Laurence Holt joined me and Diane Tavenner to unpack the current landscape and future potential of AI in K–12 education. The discussion centered on the three main AI use cases Laurence sees emerging in schools: generating materials, providing feedback, and AI tutoring. The conversation explored the vital difference between feedback and grading, the importance of instructional context for effective AI tools, and the complex challenges in cultivating curiosity and self-efficacy in classrooms. We also delved into why AI tutoring isn’t yet transformative for most students, the limitations of current chatbots, and the need for school model redesigns and tools that support social learning and durable skills.
Michael Horn
Michael Horn Here. What you’re about to hear is a conversation that Diane Tavener and I recorded with Laurence Holt, and I wanted to highlight just a few parts of the conversation for you as you begin to listen. First, Laurence named three three primary use cases for AI today in education. Number one, generating materials, number two, feedback. And third, is AI tutoring, now the only one of those ready for primetime, in his view, feedback. We then dove into why it’s crucial not to conflate feedback with grading and how, if we do so, it will actually undermine stream student motivation. Another part of the conversation that I thought was important to highlight is that Laurence observed that to be instructionally useful, AI applications need to deeply understand what students or teachers are trying to get done and then build with that in mind. That means the tool needs to have a lot more context than a chatbot that any of us would just fire up on its own.
And Laurence doesn’t believe we’ve gotten to the point yet where the tools we have understand the instructional context or have the right data on students. Finally, I thought the conversation around why AI tutoring is still falling short was very telling, particularly Laurence’s implicit observation that a lot of the blame is on a system of education that deprioritizes curiosity, developing self efficacy, and creating the time for true learning, and that it will take a lot to overcome that. Hope you enjoy our conversation on this episode of Class Disrupted.
Diane Tavenner
Hey Michael.
Michael Horn
Hey, Diane. Good to see you as always.
AI’s Impact on Schools
Diane Tavenner
It is. And what a fun season we’re having here on Class Disrupted. I’m hearing all sorts of comments from folks, I know you are too, about the conversations we’re having, which so far I would say have been pretty expansive at a higher altitude or a different perspective, which people love. But there’s also a bit of a craving for a closer look at what’s actually happening in schools. Which is why I’m really looking forward to our conversation today because we get to stay sort of at that bird’s eye view with someone who’s been tracking the full range of new AI powered companies, products, tools, programs and schools in K12, but also has been going deep with a few of those use cases. And I think it’s going to be an awesome transition into what I think is, I guess, spoiler alert, the next part of our season where we’re going to go deep with a number of folks who are sort of much closer to the action.
Michael Horn
Yeah, indeed. I think that’s a good summary, Diane. And the person we have on today is someone I’ve gotten to feature on in my writing, my work in the past just because he seems to constantly be doing really interesting explorations and finding out really interesting angles on, on things that maybe were accepted wisdom and then we find out that they aren’t what we thought. So I’m thrilled he’s joining us today. He’s none other than Laurence Holt, who is a senior advisor at XQ Institute and the Teaching Lab. And along with several others, he created an EdTech Insiders map to track over 60 use cases for Gen AI in education and over 300 Gen AI powered education tool. He was previously chief Product officer at Amplify. And again, what I love, Diane, about talking to Laurence is that look, he’s an engineer by training who went back to school to understand how people learn.
So dug into the learning sciences, neuroscience, cognitive science, and then spent years creating products that worked for or as he often says in his own words, didn’t always work in classrooms. So I love the background he brings, his experience, his humility, his humor and his ability to dig deep. So welcome Laurence, good to see you as always.
Laurence Holt
Great to see you both. Long time listener, first time caller, so welcome.
Michael Horn
Now we won’t get to say that again, so I’m glad you’re here. Let’s start high level with the EdTech Insiders map and what it tells us. What are you seeing out there in terms of emerging tools and products? And have the use cases changed much in the past year or two? Just sort of help orient us to what are the big areas for entrepreneurship, product development and so forth?
Laurence Holt
Yeah, that’s actually you hit a really interesting area straight up with a number of use cases. So we actually organized this map around use cases, meaning instead of just listing all the tools that are out there, thinking about what AI could genuinely help with and then cataloging are those indeed things that people are creating? And so you can, you know, anyone can see that for themselves. It’s on EdTechInsiders AI. It’s free with the help of our friends at Overdeck and we started in June 2023. So just after really six months after ChatGPT, Jacob Klein and I were trying to figure out how do we track all of this stuff that is bubbling up. And the first version had 40 some use cases and now we’re up to 60. So it’s not actually been that many.
We’ve added hundreds of new tools. The number of use cases is slowing, which I think tells us that the sector has gone broad. We’ve sort of had a look at almost anything that you could improve with AI. And now we’re going more deep. And in particular, the areas I like to think of are sort of the big three use cases that have emerged over time, only really one of which is sort of ready for primetime is AI good enough for. And the three I think of are generating materials. And when you interview teachers, there’s a survey, there’s a great Gallup survey that shows that’s the main thing teachers say they’re doing with AI, so creating quizzes, assignments, lessons, role plays.
There was a teacher who wanted to, a science teacher wanted to teach vacuums in middle school, and AI suggested to her that she should do a role play where the kids were 1930s vacuum salespeople going door to door and had to explain to families like, how does this thing work? Which I thought was like, that’s a really cool case. So that’s on the map. And there are lots of others that might tweak your interest. But generating materials is number one. Number two is feedback. So actually commenting or giving input to students based on their work. So not just right or wrong, not just multiple choice, but could be their writing, it could be their math written work, which AI can now do.
It could be a presentation they’ve made, so they’re just able to get way more feedback than previously. And then the third is AI tutoring, where we’ve seen just a huge upswing in the number of tools. A lot of them were around coding originally. We’re now seeing a lot around early reading and math. So those are the big three areas.
Michael Horn
Super interesting. I’m curious if you see differences in these by grade level or subject areas, or also if you’d give some commentary on those three big areas, like where are they really good today, these AI tools? And where are they still primitive and not ready for prime time and maybe won’t be ready for primetime?
Laurence Holt
Yeah. So I think the one that is very definitely ready for primetime, in my view, is feedback. And this is partly because if you look at the amount of feedback the average student gets on their writing or their math homework, it’s actually very low. And the reason for that is because it takes a huge amount of time and teachers just don’t have the time to do all of that grading. Right. So in a way, any feedback at all would be better. But there have been studies that show feedback is already, AI feedback is already as good as, say, a median teacher.
And if it’s on writing, you know, writing is the thing that LLMs, large language models are really good at.
Michael Horn
Does that depend on grade level or is it sort of equally distributed across a student’s, you know, age for how good is it’s feedback?
Laurence Holt
I think it’s been tested mostly in middle and high.
Michael Horn
Okay.
Laurence Holt
But I think, I mean, we’re certainly seeing feedback on your reading in very early grades. So I think it’s like a lot of these things, it’s kind of a jagged frontier that AI is good at and you wind up with specific point cases where it’s very strong and others where it’s not so great. So I really think of feedback as the first sort of, you know, fluoride in the water opportunities for AI. If we could just make feedback available free to every student K12 in the US or beyond K12, that itself could be transformational.
