- Thanks, Christiane.

Rebecca Winthrop, thank you so much for joining us once again.

- Thank you so much for having me back.

- You literally wrote the book about the disengaged teen, and this is something that I think a lot of parents and educators had been seeing.

They weren't really sure what to call it, and you kind of, you gave it a name, which is to say that you're concerned about motivation, sort of engagement and learning, and this is something that actually predates the concerns that surfaced during COVID.

So, but before we even get to AI and the chat bots and all of this, remind us of what it is that you were seeing that caused you to sort of ask these questions.

- My co-author and I, Jenny Anderson, did a deep dive on student engagement, which is their motivation, love of learning, paying attention, doing effort, because we were just finding that so many kids didn't like school.

It was fairly simple.

That was our question.

Why do so many kids not like school?

When the human beings have evolved to love learning, we are naturally programmed to love learning.

So what is it that sort of squashes the love of learning for lack of a better term out of kids as they progress along their school journeys?

We found that 75% of third graders say they love school, but by the time kids get to 10th grade, it's only 25% it's flipped.

- So tell us how the concerns about AI intersect with that.

And what is it again that you were seeing that made you wanna ask these questions?

- So what we found was that it wasn't so simple as kids are either engaged or disengaged in education and learning.

What we found is they show up in these four different modes.

Passenger mode, kids have physically gone to school, but they do not care about learning, so they've dropped out of learning.

You've got kids in achiever mode, they're trying super hard, they want to be perfect.

Resistor mode is what we always think of as a disengaged teen.

These are avoiding, disrupting, chronic absenteeism, not turning in their homework.

And then you've got kids in explorer mode.

And that was less than 4% of kids who said they regularly spent time in explorer mode in middle school and high school.

And that is the type of learning that prepares kids for an AI world.

They care about the learning journey, not just the outcome.

They're putting effort, they're resilient, if things are hard, they try again, they're curious, and they're interested in asking questions, not just getting answers right.

And with AI coming on, I am incredibly worried about making kids into a lot more passengers.

Wow.

The class of 2026 is the first generation to start and finish college with ChatGPT.

This is, you know, according to the tools, parent company, OpenAI.

So what is the problem with using ChatGPT?

I mean, your work identifies like specific things to be worried about.

What are the specific things to be worried about?

So the last year with my colleagues at Brookings, we ran a global task force and we were really looking at what are the specific benefits, what are the specific risks, how do they stack up against each other.

And part of why people feel worried is that chat GPT or any, I would say, general purpose AI chatbot or AI companion or friend that is built on these AI chatbots provides very, very easy answers, very easy ways to get work done without putting any effort or doing any of the thinking.

And I do hear parents say, "Well, I use it at work, so shouldn't my kids use it?

Aren't they being efficient and helpful?

Isn't it helping them if they use it in their homework?"

The problem is that literally homework or any type of learning activity, kids have to do themselves because that's how they build their critical thinking skills.

You build critical thinking skills like you learn to ride a bike.

You have to practice it over and over.

And I can't do it for my kids, you can't do it for your kids, no parent can do it for them.

Kids have to struggle, they have to make mistakes.

That is how we learn.

And if we give them a lot of shortcuts, we know from really interesting research that even includes mapping neurological activity in the brain, that kids' problem solving parts of their brain, their critical thinking parts of their brain are just not engaged while they're using ChatGPT to do homework, to write, to come up with answers.

How does the ChatGPT affect kind of the thinking part?

Because you can see the sort of advocates of the tool will say, "Well, you know, it's the first draft, and you're going to refine the first draft, and you're going to edit."

The second draft is the good part anyway, or the editing is really where the magic happens.

What's wrong with that thinking, and what did your study show about why that is not right?

So, I love that you brought this up, because I've been having a big debate with my family members who have kids, as well as my peers at work.

Lots of people have been saying, well, it's okay to use generative AI, could be chat GPT, could be whatever model you're using, to brainstorm.

As long as kids then do the real work of actually writing.

