When something is wrong in the classroom but no one knows how to say it, there is a certain kind of quiet anger that builds. An 11th-grade U.S. history teacher in Bensenville, Illinois, saw that frustration play out slowly as a group of her students spent almost two weeks following a ChatGPT-directed research thread that led them to a yellow fever outbreak that happened many years before the time period they were supposed to be studying. The students were sure of themselves. The AI seemed sure of itself. Everyone moved forward. After that, everything broke down.
The mistake is not what makes that story stick. Everyone in school always takes the wrong turn. The faith is what makes it stick. The group didn’t know how to question what the tool was telling them because no one had taught them that. That AI “can generate information instantly, but it can’t tell my students whether that information is meaningful, relevant, or true” is also a point that Torres makes that isn’t really talked about in the policy debate.
A lot of the talk about AI in schools has been about who can use it and how it should be supervised, like whether school districts allow it and what rules should be in place to protect student privacy. These are good reasons to be worried. There are more than 134 AI-related education bills in 31 states this year alone.
These bills cover a wide range of topics, from limiting AI use to students under the age of six to protecting data to outright banning AI in all school settings. That activity in the legislature shows a real sense of urgency. But most of it is still based on managing a tool that already exists and was made by people who didn’t have eleven-year-olds, students who are having trouble reading, or students who have never been told how a language model works in mind.
The way the Daily Herald puts it—that students should help make responsible AI frameworks, not just use them—cuts through a lot of that. The argument isn’t as loud as the ones that are getting a lot of attention, but it might last longer. Torres helped get a law passed in Illinois that set the first statewide guidelines for AI in schools. Her next project is all about adding student voice to the framework itself. That’s a difference that matters. A policy that was made with students in mind is not the same as a policy that was made with students in mind.
There’s a connection that seems important to note. When social media came to schools, adults spent years trying to control it after the fact, with policies that banned cellphones, rethought laptops, and changed rules. Some people have already said that this pattern will happen again with AI because it happened so often. One difference is that AI is harder to keep in a safe place. This person from the school superintendents’ association said that it’s “only one personal device away” from being in the classroom, no matter what the school policy says.
Because of this, the design question is more important than it might seem at first. Since 92% of students around the world used these tools in 2025, limiting access might not be the most important thing to do. Instead, shaping the experience might be more important. And it’s hard to make an experience good if the people who will be most affected by it didn’t help plan it.
Now that I think about it, this seems almost obvious. Most people who work on AI have never thought to ask students about how they learn, where they get stuck, and what kinds of feedback they find helpful and annoying. When you design with them, you don’t have to give up control of the technical architecture. For this to happen, their point of view needs to be gathered before the system is built, not after it’s been used in 40 million classrooms.
Sofia’s group finally began again. Their sources were called into question this time. They compared their points of view. Torres said that the final project was not perfect, but it was theirs. It’s not as simple as that phrase might seem.
