like, what is a seventh grade middle school’s computer science teacher supposed to do?\n\nLike, what would AI literacy in that class look like until Google can figure it out? That is an excellent point. So you mentioned earlier this idea that, in general, previous educational technologies have – if they’ve had a positive impact, it’s been toward the students who are already either high-performing or come from very high-resourced schools.\n\nAnd I wanna ask you a little bit about special education, and in particular, kind of the extreme edges of special education. So this is kind of personal for me. My 11-year-old now, she has a rare genetic disorder, and I would say she does fall in that sort of extreme end of the continuum. Like, she literally will not look at a piece of paper if it hasn’t been, like, personalized with things that, you know, her teachers and aides and therapists know will draw her attention.\n\nAnd so there’s certainly this kind of low-hanging fruit that I could see being very easy helping, you know, busy professionals in the classroom. But I do see the possibility of something a little bit more transformative in a positive way – for kids whose brains just work so differently that the professionals involved don’t always have a ton of intuition about what will work.\n\nSo we talk a lot about AI having these jagged skills that are hard to understand from the outside, so being amazing at one skill and, like, hilariously bad at another. But there is a population of students for whom this is also true. I think a very concrete example here is severe language disabilities.\n\nSo right now, the vast majority of educational instruction is done via language. If you have a kid whose language skills are significantly more impaired than their other skills, how do you teach them? How do you assess them? Non-verbal assessments do exist, but guess how the instructions are given? They’re given in language.\n\nAnd it just seems like this place where a tool that can radically personalize, that’s completely agnostic to how a student chooses to answer a question, that has zero preconceived ideas about what will or won’t be an effective learning tool, could be transformative. And I know this is very hand-wavy, it’s very in the distance, but is there something here?\n\n**Justin Reich:** Oh well, for sure. So first, I’m definitely rooting for these people – I mean, I’m always rooting for the people who are making education much, much better. My, like, very boring, sometimes sad job is to, like, hop into these conversations and be like, “That would totally be great. Just, we should remember that people have been working on this for decades, and progress tends to be more measured.”\n\n**Jess Love: **That’s why they call you the skeptic.\n\n**Justin Reich: **There’s this whole field called universal design for learning, which, you know, has observed for a long time, sometimes better to think of curriculum as disabled than people as disabled. The curriculum is just not presenting information in ways and in mechanisms that people with different kinds of ability can access. And so we should do a better job of reinventing our curriculum and, you know, in fact, as we start bringing generative AI into the application of these kinds of things in special education, we don’t have to start from scratch. We can start from decades of effort of people using computers to do these same kinds of things.\n\nYou know, translation is one thing that we mentioned, putting learning resources into different kinds of modalities. So if there are people who don’t read text well, then we can just have the machine speak the text. You know, a strategy that we’ve tried a lot is to build just-in-time learning supports for people into resources, saying like, “Okay, if this learning resource in its current form isn’t working for you, like, push this button and it will talk. Push this button and the reading level will change. Push this button and this other kind of feature of it can be modified to suit your needs.”\n\nWhat we found historically is that the kids that we most want to push those buttons are not the ones who push the button. The like high-performing kids push the ‘Help Me’ button and the kids who we most wish would push the, like, ‘Give Me Some Extra Resources’ or ‘Change This to Support Me’ button would.\n\nYou know, when we talk to teachers across the country, adapting resources to folks with different kinds of abilities is one of the things that they’re most enthusiastic. The you know, generative AI technologies might be able to help them do, and I’m rooting for them, and I hope that there are companies that figure out ways of doing this more sustainably and at scale. And I wouldn’t be surprised if we saw some potential benefits from that. And those benefits are most likely to emerge not in the places where people download the, you know, the software that personalizes things for students with different learning capacities and things like that. It’s gonna be where whole communities are able to, like, rethink the way that they do special education in the context of that.