Tim Hwang: It’s 2028, three years from now, and you are 12 years old. How much of your learning is done with an AI assistant? A lot, a little, or none at all? Phaedra Boinodiris is the Responsible AI Leader for Consulting. Phaedra, welcome to the show for the first time. What do you think?

Phaedra Boinodiris: I think the answer is, it depends—in particular, on one’s socioeconomic background.

Tim Hwang: Yeah, we will definitely be talking about that. Skyler Speakman, Senior Research Scientist, welcome back to the show. Skyler, what do you think?

Skyler Speakman: A lot, but a little—and that is intentional for my three kids growing up in Nairobi, Kenya.

Tim Hwang: Awesome. And last but not least is Marina Danilevsky, Senior Research Scientist. I believe we’re talking about your kid here, who will be 12 in 2028. What do you think?

Marina Danilevsky: My son will be, and I think it’ll be a little, also intentional—even though, and maybe especially because, I live in the Bay Area.

Tim Hwang: All right, awesome. All that and more on today’s “Mixture of Experts.” I’m Tim Hwang, and welcome to “Mixture of Experts.” Each week, we bring you the analysis, hot takes, and banter that you need to keep up with the ever-hectic world of artificial intelligence. Today, we’re going to focus the entire episode, as we get to the end of the year, on AI and education: what that means for AI to be used in education, the risks and opportunities, and where we think it’ll go into the future. So, let’s just dive into it.

Phaedra, I want to turn to you first to just kind of set the stage for our listeners. I think AI and education is kind of widely hyped, and it’s sometimes difficult to know what it is that’s going on. So, I want to give our listeners just a lay of the land to start. What are the big uses for AI in education right now, and where do you expect it’ll go in the next few years?

Phaedra Boinodiris: Well, I think it’s really important, if you’re thinking about using artificial intelligence in an education context, first of all, to be focused specifically on personalized learning and having customized curriculum. But then also, utilizing it to help teachers curate that curriculum, to be able to augment their day-to-day. Additionally, there’s all kinds of things that happen in the background—back office—to help with operations. But I think, in addition to having a conversation about what is the usefulness of artificial intelligence in an education context, I think it’s also important to have a conversation about how we need to be changing our approach to how we’re even teaching the subject of AI in schools today, and how that needs to be changing going forward.

Tim Hwang: Yeah, that’s great. I mean, I guess your short answer is it’s kind of happening everywhere—front of the educational experience, school operations, teaching people about AI. I guess, Skyler, Marina, I don’t know if either of you want to jump in as both parents yourselves. I’m kind of curious about what you’re seeing on the ground with your kids. I mean, are you seeing teachers starting to use AI tools, or people being encouraged to learn about AI? I’m curious about how that’s all kind of playing out in your experience.

Skyler Speakman: I am seeing it come out a bit more with our kids and their teachers. I think one of my hot takes on this is, at least for primary education, this is an opportunity for us to channel all of the people-hours that ideally will be made available with the advent of AI. So, I think it would be really great to really keep the personal touch in the primary education space because of all the other enablement we’ve had in other sectors. So, I think really keeping this balance between the role of AI and that human connection in the classroom is so important for what, at least, we are looking forward to for our kids.

Tim Hwang: Yeah, for sure. Marina, what are you seeing? I’m kind of curious. I mean, Skyler’s in Nairobi, you’re in California—very different places—but are you seeing the kind of AI wave appear with your kids’ education?

Marina Danilevsky: Yeah, absolutely. I think that there’s a lot to be said for some very interesting games and gamification that is going on in the educational space. So, one thing I’ll call out without trying to be a sponsor is Osmo—really great games that would not have been available even a few years ago because of the capabilities of, like, the tablet camera to be able to see and directly interact. So, it’s this really lovely mix of what’s going on on the screen, but also being able to do things that are physical for spelling, for math, for coding. Those kinds of things are really great. Same thing with programmable robots—Botley and things of that nature. That’s the kind of thing that’s showing up as well. So, I think that it’s very interesting; it gives a lot more options for how kids could be exposed to these concepts, and that seems to be a good thing since kids learn differently.

