Rebecca Winthrop

Like this is how it works. It's not some magical thing. It's not another human being. So when kids get older, we need to teach them about that. And then they need, when they get older, they need to start playing with it, playing with it, using it. But my huge caveat is with AI that is designed for kids. Right now, there is a spring fling race by the large AI labs to get students to sign up.

You've got ChatGPT giving two months free of GPT Plus. Then you got XAI come in, two months free for SuperGrok. And then Google, not to be outdone, is like, well, you can get a year free and I'll give you two terabytes of storage. And these are largely for college students. And Google just made Gemini available for kids through Parents with Family Plan. And they are racing.

You've got ChatGPT giving two months free of GPT Plus. Then you got XAI come in, two months free for SuperGrok. And then Google, not to be outdone, is like, well, you can get a year free and I'll give you two terabytes of storage. And these are largely for college students. And Google just made Gemini available for kids through Parents with Family Plan. And they are racing.

to get allegiance of young kids. This is terrible because those products are not designed for children and for learning.

to get allegiance of young kids. This is terrible because those products are not designed for children and for learning.

So I think that AI has real potential for very specific use cases, particularly around access gaps. And in Nigeria, what was done was after school, twice a week, an AI tutor helped kids learn English. And it was for six weeks, which is not long. It was June, July, I think. And it was a randomized controlled trial.

So I think that AI has real potential for very specific use cases, particularly around access gaps. And in Nigeria, what was done was after school, twice a week, an AI tutor helped kids learn English. And it was for six weeks, which is not long. It was June, July, I think. And it was a randomized controlled trial.

We're still waiting for all the evidence to come through, but 0.3 standard deviations, which is pretty good, equivalent to maybe two years of average sort of English learning. And we see that difference with other technologies, too. It doesn't have to be Gen AI. It can be rule-based AI. It could be predictive AI.

We're still waiting for all the evidence to come through, but 0.3 standard deviations, which is pretty good, equivalent to maybe two years of average sort of English learning. And we see that difference with other technologies, too. It doesn't have to be Gen AI. It can be rule-based AI. It could be predictive AI.

We've seen sort of similar benefits, for example, in Malawi, teaching literacy and numeracy to kids with offline tablets where teachers have maybe 80 to 100 kids in a class. And each kid is having sort of a personalized adaptive learning experience. That is hugely beneficial as well. So that's one use case. Another use case that I think is really great is neurodivergent kids. Super helpful.

We've seen sort of similar benefits, for example, in Malawi, teaching literacy and numeracy to kids with offline tablets where teachers have maybe 80 to 100 kids in a class. And each kid is having sort of a personalized adaptive learning experience. That is hugely beneficial as well. So that's one use case. Another use case that I think is really great is neurodivergent kids. Super helpful.

There's all sorts of kids that have different learning differences that struggle in school, don't have access to the specialists that they need, that would benefit greatly from being in a classroom where they could have a little assistant to help them navigate. My youngest son has dyslexia and they sort of read and write, text-to-speech, speech-to-text has been game-changing for him.

There's all sorts of kids that have different learning differences that struggle in school, don't have access to the specialists that they need, that would benefit greatly from being in a classroom where they could have a little assistant to help them navigate. My youngest son has dyslexia and they sort of read and write, text-to-speech, speech-to-text has been game-changing for him.

There's also use cases here in the U.S. You see AI being used and experimented around supporting wellness advisors who kind of fill the gap for school counselors in rural school districts, for example, where they don't have school counselors. which is actually an actual person, but AI is boosting that person's ability to have a helpful conversation with a kid.

There's also use cases here in the U.S. You see AI being used and experimented around supporting wellness advisors who kind of fill the gap for school counselors in rural school districts, for example, where they don't have school counselors. which is actually an actual person, but AI is boosting that person's ability to have a helpful conversation with a kid.

And it's bringing through tech a mental health resource into a community that didn't have one. So there's lots of use cases, actually, if done well, contained well, designed well, and we humans have our hand on the steering wheel.

And it's bringing through tech a mental health resource into a community that didn't have one. So there's lots of use cases, actually, if done well, contained well, designed well, and we humans have our hand on the steering wheel.

Exactly.

Exactly.

Absolutely. And the key is what you said, contain the system. We can't sort of just bring commercial tech into our schools and hope it will solve these problems. It has to have guardrails. We have to make sure that the data that it's being trained on is legit and not going to create harmful prompts for kids.