Schools Rush Into AI With Thin Evidence, No Policy

School districts across the United States are pouring money into artificial intelligence tools — from social-emotional chatbots to humanoid robots — even as rigorous evidence of their effectiveness remains scarce and formal policy guidance is largely absent, according to an NPR investigation published Monday.

In Dayton, Ohio, students at an alternative school spent weeks confiding in “Jordan,” a chatbot designed to present as angsty and to solicit advice from students rather than dispense it. In Wichita, Kansas, officials have rolled out AI tutoring in some high school math classes. And in a rural corner of upstate New York, a district spent nearly $58,000 on a humanoid robot before community backlash paused the plan.

The rapid, scattershot adoption has experts warning of a “Wild West” with few guardrails. “It feels very Wild West in terms of the randomness,” said Robin Lake, director of the Center on Reinventing Public Education, which tracks AI adoption in K-12 schools. “We need state and federal leadership.”

A Research Base That Lags Far Behind Deployment

The most comprehensive effort to date to assess the evidence — a Stanford University review released in March — analyzed more than 800 academic papers and found only 20 high-quality causal studies examining how AI tools affect students or educators. The report’s authors noted that the repository has since grown to more than 1,100 papers, but the share of rigorous research remains tiny.

The findings, such as they are, point to a central concern: a “durability gap.” Students tend to perform better on tasks when they can use AI, but results are mixed when the tools are taken away. The Stanford review frames the question bluntly — whether AI tools are helping students complete tasks, or helping them develop durable learning and skills.

The review also found that tool design matters. AI systems built with pedagogical guardrails — tutoring programs that give hints or guide reasoning — show more promise than general-purpose chatbots that simply supply answers. There are no high-quality causal studies of student AI use conducted in U.S. K-12 classrooms, and very little research on equity, student wellness, or social development.

That evidence vacuum leaves districts improvising. Mike Kentz, an external consultant hired by Dayton to design and evaluate its chatbot pilot, said the district gathered enough positive signals to continue testing — but not enough to scale up.

“Every test, every experiment that I try to measure in some sort of concrete way, I can only get as far as perception,” he said.

Katelyn Schoenhofer, the AI specialist for Wichita Public Schools, described the same problem from a different angle. “We wrestle with what does success really mean when it comes to AI use? And what metric do we want to use in order to evaluate that? And we have not come to an answer yet,” she said.

A Chatbot That Asks Students for Advice

Dayton’s experiment offers a window into how far some districts are willing to go. The district paid the ed tech company SchoolAI $72,600 last academic year for multiple AI tools, including Jordan, and renewed the contract for the current year at $78,650.

Unlike the agreeable, sometimes sycophantic AI chatbots common elsewhere, Jordan is designed to seem troubled. “Lately, I’ve been feeling like what’s the point of any of this?” the chatbot told students during NPR’s visit. “I’m not failing. I just don’t care. You ever feel like that?”

Rather than asking students directly about their emotions, Jordan asks for advice and then grades the responses for engagement and authenticity. Chat logs are viewable by school employees, an external evaluator, and SchoolAI staff, according to administrators, who said families and students were notified.

One sophomore, identified only as K to protect his privacy, described a genuine attachment. “I built a little relationship with him,” he said. “It’s Jordan — like, I’m texting somebody.”

Superintendent David Lawrence defended the district’s willingness to test the technology on vulnerable students. “I think when the history is written, it’ll say that ‘Hey, they were really cutting-edge, innovative and ahead of the curve in the interest of children,’” he said.

Backlash in New York, Bans in the Nation’s Largest Districts

Not every community has welcomed the experiments. In Salamanca, New York, a district serving many Native American students on Seneca Nation territory paid the robotics company Realbotix nearly $58,000 for a humanoid robot nicknamed “Sally.” Superintendent Mark Beehler had planned to deploy it in high school classrooms this fall.

Concerns about student data privacy — and about Realbotix’s corporate ties to a sister company that makes hyper-realistic sex dolls — quickly mounted. “A robot built by a company associated with sex dolls has no business in our classrooms,” said Melinda Person, president of New York State United Teachers. “Our students don’t need robots. They need real relationships with caring adults.”