Tim Hwang: Yeah, for sure. I think it’s one of the things I’m most excited about is, like, you’ve got all these options for learning the same topic now, which feels really interesting. Phaedra, let me ask you this: we’re looking ahead to the next year, 2025. I think we’re going to hear a lot more about AI and education. What is the big trend? Do you have like the one thing where you’re like, “Wow, this is really the thing that’s going to knock people’s socks off in the next 12 months”? I’m kind of curious about what our listeners should be paying attention to.

Phaedra Boinodiris: Oh, well, I was interested this week to see an article come out of PBS and their use of artificial intelligence, in particular, enabling children to have conversations with some of their favorite characters in the PBS learning shows. They are not using generative AI, but more traditional forms of AI, and it’s specifically targeted towards younger kids who naturally talk to the TV. So, I thought that was really interesting. I think we’re going to see more very clever ways, as Marina said, about the intersection of play and AI. And I’m really looking forward to seeing how that shapes the realm of education, and in particular, ways of being able to harness that to address more equitable outcomes in education.

Tim Hwang: Yeah, for sure. And I do want to talk a little bit about some of the concerns here. I think, Phaedra, do you want to go into that point just a little bit more? I know you mentioned it at the top of the episode as well—kind of these concerns ultimately about the equity of these kinds of tools.

Phaedra Boinodiris: It’s really important that we’re teaching people how to be critical consumers of technology at large. It’s really, really important, and in particular, how to teach people what is the real nature of AI and the real nature of data. One of my favorite definitions of the word “data” is that it’s an artifact of the human experience, right? We humans, we generate the data, or we make the machines that generate the data. But it’s important to recognize: we humans have over 180 biases, and counting. So, what’s really interesting about AI is that it acts as a mirror that reflects our biases back towards us. But we have to be brave enough and introspective enough to look into the mirror and decide, does this reflection actually align with my values, my organization’s values? If it does, it’s important to be transparent: why did you pick this data? Why did you pick this approach? And if it doesn’t align, that’s when you know you need to change your approach. And it goes back to the conversation I had at the top of the hour about not just having a conversation on how AI can be used to transform education, but how we really need to be teaching this in schools. Because if you’re lucky enough to be able to take a class on the subject of AI or AI ethics or data ethics, you’re probably in a higher institution, and you have self-categorized as a coder or a machine learning scientist or a data scientist—and literally not everybody else on the planet. So, I think we need to be thinking about how do we bring this kind of curriculum that’s holistic and multidisciplinary much earlier in people’s academic careers. In fact, you know, I see no reason why we shouldn’t be teaching this in middle schools, and in particular, in social studies class versus computer science class, which I think is where this subject ultimately belongs.

Tim Hwang: That’s great. And we want to—I want to get some more concerns out on the table. I think I do want to talk a little bit about how we approach these sorts of issues and how we address them. I know both Marina and Skyler said, “Well, hopefully a very little, and that’s by choice,” in terms of kids using AI to learn. And specifically, “AI assistance” was kind of how I had teed it up. Marina, maybe I’ll choose you first, and then we’ll go to Skyler. Why is that? I mean, what are your concerns there? Why would you want to limit access to these tools as a way of learning?

Marina Danilevsky: Well, I think because the nature of the tools right now—if you’re going to generative AI and not sort of the more traditional ML AI—is that it wants to adjust itself almost a little too much to the person. And it’s a good way to fall down rabbit holes that kids are maybe not yet very well equipped to handle. There needs to be some structure around that. So, on the one hand, it’s good to have the adjusting personalization; on the other hand, it can be dangerous. So, I hope that there’s going to be a decent amount of oversight for that kind of thing, and a way also of doing critical thinking. So, I think that from a very early age, what kids can be taught—again, in a gamification way—is how do you trick it? How do you break it? How do you make it lie to you versus tell you the truth? Then, you start to really understand, even as a kid, how to set the expectations that it’s not an oracle; it’s more potentially like Loki, the trickster. And see that that’s the kind of back-and-forth, mildly antagonistic relationship you might want to have with it. It’ll also help with critical thinking.

