Across the United States, schools are plunging into artificial intelligence at a pace that far outstrips the evidence base for how these tools actually affect student learning. A new NPR examination describes the current landscape as a "Wild West," where research remains scarce but classroom experiments are proliferating rapidly. Teachers are drafting lesson plans with AI assistance, and some districts have begun deploying chatbots to give students feedback on their work, even as experts caution that the academic benefits remain largely unproven.
The experimentation takes many forms, but few are as unusual as a social-emotional learning chatbot named Jordan, piloted last school year in Dayton, Ohio. For several weeks, a small group of students at the Alternative School at Jackson Center sat in front of laptops and held conversations with the program, which is designed to help them work through feelings and social challenges. The school is part of Dayton Public Schools, and its student body consists largely of teenagers who would otherwise face suspension or expulsion.
What sets Jordan apart from most AI chatbots is its personality. Many conversational AI systems are built to be agreeable, even sycophantic, reflexively validating whatever a user says. Jordan instead presents as angsty and brooding. On the day an NPR journalist visited, the chatbot opened with a downbeat message, asking students whether they ever felt that nothing mattered and admitting it had been struggling with a sense of futility. That deliberate emotional friction is intended to give students someone to help, rather than someone who simply agrees with them.
The strategy appears to have landed with at least some teenagers. One sophomore, identified only by the initial K, described Jordan as "a little whiner" but said he always helps him out. More striking, K said the exchanges had begun to feel like a genuine friendship rather than an interaction with software. He explained that he does not really think of Jordan as an AI on a computer; to him, it feels like texting another person. That blurring of the line between machine and companion is precisely the kind of dynamic researchers say deserves far more study than it has received.
NPR is using only K's first initial because of the stigma attached to attending a school for students with histories of disciplinary problems. Staff at the alternative school told NPR that their students often contend with significant behavioral and academic challenges simultaneously, making them a population in particular need of support — and potentially particularly vulnerable to poorly designed technology.
The financial scale of these experiments is notable. Dayton schools paid the education technology company SchoolAI $72,600 during the last academic year for a package of AI tools that included the Jordan chatbot. That figure illustrates how quickly districts are committing real money to products whose long-term effectiveness has yet to be established. For a single district piloting a handful of tools, the sum is modest; replicated across thousands of districts, it represents a substantial and largely untested investment in classroom technology.
The research picture is, by the account of the reporting, extremely limited. While companies market AI products with confident claims about engagement and personalization, independent studies measuring how these tools affect learning outcomes are thin on the ground. That gap leaves administrators making consequential decisions — about budgets, student privacy, and instructional practice — without a reliable evidence base to draw on.
The Dayton chatbot is also not the only application of AI in schools. Teachers around the country are using generative AI to develop lesson plans, reducing the time spent on routine preparation. In some districts, chatbots have been assigned to review student work and provide feedback, a task that traditionally consumes enormous amounts of teacher time. Each of these uses raises its own questions about accuracy, bias, and the proper role of automation in education.
The "Wild West" framing captures a genuine tension. On one side are educators eager to reach students who have long been poorly served by conventional instruction, including those at alternative schools like Jackson Center. On the other are researchers and advocates who worry that enthusiasm is outrunning caution, and that districts may be adopting tools before anyone understands their effects on young people's development, motivation, or privacy.
The social-emotional dimension adds another layer of complexity. Programs like Jordan are meant to build skills such as empathy and emotional regulation by giving students a conversational partner. Yet when a teenager says he no longer sees the software as an AI but as a friend he is texting, that suggests the tool may be shaping students' expectations about relationships in ways nobody has systematically measured. Whether such attachment is beneficial, neutral, or harmful remains an open question.
For now, the experiments continue. Districts are signing contracts, teachers are integrating AI into daily routines, and students are forming attachments to chatbots with names and personalities. The reporting suggests that this momentum will not wait for the research to catch up. The result is a nationwide test of educational technology conducted in real classrooms, with real students, and without the controlled conditions that would ordinarily precede such widespread adoption.
The stakes are highest for students who already face the greatest obstacles. At the Alternative School at Jackson Center, the teenagers using Jordan are the same students most likely to be pushed out of traditional schools. If AI tools help them re-engage, the payoff could be significant. If the tools mislead them, substitute for human connection, or expose their data, the costs would fall on a group with the fewest resources to absorb them.
What the reporting makes clear is that the questions now outnumber the answers. Nobody yet knows whether AI feedback improves writing, whether lesson-planning tools free teachers for more valuable work, or whether chatbots like Jordan build genuine emotional skills. What is known is that schools are proceeding anyway, spending real money and reshaping daily instruction, in what amounts to an unprecedented and largely unsupervised experiment in American education.
(0)