What we found in our task force work is that actually writing is thinking.

It is a way that kids train themselves to come up with ideas.

It's not the only way.

You could have long-form debates, but writing is a really good way to do that.

When you use AI to brainstorm, it short-circuits kids' own creative ideas.

We did find that there's lots of researchers out there, including those at Georgetown University who are leading this, who found that when students use AI to write important essays, I'm not talking about logistical emails, I'm talking about really writing to come up with ideas.

It undermines their creative thinking.

They are less creative, they come up with less creative ideas.

In fact, humans have eight times more creative ideas than AI, if you just look at writing essays about something meaningful to yourself.

The problem is though, that even though it undermines your creative process, and I would say you should brainstorm first on your own and write your first draft, no matter how bad it is.

And then at the end, use AI to help with your grammar and polish the flow.

I think that's the right order for really harnessing and exercising our creative thinking.

But the problem is that even when you do use AI to brainstorm, it sounds better.

So AI kind of tricks us.

That, you know, there was this huge data set looking at over 300,000 college essays for high school students.

- Yeah, I was gonna ask you about that.

- Yeah.

- Tell us about that.

- So this study led by Adam Green out of Georgetown University was a natural experiment.

They started eight years ago, the team looking at high school students who are applying to college and examining their college essay, their personal statement, and assessing how creative it was.

So they have eight years of data.

So they saw the difference pre-Chad GPT and post-Chad GPT.

And one thing they know for sure from looking at this data set in many different ways is kids are definitely using AI to write their essays because they look different.

And they found that humans who were assessing the essays judged them to be more creative because they had better, more sophisticated vocabulary.

But if you looked across all the ideas that young people were writing about, the ideas got a lot more similar, which is this really weird paradoxical thing that AI is doing to our creative thinking.

On the surface, it makes it sound good.

It sort of surface sparkle masks underlying sameness.

Did the sort of the AI companies respond to this study?

What do they say about it?

- I mean, I have, I talked to technologists, I interviewed technologists, I've asked technologists particularly about this idea of homogenization worry.

And the folks I've talked to have said, "Huh, they weren't super aware of it."

And then they were also trying to grapple with, well, part of the reason there's homogenization is of course they're trying to reduce the really terrible things that come out for safety out of AI chatbots.

So they're trying to eliminate some ideas that we don't want.

We really don't want kids going online and figuring out how to, you know, make a chemical weapon and bring it to their school, right?

So they say they're sort of caught between that, but ultimately what they say, really, when I talk to folks is they just don't know how the models really work internally yet, and they don't quite know why it's doing this.

But to me, I think the implications are exactly what you raised.

I worry about kids not being able to develop their own unique voice.

Let me give you the case from the AI companies.

Open AI argues that AI doesn't replace ambition, it amplifies it, allowing students to learn new skills, prototype ideas faster, and contribute in ways that once required far more resources.

I think what they're saying in a way is it democratizes something that elites have always had access to, which is consultants, for example.

The use of consultants to help the most privileged kids get into these places, right?

To sort of shape their journey as it were, and look at their essays and doing all that.

And that's something that less resourced people don't have access to.

What would you say to that?

So what I would say to that is that some ways of using generative AI can be good for learning.

And it is the difference between what I would call narrow AI use and wide AI use.

And narrow AI use can indeed help level the playing field a bit.

So for example, narrow AI use is when usually teachers or a tutor, though not always, sometimes students directly, are using an AI tool that has a very, very specific purpose.

So assistive technology, maybe you have dyslexia or dysgraphia or aphasia.

Aphasia is when you have communication problems.

One of the most moving examples of narrow AI use I've seen is kids with aphasia getting a synthetic copy of their voice, thanks to generative AI, and being able to communicate with their teachers and their peers in the classroom.

Incredibly transformative narrow AI use, or teachers using it to help free up some of their time.

Gallup and the Walton Family Foundation did a nationally representative survey and found that teachers in the US are saving on average six hours a week from automating some of this administrative work with generative AI.