Tim Hwang: Yeah, I love that. I like part of the education here is like getting kids to break these technologies. It feels very, very rich and something we should talk more about. Skyler, do you want to jump in? I’m not sure if you share Marina’s concerns, or if you kind of think about your worries about this technology in a different direction.

Skyler Speakman: First of all, plus one on the gamification. I think that really is such an important catch for these kids going into this technology space. But I do want to give this example that I saw earlier on Facebook this week, I think, of just a really great balance between generative AI technology and classroom leadership. One of my friends from undergrad is a primary education teacher in the US, and he had this really cool post on Facebook where he generated some prompts that he used in his class that wrote Act One of a play, and then his students were going to act it out. And the students then had to write Act Two of the play. And so, having that type of dynamic leadership in front of the classroom, with content coming from both generative AI and from the kids themselves, playing back and forth on that, I thought that was just a really cool example of balancing the roles that come from generative AI, social interactions, and leadership from the classroom. So yeah, shout out to Donny Pieri on that.

Tim Hwang: And Skyler, I guess to kind of close this section before we talk a little bit about ways of addressing these concerns, I’m curious if there’s any other items you might want to throw on the table. I know Phaedra talked a little bit about the equity concerns; we just talked a little bit about dependence and personalization as things that we might worry about. I’m curious if there’s other things that you might want to put on the table that come to mind as we think about how to responsibly deploy this type of tech.

Skyler Speakman: Yeah, I think that those are some really great examples on the responsible and the equity angle, in particular. My kids do go to a private school here in Nairobi, Kenya, and that does look quite different from the global majority from around the world—north, south, east, west. And so, I think making sure that that is recognized and top-of-mind for how these things are deployed across schools of all sorts of socioeconomic backgrounds is a key point that Phaedra just started off the conversation with.

Phaedra Boinodiris: I do a lot of volunteer work with the Girl Scouts, and we have used games to introduce the girls to things like algorithmic bias. But then, I think it’s important to have conversations with them, like, “You know, give me examples of where AI has delighted you. Now, give me examples of where you were playing around with an AI, and the output made you feel really bad—like you knew it was wrong, or it didn’t make you feel good.” And listen to what they say. It is, I think, really telling when you invite a young person to be critical consumers of the tech and really be thinking about things like disparate impact or unfair outcomes. It’s very, very telling. And again, it goes back to what I was saying at the onset of this conversation, which is this is far more about social studies—like, whose worldview is actually being depicted in this AI model? Beyond just, “Can I trust the outputs?” Whose worldview is being reflected? Also, teaching them to ask critical questions like, “Who’s accountable for this model? How much better does it perform compared to a human being?” I just think these are all important things we need to be teaching the next generation.

Tim Hwang: I think that’s a great lead into the next segment, which is thinking a little bit about how we address some of this—these concerns in the technology. And I know, Phaedra, you’ve been thinking a lot about these issues, done a lot of work on it in the last few years. In particular, I know right before this episode, I was reading a little bit more about your work with Smarter Balanced. I’m kind of curious if you want to talk a little bit about your work there and how it applies to some of the issues we’ve been talking about.

Phaedra Boinodiris: Yes. This ed-tech company out of the state of California was interested in addressing inequity in traditional educational assessments. There’s been a lot of research that shows that traditional educational assessments are inequitable for a wide variety of different reasons, including that English might not be your primary language, or you might suffer from test anxiety, or you might be neurodivergent. There’s many countless reasons why traditional tests might not work, right? So, they wanted to dive into discussing and experimenting on whether artificial intelligence could directly address some of this inequity. And so, one of the things they tasked us to do was to form a think tank that included students from all over the world, and teachers in elementary, middle school, and high school, as well as people who had leadership in neurodivergent communities, etc. We pulled together this think tank and really dove into some very specific use cases for these AI models. Like, if you were going to use an AI to be able to ascertain the skill set of, let’s say, a sixth grader’s ability to comprehend a passage of text and have deeper conversations about that passage of text, what would unintended effects of such a model be? And then, given those potential categories of harm and the principles that this think tank came up with, how would you detail what are the functional as well as the non-functional requirements needed to be seen in such a model? And the principles that the think tank came up with, I think, were really interesting. Like, IBM, for example, you know, we detail fairness and explainability and robustness against adversaries and transparency and data privacy. This think tank, when thinking about AI models that are going to be used by children, included principles like kindness, and data sovereignty, and agency. And so, a lot of the work was thinking through, “What does it mean for an AI model to reflect a principle—a human value—like kindness? What does that look like in terms of feature and function?” It was absolutely fascinating work, and that report is being made public.

