We're Raising the First AI Generation. They're Pushing Back. (Rebecca Winthrop, Brookings)
“Young people harbor real anger about AI because they’re already experiencing its consequences in their schools, relationships, job prospects, and feeds.”
I end every interview with the same question: What’s one small signal you’re paying attention to?
Late last year, I started hearing the same answer: AI is becoming a political and cultural lightning rod.
Former Labor Deputy Secretary Seth Harris saw it in politics. In our conversation, he pointed to Georgia, where two Democrats won statewide Public Service Commission races after campaigning on the risks data centers pose to energy affordability, a race Democrats hadn’t won in decades.
Soon after, futurist George Siemens described AI as a vessel for much broader anxieties about work, power, and the future.
Today, that trend is no longer a small signal. It’s in full swing.
The data tells the same story. A Gallup and Walton Family Foundation survey found that Gen Z’s use of AI has remained steady, but their optimism has fallen sharply. The share of young people who say they feel excited about AI dropped from 36% to 22% in a year, while those who say they feel angry rose from 22% to 31%.
Rebecca Winthrop, who directs the Center for Universal Education at Brookings, helps explain why. Through years of research with students, teachers, and parents, she has watched young people’s views shift as AI begins to shape their learning, relationships, job prospects, and sense of the future.
As Rebecca told me:
“Kids harbor real anger. They’re pissed about climate impacts. They’re upset about job prospects. They’re outraged about AI being used to plagiarize writing and produce counterfeit artwork. Their relationships are being affected, and they’re seeing deepfakes in their feeds.”
Our conversation explores what follows from taking those concerns seriously. We discuss cognitive stunting, the case for small, purpose-built AI models over frontier models in education, why students should spend far more time in explorer mode, and how schools can cultivate agency, curiosity, and independent thinking in an age of AI.
The interview ends with one of Rebecca’s most practical recommendations: every school should have a student AI council, not as a symbolic gesture, but with real influence over the tools their schools adopt, the ways they’re used, and the data students are asked to hand over.
Every generation of technology seems to relearn the same lesson: build with people, not for them. AI should be no exception.
Enjoy,
Allison
On the Potential of Intentional, Niche AI Applications
ALLISON: What is something you have changed your mind about in the last year or so?
REBECCA: I’ve evolved my thinking on the dosage of AI strength different tasks need. In my view, this is one of the most important areas of inquiry going forward.
We don’t need the collective power of the Internet to level fourth grade reading materials for second or third grade readers; we need a tiny fraction of it. What smaller models might better enable that task—and what energy and data sets do we need to power them? There’s so much opportunity to explore local models run on desktops, or small language models that need less data or power. I’m honestly a bit obsessed about this now.
ALLISON: What’s so interesting about that notion to you?
REBECCA: This graduation season has demonstrated the breadth of pushback against AI among young people in the US. They’re booing anyone who talks positively about AI and cheering for AI skeptics. And I get it. One college junior told me, “It’s hard to be optimistic about AI when it’s actively undermining my future in terms of my career.” They’re also upset about the climate footprint—young people are really pissed at what looks to them like a deprioritizing of climate change amid the AI boom.
But small language models and locally run models don’t carry the same climate impact as LLMs, and they can be deployed in ways that amplify generative AI’s benefits because they can be purpose-built for the task. Our AI task force calls these technologies narrow, versus wide, AI—we borrow it from the Center for Humane Technology. Intentional, careful, niche deployments of AI can outperform general-purpose uses, including chatbots and AI companions, on the problems those uses create.
ALLISON: What are some of the most unresolved tensions in your thinking on AI and education?
REBECCA: I’m starting research on exactly where the line is between cognitive tasks we should do on our own and those that are acceptable—or even beneficial—to let AI help with. I just wrote a New York Times op-ed on creative thinking and writing; there’s interesting data showing that if you use AI to write, even when you’re putting in ideas and going back and forth, it will improve most high school writers’ syntax, style, word choice, and overall prose quality. But if you look across the cohort, their ideas were getting homogenized; their essays are very, very similar to each other. Using AI appears to have reduced creative thinking for these writers.
So—where is this line? How do we nurture and protect creativity in an AI-powered world? It’s clear to me that you should brainstorm without AI, since using the tool at that stage encourages you to latch onto one idea rather than formulate a divergent one. Could you design an AI tool for brainstorming that actually supports divergent thinking? I don’t know. I don’t even know if that’s possible. The frontier models don’t even know why that’s happening. In my own practice, I brainstorm and write the first draft alone for that reason, then ask AI to polish it at the end. But substantial editorial changes happen in “just a light copyedit,” too. So what’s the right dose of AI for creative work? Too much? Too little?
ALLISON: I’ve recently come across the concept of cognitive sovereignty: the idea that, even as AI becomes a bigger part of our work, we should remain in charge of our own thinking. It’s not about avoiding AI, but about being intentional about when we delegate, when we don’t, and staying aware of how it’s shaping our reasoning.
One of my team members, Karen, took that idea and coded a prompt into her Claude instance that holds her accountable to cognitive sovereignty throughout her work. I loved it because it’s such a concrete example of expressing your values to AI so it can help you live them, rather than simply making you faster.