So all of that is narrow AI use and very good and effective and does help democratize supports.

The problem is that most kids we talk to are not necessarily just accessing AI that is sort of narrowly deployed.

They're accessing what we would call wide AI use, which is general purpose frontier model AI chatbots.

So your chat GPTs and AI friends and companions.

And those are not designed for learning, not designed for kids.

And that is really where if kids are interfacing with them for a long period of time, like the kids going back and forth, developing their personal statements for their college applications, they are undermining their unique voice, their critical thinking ability, and even their ability to interact and relate to other people because the AI chatbots and friends are really programmed to agree with you all the time.

- To that end, according to a RAND survey, 62% of students reported using AI for homework by the end of 2025.

But it's interesting that 67% said they believed that AI could harm their own critical thinking skills.

But the other thing I wanted to sort of ask you about is another survey by Common Sense Media found that roughly one in three students said that they'd rather talk to an AI companion than a real person.

And as a person, the reason I'm sort of raising that is that one of the things you talked about in your work about the disengaged teen.

You weren't just talking about kids being checked out from school per se, but sort of being checked out from life, you know, from kind of learning how to live in the world.

And I just wonder what you make of that.

So this is one of the really worrisome, worrying statistics that keeps me up at night, because this is a relatively new technology.

One out of every three teens preferring to talk to a friend, an AI friend on their phone.

And the problem with the AI friend on their phone is that they are designed to be sycophantic.

I'll give you an example.

If I'm a teen and I tell my AI friend, "My mom is such a bummer.

She is making me clean my room."

This is hypothetical.

The AI friend would say, "I'm so sorry.

I am here for you.

You are so right.

This is terrible.

Yeah, this is terrible.

Come talk to me."

A real friend would say, "Dude, my mom makes me clean my room too.

What's your problem?

Get over it."

That friction, that interaction is what children and young people need to grow up and not only think independently, but live in this world and be able to relate to others and actually be able to have relationships with other human beings.

We know there was a study in Nature that showed even small amounts of interaction with AI friends and companions in adults, mind you, reduced the frequency of people repairing ruptured relationships.

So imagine what that is doing to kids.

The other reason that kids say they like talking to AI friends is really sad, which is they say they don't have anyone else to talk to.

So we need to really lean in on our relational infrastructure.

So you know, teachers and students when trusting relationships, which AI is fraying, by the way, we need to lean in there.

We need to lean in into community organizations and third spaces where young people can participate and build friendships and community.

And before I let you go, parents who might be listening to this conversation, and kind of have like this vague concern, don't know what to do about it.

What would you say?

Like, how would you advise them to even talk to their kids about it?

- Well, this is a topic that we've gotten so many requests for, that we don't usually do this with Brookings Reports, but we're making little parent tip sheets out of our Brookings Report because so many parents are desperate for guidance.

So people can go, go, they're freely available at the Brookings website, but conversation starters, talk to your kids about AI first.

Don't be hugely judgmental, see where they're at, see where their opinions are.

A lot of kids do not like AI and are quite skeptical and that percentage is growing.

Two, talk to them about where skills are built.

In the effort of doing hard things, you can become a master and face anything in life.

So if you want to be a fully developed human being who can weather the changes that AI brings, you need to be a really good learner.

And to learn to be a really good learner, you can't just do everything outsourced to chat GPT.

And so we have some of those conversation starters.

And then, you know, experimenting with AI, if kids are old enough, and they're using it, doing it side by side, there's certain things that could be great.

Maybe a young person has a great idea for a film they want to make.

Well, guess what, you can vibe code a film by talking to Gen AI in any of the creativity AI tool suites and it will create a short film.

You know, I think of it as the next generation of when I was in school, we would cut out pictures and make from magazines make collages.

My kids are making little documentary films with the iPhone.

You know, these kids are vibe coding films just from their brain and speaking to it, but it's their ideas that the technology is bringing to life.

It isn't, the technology isn't sort of subtly giving them ideas that they begin to interpret as their own.

.

Thanks so much for talking to us.