Tim Hwang: Yeah, Phaedra, I think that’s great. I think one of the things I’m really excited to see is all of these groups starting to articulate a lot more crisply what the values are that they want out of these technologies. And I think that’s such important work because it helps to really set up the goals—like, what do we need to do in order to make sure that these systems are doing what we want? Well, part of it is we need to know what we want in the first place. Skyler, I know I wanted to give you a chance to give a little bit of a travel report. I know you were at the AI Safety Institute’s conference, which, as I understand, is very much involved in the process of trying to develop evaluations and standards for the space. Did any of these topics come up? I’m kind of curious about how that might plug into what we’re talking about here.

Skyler Speakman: Yeah, I think it came up in two ways. One was directly with education as a use case, and the second one was a bit more indirectly, which is what are these kind of international AI safety institutes doing for capacity building and awareness? So, we’ve already hit on these two topics about how important it is for these young consumers to be critical about that technology—that’s the capacity building and awareness side. And a bit more on the policy side, this technical gathering for these AI safety institutes was really trying to spell out how we do risk assessment—everything from the “doomers,” you know, end-of-the-world type scenarios, to addressing the day-to-day harms that we already see in these deployed models. So, it was really a fascinating couple of days between technology experts, academics, and policy makers, trying to come together and put language together so that in Paris in a few months from now, in February, these countries can come together and sign these multilateral agreements about where they want to prioritize AI safety—again, from education, from healthcare, from market competition. Really, really cool space to be a part of, and that all just concluded last week in San Francisco. I was there representing the Kenya delegation. Quite an interesting event.

Tim Hwang: Yeah, that’s really exciting. And yeah, I think part of it is, especially in the US, education is regulated on such a regional level. It’s exciting to hear that at the international level, we’re trying to develop these global standards.

Skyler Speakman: You used two key statements there: regulating and standards. And the Secretary of Commerce presented at this conference, and she was incredibly clear: the AI Safety Institutes are not regulators. They are there to catalyze and provide standards. So, it was a really cool conversation to have in there. Both of those areas have a role to play, but these AI Safety Institutes are much more about catalyzing and forming standards, and not yet on the regulator side.

Tim Hwang: So, Marina, maybe I’ll present to you a hard question that I’ve been mulling over. I think, as we’ve talked about, there’s huge opportunity with technology; there’s certainly risk; but there’s a lot of work being done to try to mitigate them. But I’m sure some of our listeners will be listening to this episode and saying, “Well, there’s maybe one thing we haven’t talked about, which is: can someone just refuse to use AI in the future?” I feel like, should we give students the right to entirely opt out of AI? It seems like a lot of the discussion we’ve been having is, “Well, the technology will be here; we’ll just have to mitigate its risks.” I’m curious about what you think about that. Is that something we should be trying to protect as we build this new educational ecosystem? Or, ultimately, is it very challenging, given how AI appears to be headed to be ubiquitous in the future?

Marina Danilevsky: Well, actually, that’s interesting because I would ask you, what do you think would be the motivations for a student to decide to opt out? I can see a couple of things: it could be parent-driven; it could be because a student sees that they want their voice to remain theirs and not have any AI-assisted anything. I mean, can you learn things without AI? Yeah, we’ve been doing it for a while, so probably. What would be the motivation, do you think, for opting out?