REBECCA: Cognitive sovereignty is so important, and will change as AI comes to include ambient AI, wearables, and brain computer interfaces. We won’t always be interfacing with AI through a screen. And we will need to be prepared to protect our sovereignty as AI continues to evolve.
On Taking AI Resistance Seriously
ALLISON: Since you mentioned it, let’s talk about the booing at the commencement speeches. What is going on there? How does your data tell you about students’ AI skepticism?
REBECCA: I’m intrigued, but not entirely surprised, by the increasing negativity on AI from students in our focus groups. Walton Family Foundation and Gallup also just released a large survey showing Gen Z is more negative now than they were a year ago. Climate is a big reason. They also want to protect their own learning and development. Young people are hungry to learn, want to be engaged and challenged, and want to find their path through the world—and they feel AI will hold them back.
Instructional design is honestly part of the problem. Students who want to do the work themselves face a Catch-22: struggle through the hard work of learning and maybe get a C, or use AI like everyone else and get the A that gets them into grad school. That’s not a position students should have to be in.
I heard another example on Justin Rice’s podcast: a high school student said, “I don’t use AI, but all of my peers do. The teacher doesn’t know that we don’t understand; she’s moving really quickly through the content, and now I’m getting farther and farther behind. Now I feel like I have to use AI to catch up.”
ALLISON: Interesting. So the teacher believes that the class has a brilliant understanding of the material based on the work they are submitting.
REBECCA: Right. Because how else is a teacher going to get that understanding?
AI is requiring that teachers rethink assessment. Our kids deserve to have instruction that genuinely reaches them where they are.
On Two Futures for AI and School
ALLISON: Describe two extreme futures for AI and learning. What happens if we get AI and education right? What happens if we get it wrong?
REBECCA: The positive vision looks something like this: intentional, strategically deployed AI becomes a deep, catalytic mechanism for redesigning schooling, in ways that educators, neuroscientists, and learning scientists have long advocated for. School activates and supports students in explorer mode much of the time, and protects social interaction and learning eye to eye, shoulder to shoulder—how humans evolved to learn. We won’t evolve differently in the next 200 years, let alone 20. Sitting alone in a room with a screen doesn’t lead to happiness and flourishing.
School that prioritizes explorer-mode learning increases rigor—students master knowledge and content for purposeful application. They learn the Pythagorean theorem because they’re working on a project that needs them to use it, teach it to a teammate, and practice it in different contexts. That’s where deep learning transfer happens.
Finally, I hope kids can be excited to learn. I don’t have a crystal-clear vision for how AI achieves that, but I imagine it involves a lot of back-office support and perhaps assessment work too. Assessment has a stranglehold on our schooling design. But when you can see where a kid’s learning journey is stuck, and assess their competencies and ability to apply knowledge, it makes a huge difference in their forward momentum.
I don’t see kids using AI to do a lot of actual work, like learning to read and write. Writing is already a highly efficacious pedagogical tool for teaching critical thinking at scale; Socratic dialogue, too, was always imagined as an instructional technique that scales. We may develop new ways of teaching critical thinking at scale, but if AI lets students and teachers spend more time reading, writing, and discussing, that’s a net good.
If we give in to wide AI use, here’s the negative vision: school stays centered on task completion, students use AI to do the work, and we don’t redesign teaching and learning around the activities that matter. Kids are completely demotivated, as they’re already beginning to be. I hear kids say all the time: “Why do I have to do this task if AI is so much better and faster at it than I am?”
Demotivation leads to lack of engagement, effort, resilience, and cognitive stunting—to use your fast food metaphor, they don’t get enough of the right nutrients to develop cognitive strength and breadth for life. This isn’t just cognitive offloading. It’s fundamentally changing the conditions of their development.
ALLISON: Right. Adults offload because they are functionally delegating work to the machine, and managing that work—that’s fine for some things. But stunting is about never even developing the skill to begin with. I think that’s really useful language.
REBECCA: That’s exactly right. What happens if generations of people don’t get enough effortful learning to lay the groundwork for independent, consecutive, critical thought? I’m also concerned about loneliness, and how AI is poised to exploit and amplify it for everyone. This kind of general-purpose AI isn’t tooled for teaching, and it’s not safe for kids.
ALLISON: Do you think we are on one of these future paths already?
REBECCA: Our task force report, a pre-mortem on generative AI in K-12, suggests we’re on the negative future path. To be clear, we weren’t looking at AI in back-office work; we were looking at students, learning, and development. We found that yes, there are some benefits from narrow AI use, but the risks currently outweigh them, and are of a different nature.
We’re already seeing cognitive stunting, reduced relationship abilities, reduced trust in teachers and fellow students—it’s already in full swing.
ALLISON: What action, taken now, could move us away from that future?
REBECCA: The task force developed three levers, and they can be acted on in any order.
One, we have to shut down wide AI use in K-12. There’s specific policy language in the report, but it boils down to a single directive: do not let kids under 18 access AI friends, chatbots, and companions. Anything that pulls from the entirety of the Internet isn’t safe for kids, and isn’t designed for learning.