\n\nAnd, you know, and it’s probably gonna be that students who live in more affluent places are gonna have the kinds of systemic resources that allow for all that training and adoption to occur.\n\n**Jess Love: **This idea of resources, it’s an important one, and not just for special education because every dollar, every hour of an educator’s time, it comes at the expense of money or time spent elsewhere, including on things that we do know work.\n\n**Justin Reich: **And the worst case scenario, which I think we’ve seen a lot of places, is that we make substantial additional investments, both in technology platforms and then in a bunch of extra humans to manage those technology platforms.\n\nAnd so the cost of schooling goes up, but because at best, the gains of those technology platforms is pretty moderate, you’re like adding a whole bunch of additional expense. Like, you know, you’re basically like in the Chicago Public Schools, like you bought all these Google Docs, and you bought all of these computers so that all the students can use them, and you bought all these IT professionals because the computers break all the time, and your husband like, is like, “Well, actually, the best thing to do is to have them write essays on pencils and paper.”\n\nWell, that’s an awful lot of money that we’re spending on all of this infrastructure to sit in a closet while your students are writing in composition notebooks.\n\n**Jess Love: **Will AI transform K-12 education? Justin really doesn’t think so. And frankly, I’m not sure whether to be disappointed or relieved by that.\n\nLike, there is something reassuring about a world where students continue to learn the same kinds of things that I learned when I was in school. It’s certainly preferable to one where the whole educational system grinds to a panicked halt because they’ve decided students don’t need to learn anything anymore.\n\nBut on the other hand, it seems my dreams of some automated tool that can magically help my daughter learn in a way no human has yet managed to is probably not right around the corner either. And I still think that experiments will be important, including the kind of ambitious, dare I say transformative experiments that folks like my colleague Sébastien Martin are pursuing.\n\nWe have to know what’s possible, and then we’ll need to roll up our sleeves. Test, learn, make sure that whatever gains we see in one classroom with one teacher can eventually benefit a much larger group of students. Because, and I’ll end with this, even our resident skeptic Justin agrees. At the end of the day, alongside all the headaches, new technologies do bring new capabilities. And little by little, these capabilities can tangibly improve the status quo.\n\n**Justin Reich: **I mean, the things that do work, it’s probably going to be more like 10 or 20 years of development rather than sort of stumbling across something which works super well in a year or two.\n\n**Jess Love: **So it could be transformative for the better, but it’s just going to take a ton of work and dedication and probably resources to get there.\n\n**Justin Reich:** My colleague Ken Kaedinger says that step change is what 25 years of incremental change looks like from a distance.\n\n**Jess Love:** I’m Jess Love. Life, Automated is a project of the Ryan Institute on Complexity at the Kellogg School of Management at Northwestern University. We’re distributed by KQED. Special thanks to today’s guest, Justin Reich. Jesse Dukes is our producer. Music by Steven Jackson. Recording help from Will Feeney and George Christensen. Marketing support from Ananya Mallapragada. Administrative support, recording, and wise counsel from Stacia Sliger.", "url": "https://wpnews.pro/news/why-ai-might-not-change-education-much-at-all-with-justin-reich", "canonical_source": "https://ww2.kqed.org/news/2026/08/19/why-ai-might-not-change-education-much-at-all-with-justin-reich/", "published_at": "2026-08-19 10:00:00+00:00", "updated_at": "2026-08-19 10:11:16.386263+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-policy", "ai-ethics"], "entities": ["Justin Reich", "MIT", "Teaching Systems Lab", "DeepMind", "Kellogg", "Ryan Institute on Complexity", "KQED", "The Homework Machine"], "alternates": {"html": "https://wpnews.pro/news/why-ai-might-not-change-education-much-at-all-with-justin-reich", "markdown": "https://wpnews.pro/news/why-ai-might-not-change-education-much-at-all-with-justin-reich.md", "text": "https://wpnews.pro/news/why-ai-might-not-change-education-much-at-all-with-justin-reich.txt", "jsonld": "https://wpnews.pro/news/why-ai-might-not-change-education-much-at-all-with-justin-reich.jsonld"}}