Tim Hwang: I would say there’s probably a lot of fear of the technology itself, which is to say, “I don’t know much about it. I learned the old-fashioned way.” I could imagine that being a very strong incentive: “I learned with books; I don’t know why we need these new AI assistants.” I think that’s probably one of the risks. I’m sure there’s also a privacy risk. I’m sure some parents say, “Where is all the data about my kid going? Do I have any control over that?” So, you’re right, I think there’s a couple reasons why someone might be concerned about it. But, like any new technology, I think there’s just a lot of fear over what it is and what it might be doing to your kid.

Marina Danilevsky: The data is a really fair risk, although that’s something that maybe parents understand better than their kids do—especially today’s kids; they’ve grown up not even thinking about the fact that everything they do is online. But the idea of what does it mean to learn with it? I think this goes back to a lot of interesting things that Phaedra pointed out: are you going to be subjected to biases without even understanding that you are? Are you going to end up in some sort of an echo chamber? Are you going to not have the breadth and depth of concepts that you are trying to go through, that a human might find the appropriate time to push back, to stop, to pause, to redirect? And AI is not going to do that most of the time. The AI assistants, what they really want to do is keep hurtling along at speed in the direction that they’ve been pointed—at least so far; maybe things will change. So, on the one hand, part of education needs to be, how do you function in society? And even if you opt out, you do need to know how to handle it when it comes your way, or when it comes the way of your friends or your family. So, even if you have that critical of an eye, I think it’s not great to say, “I’m not going to learn it.” It’s like, “I’m not going to learn to follow traffic signals.” Well, I guess you can opt out, but it’s probably not a very good way to be a part of society. So, you at least have to learn about it, even if you don’t want to fully participate.

Tim Hwang: Yeah, and I think this is the third topic I really wanted to touch on: we’re now moving away from the AI being the teacher to the difficult, really interesting questions around AI literacy, which is, “You might opt out, but we actually think it’s really important because you need to know how to work with these systems in the future.” Skyler, you’re smiling; I guess you might want to jump in.

Skyler Speakman: Well, I was just reflecting a bit: do you think that “opt-out” conversation is happening at the family level, at the classroom level, at the school level? I mean, where do you think—maybe not the opting out, but the decisions to really engage with this technology—how do you see that working out on a practical level? What level of decision-making do you think is going to drive that type of adoption?

Tim Hwang: Yeah, I think it’s complex. The short answer is, I can see it emerging across any of these options. A school district might say, “This is untested; we’re going to opt out.” I could imagine a parent saying, “I don’t trust this technology; we’re going to opt out.” I could also imagine a kid just saying, “Hey, I don’t learn great this way. I learn best with books; I want to opt out.” So, I can see it happening across all those levels. I know, Phaedra, you’re right in the middle of it; I don’t know if you want to jump in and respond.

Phaedra Boinodiris: I would say the reason an individual or a group or a school or a state would want to opt out is because they don’t trust it. They don’t trust it. And there are many reasons why someone might not trust an AI model. And earnestly, it takes a lot of work to earn somebody’s trust. It takes a tremendous amount of work, and it’s not strictly a technical problem at all. It is a socio-technical problem. And with any socio-technical problem, it has to be approached in a very holistic way, first beginning with accountability. Like, do you actually have a group of individuals who are being held accountable for making sure that this model is behaving in the way that it’s intended to behave? Are they being transparent about this model and, again, the worldview that has been embedded within this model? The data—was it gathered with consent? Is it representative of all the different communities which have to be served in an educational system? Is it the correct data to use according to real domain experts who understand the context of this data and the relationships between this data? I’ll tell you, I think it’s very unfortunate that so many organizations are ill-prepared to be held accountable for these models. And again, it goes back to why the emphasis on AI literacy and really understanding what is the level of effort that is needed to put into these AI solutions in order to be able to earn people’s trust. And honestly, the hardest part, as I said, is not technical; the hardest part is the social part, and making sure that you’ve got the right organizational culture and the processes in place, as well as the tools and AI engineering frameworks, to do this work in a responsible way.