Two, prepare, prepare, prepare. Encourage everyone across education — administrators, teachers, parents, students — to understand the field of AI, the technologies it’s created, and how those technologies have been and could be applied, in their field and others. We all need this foundational understanding to use it carefully. Plenty of students told me they use ChatGPT and an AI “humanizer” to complete work—AI is everywhere kids are, and they’re already misusing it.
Three, adults need to harness narrow AI to help kids prosper. It’s not fair to put teachers on the front lines of that work. But it’s now on their shoulders to shift their pedagogy in response. We talk about pedagogy that’s AI aware, AI assisted, and AI resistant. AI aware assignments aren’t easily completed with AI. AI assisted teaching intentionally asks students to use AI to achieve the learning objective in targeted, specific ways. AI resistant pedagogy is about creating time and space for learners to do things on their own, without AI support. All three are critical for keeping cognitive stunting at bay.
On How to Make School a Place for Exploration
ALLISON: As your research has noted, learners spend only 4% of their time in “explorer mode,” meaning work where they have meaningful agency to define the problem, choose a path, test ideas, and create something that is not predetermined. I feel some paralysis hearing that number. It suggests the future of school we want to build is not an incremental shift, but a sea change. Can you ground me a bit? What would a healthy percentage of time in explorer mode look like, and what would need to change about school to make that possible?
REBECCA: That’s an interesting question; I’ve never thought about it in terms of percentage of time across the school day. But we know we don’t need to be in explorer mode 100% of the time to learn well. There’s a time and place for each mode, actually. But we certainly need many more explorer moments in school, and out.
I’d love to see kids in middle and high school spending 50-60% of their time in explorer mode—that would be transformative for young people. Explorer mode is where agency meets engagement, and learning becomes fun. When you hit a roadblock, you ask for help. You’re more advanced, happier, healthier. You’re doing better in school, but also better in life. You become nimble at navigating change, which is critical in the era of AI.
ALLISON: 50-60% of time in explorer mode is a big change, and it’s not what our system is incentivized to do. What would you advocate for to help realize that shift?
REBECCA: All kids can get into explorer mode. There’s tons of evidence to show that. So it’s not really on the kids. It’s on the adults to create learning environments to enable it. There are three design premises that can help us create that environment.
One, pairing knowledge acquisition with knowledge application. That can be done many ways—high-quality project-based learning, experiential learning, inquiry-based learning, service learning. Across this suite of pedagogies, students pair theory with practice. To know isn’t enough; you also have to apply.
This pedagogical shift necessitates rethinking assessment. So, two: we need to rewire assessment around applied competencies and skills, not only content knowledge. Can you do the math, then apply it to a real-world circumstance? Can you read a text, then connect the argument to a cultural conversation you’re engaged in? Can you apply your knowledge of physics and chemistry to fixing something, like a broken motor or appliance? We’re already seeing some schools go all in on AI to help with that work—Anaheim Public Schools is one example—and it’s helping them track knowledge acquisition in a more visible, rigorous way.
Three, we need to think more creatively about where learning can happen, and who kids can learn from. Real schools and communities are the canvas for theorizing explorer mode, and some districts I’ve visited over the years have massively transformed equity and outcomes by rejiggering elective credits in high school. More applied pedagogy in those courses generated engagement, agency, and excitement that spilled into core subjects too. That’s a smart way to start shifting more of the school day into explorer mode.
One Small Signal
ALLISON: What’s one small signal to which we should be paying more attention?
REBECCA: Pushback from young people—we shouldn’t dismiss it as youthful rebellion. They feel we’re crafting a future without their input, and without consideration for their needs and desires. That’s something we should take seriously.
Our task force recommends every school have a student AI council, and not as a token gesture. Students want to beta test products. They want to give teachers tips about how assignments did or didn’t work. They want to review procurement contracts; they know it’s their data being handed over, and they want a say in how it happens.
ALLISON: Do you feel like people are taking youth pushback seriously?
REBECCA: I think people are surprised, and willing to brush it off as “influence” on behalf of teachers. But if you talk to kids, they harbor real anger: they’re pissed about climate impacts. They’re upset about job prospects. They’re outraged about AI being used to plagiarize writing and produce counterfeit artwork. Their relationships are being affected, and they’re seeing deepfakes in their feeds. They don’t like it because they’re experiencing real adverse impacts, right now.
Most adults I talk to misunderstand how keyed in kids are to AI. Kids have been having conversations about AI from a young age. I’ve done a lot of focus groups with kids, talking with school districts, students, parents, and teachers. When I ask, “What are you most excited about?” kids as young as fifth grade express apprehension. “Maybe a robot will do my chores… but that might make me very lazy.” When I ask, “What are you worried about?” they share a litany of concerns: “What if an AI doctor makes a mistake in my care? What if deepfake videos spread untruths about me? What job will I do? How will I buy food if I can’t find one?”
These are the worries of 10-year-olds. It’s a call for all of us to begin any conversation about AI literacy by talking to kids and seeing where they’re at. We need to understand what they are excited about, and why they are afraid.