By the end of July, the district had put the pilot on hold. Beehler insisted the pause was not a retreat from technology. “There is no possible way a robot can replace a human in a school,” he said. “Teaching is a human-to-human process.”

The country’s two largest school systems have moved in a different direction. New York City Public Schools, which serves about 600,000 students in preschool through eighth grade, imposed a one-year pause on student-facing generative AI beginning this school year, allowing only limited high school use for AI literacy and career readiness. Teachers are barred from using AI for grading and decisions about students’ graduation or promotion.

The Los Angeles Unified School District followed suit, blocking generative AI tools on district-issued devices for roughly 378,000 students through web-filtering software called Lightspeed. The announcement, made at the first meeting of a new AI committee, surprised board members and parents.

“This is an emerging technology,” said Pia Sadaqatmal, LAUSD’s chief academic officer. “We will continue to be adaptive as needed, but we do know that the heart of it is really making sure that we keep teaching and learning at the center of our decision.”

Teachers Worry, and Public Opinion Sours

A June NPR/Ipsos poll of K-12 teachers captured a profession caught between utility and anxiety. Nearly three in four respondents said AI’s implications for education would outstrip past innovations like the internet or computers, and roughly eight in ten said schools should teach responsible AI use.

But a majority also see trouble. Fifty-four percent said AI makes it harder for students to learn critical thinking; 55 percent called it mostly a shortcut to avoid work; and nearly six in ten said it is eroding trust between students and teachers.

“What that tells me is that they are trying to navigate some very complex challenges in an environment that is already rife with mistrust,” said Mallory Newall, a senior vice president at Ipsos.

Many educators are improvising without institutional support. Among teachers whose schools provide AI software, only 35 percent said they had a formal policy on teacher use. About half said their school offered no AI guidance — or that they were unsure what it was.

Rebecca Winthrop, a senior fellow at the Brookings Institution who researches AI and education, argues that caution is warranted. “I don’t think you should just sort of wildly experiment, especially with really marginalized communities,” she said, warning of “cognitive stunting” — diminished learning potential and altered socialization. She does not believe anyone under 18 should have access to general-purpose chatbots like ChatGPT and Claude.

A Student-Written Blueprint Fills the Void

With adults deadlocked, some students have tried to write the rules themselves. In late July, 98 high schoolers from all 50 states gathered at a replica of the U.S. Senate chamber in Boston to draft and debate what they called the Students First Act, which passed overwhelmingly, 82-16.

The student-written bill would require schools to teach AI literacy as soon as students gain access to classroom devices, ban AI use on graded tests, and allow teachers to require an oral defense of any project from a student suspected of inappropriate AI use. It is not law — but AASA, the School Superintendents Association, plans to circulate the text to its 10,000 members, many of whom are writing AI rules right now.

“We missed the mark with social media and phones,” said Hunter Wurzel, a high school senior representing Ohio. “That’s been a huge talking point already — how we failed to implement those things into education and teach the youth how to use them.”

An Expensive, Unsettled Bet

The spending continues regardless. A San Diego charter chain, Altus Schools, spent $500,000 on two ChatGPT-enabled humanoid robots named Ameca — an investment that drew sharp criticism from researchers. “The physical robot in the room is just complete bullshit — a show, a charade, a spectacle,” said Neil Selwyn, a Monash University professor who has studied technology in classrooms for decades.

Wayne Holmes of University College London was equally blunt about claims that such tools can teach. “It’s a parody of education,” he said. MIT’s Sherry Turkle warned of a broader trap: “Generative AI is a marvel,” she said. “But the fact that it’s a marvel means that we’ve gotten into a kind of delirium with it. It’s not hallucinating, we are.”

Even the optimists acknowledge the uncertainty. Jay Garrity, Altus’ head of instructional services, called the robots “a long-term investment.” “We’re still at the beginning phases, but we are optimistic,” he said.

What happens next depends largely on whether policymakers step in. In the meantime, districts are writing checks, teachers are improvising, and students like K are forming bonds with software that no one has fully decided how to measure — or whether it belongs in a classroom at all.