Tim Hwang: Yeah, for sure. And I want to unpack that a little bit more, Phaedra. What does this look like exactly—AI literacy in practice? I mean, is it, “Okay, districts, okay parents, okay kids, here’s a curriculum; you have to go through the AI 101 class”? Or is it something else that you’re envisioning?

Phaedra Boinodiris: Oh, heck no! No, no. First of all, it has to be multidisciplinary. When I say multidisciplinary, I mean, get it out of just strictly computer science class. Have it be where you’re bringing in schools of philosophy, schools of government. It is a truly interdisciplinary... The challenge, I think, at least within the United States—I’m not going to speak for other countries—but public school systems, even higher institutions within the United States, have been extremely siloed with respect to how they teach disciplines like artificial intelligence. As I mentioned at the beginning, if you’re lucky enough to take it right now, you’re in a School of Engineering most likely, and you’re not bringing in linguistics professors; you’re not bringing in philosophy professors to talk about worldviews and ethics or even disparate impact. To give an example, I’ve come across AI practitioners who are developing AI models to do something like offer predictions on what percentage interest rate people should be given with respect to a home loan, that don’t know what the word “redlining” is. They’ve never heard it before. And again, this points to why we desperately need to have a multidisciplinary, interdisciplinary approach to how we teach this subject. In other words, AI is not the death of liberal arts education; if anything, it’s more important than ever.

Marina Danilevsky: That’s right. She’s right; she’s absolutely right. And even when you look at generative AI, look at how much it’s being used to do coding now. What does that mean in terms of the programming profession? Whereas now, people are saying we need more English majors to be able to craft the right prompts, right? So, she’s right: liberal arts education is now more important than ever—that we understand what is inequity, what is human history, what is disparate impact, how do we approach ethics in a way that’s holistic and representative of all the people that we need to serve. I’m just now so much more optimistic about my undergraduate liberal arts degree.

Tim Hwang: Yeah, thanks—it was all worth it! Yeah, for sure. It strikes me, Phaedra—I don’t know if you’d agree with the statement—that the stakes are pretty high here in terms of getting this AI literacy bit to work properly, because it does seem like irresponsible deployment of the technology could lead to some incident that really reduces public trust, that means there’s going to be less use of that technology going forward, less opportunities to show that the technology can create real benefit. It almost feels like getting the trust in education bit is going to be the thing that ensures we can actually get to all the opportunities we’ve been talking about. I don’t know if you’d agree with that.

Phaedra Boinodiris: I think, in order to be able to get to the opportunities that we’re describing—where you’re creating models that earn people’s trust—you need to educate people on what the heck we’re even talking about. Like I said, what is the real nature of data? Because, interestingly, working with the clients that I do, so often, real domain experts who desperately need to be part of the conversations and have a seat at the table, their perception in their mind is, “I’m not a machine learning expert; I’m not a data scientist; I don’t have a degree, so why do I really belong here? That’s not really my swim lane.” And that’s what we’ve been communicating to people for decades—is that they don’t belong—which, in fact, they desperately do. We desperately need to have their voice at the table. And in addition to those domain experts—again, where you’re trying to solution something in their domain—we’ve got to have far more diversity and inclusivity in terms of who’s developing these models and the systems of governance around these models. And that doesn’t just mean gender, race, and ethnicity, but earnestly, people who have different lived-world experiences coming to the table to have discussions about, “Does this artificial intelligence solve the problem? Is it reflective of the needs of a wider variance of human beings? What are the unintended effects of these models? How do we design this in a way to earn people’s trust?” And as I mentioned, these aren’t strictly technical challenges.

Tim Hwang: Yeah, for sure. Marina, I’m curious how you respond to all this. You’re someone who spends a lot of time directly in the research, and I’m sure, again, when I talk about this with some of my friends in the machine learning space, they’re like, “This is overwhelming. We’re just trying to get these models to work; now you want to worry about all this other stuff.” I guess I’m curious: do you think, in effect, what Phaedra is proposing is that people who do machine learning in the future will look really different from the people who are mainly doing it today? And in part, it’ll be that they will have to be so strenuously interdisciplinary that it might end up looking quite a bit different from what we expect at an ICML or a technical conference today. I don’t know if you’d agree with that.

Marina Danilevsky: We used to think that only specific people needed the training to learn calculus, and that wasn’t because you’re going to be doing calculus forever; it was because you needed to learn what it is, and how it shows up, and what does it mean to have a structure and a proof, and things of that nature. I’d make a plug to join Phaedra’s social studies class. Statistics—early statistics. Statistics often, because part of what you really need to do is understand how these models even remotely work—just an intuition, not the deep math. But that’s what’s going to help you combine that along with your work in linguistics, your work in history, your work in language, and all the rest of it. I do find my own, slightly more liberal arts background, coming up a lot when it comes to trying to talk to people with examples that they can understand, but also, again, intuition from my stats classes comes back time and time again—the explanation of what these generative models do: they’re playing “guess the next word.” Simple things—they might not be completely accurate, but simple things. Don’t try to boil the ocean. If everybody has just a little more intuition, then you’re going to be more effective. Again, another example: look at cars. None of us understand how they work, but we understand how to drive them, we understand how to regulate them, we understand in general how we live with them and use them and what the effects are. It’ll get to that point. So, I’m not worried; I just hope that we’re not going to be rushing it. It’s going to take a little time for this to become pervasive and become natural and second nature.

Skyler Speakman: To the point about how this is going to take time, again, look at the traditional school systems today and how siloed the approach is, and how hard it is to get these different schools to actually work together on a collaborative curriculum like that. I think that is what’s going to be the hardest thing to move.

Skyler Speakman: Yeah, just last week, I was helping my 10-year-old make a probability wheel, which is a spinner, and it can fall in one of these things. And then I told him that, you know, his dad—me—I do probability day in and day out at my job. And I could just see his wheels spinning: “What do you mean?” You spin this probability wheel... But it goes to Marina’s point about starting those conversations early and the importance of that type of background and intuition. I’m seeing it play out already in some of these young lives. So, yeah, again, just a great comment, Marina, and backing that up with a real-world example from just a week ago.

Tim Hwang: Yeah, that’s great. I love that your kid imagines you just sitting in your office with a bunch of wheels, spinning them.

Skyler Speakman: Exactly. He couldn’t quite get it, but I told him that this is really important, and I use this on a daily basis.

Tim Hwang: All right, for our last segment, it’s the end of November; we’re starting to think about the new year. I want to go around and just ask each of you to tell us your greatest hope for the new year. If you could change one thing, what would that be? Marina, I think we’ll start with you.

Marina Danilevsky: As much as possible, get teachers up to speed and educated and comfortable and able to own what’s going on. They are, after all, the folks that drive how it’s really used on the ground. And any way that we can offer support to teachers, to meet them where they are and make this be something that’s positive in their classrooms.

Tim Hwang: That’s a great one. Skyler, you next.

Skyler Speakman: Doubling down on supporting the teachers, but with their outside-the-classroom work—you know, the extra work. I think there are some areas that could be lifted off them to make them so much more impactful and involved from the front of the classroom. So, I think AI’s got a role to play helping teachers from the front of the classroom, but also, I guess what we’d call back-office stuff as well, that could really, really change the lives and aspirations of teachers.

Tim Hwang: That’s a great one. And last but not least, Phaedra.

Phaedra Boinodiris: Well, as I mentioned, I want AI in social studies class, and I want it taught much earlier—like I said, middle school, if not elementary school; you could twist my arm. But then also, I would love to be able to see more schools making a concerted, deliberate effort to make more room at the table—pull the seats out—and invite students who don’t see themselves as being technologists, and say, “Hey, having a conversation about AI and what it means for you, and does it reflect you, is core to you having a seat at this table to be a critical consumer of this tech.” That’s something I would desperately want to see within the coming years.

Tim Hwang: Phaedra, Marina, Skyler, thanks for joining us, and we’ll have to have you back on in 2025 to talk more about this. And thanks to all of you listeners for joining us. If you enjoyed what you heard, you can get us on Apple Podcasts, Spotify, and podcast platforms everywhere. And we’ll see you next week on “Mixture of Experts.”