Recently, the nation’s two largest school districts hit pause on artificial intelligence. New York City placed a one-year moratorium on generative AI for preschool through eighth grade, and Los Angeles blocked student access on district devices.
Six days later, researcher Jacob Coxon resigned from Anthropic, posting on his X account that AI companies are “racing straight to self-improving superintelligence and gambling with our lives.” And at the UN Security Council, Sam Altman of OpenAI and Dario Amodei of Anthropic repeated calls to slow development, urging international leaders to cooperate in order to keep AI under human control.
These stories share a common concern about AI’s development: speed. The growth of AI’s capabilities is outpacing the governance to manage its risks. Consider mathematics. In 2021, leading AI systems could solve fewer than 1 in 14 problems taken from challenging high school math competitions. Four years later, an AI model earned a gold-medal score at the International Mathematical Olympiad, the premier competition for pre-university mathematicians.
Get stories like this delivered straight to your inbox. Sign up for The 74 Newsletter
Previous transformative technologies required new infrastructure. Electricity needed grids. Automobiles needed roads. The internet needed networks. Those systems took decades to build, giving society and government time to adapt. But generative AI arrived on infrastructure that already existed. Its debut is leaving everyone a bit breathless.
The impulse to pause is understandable, and appropriate for the youngest students. But teachers and school leaders should be allowed to use AI in their own work while districts determine what responsible student use should entail.
The reason is simple: You cannot govern what you have not used. Many of AI’s benefits and pitfalls emerge through careful use in real learning environments. No committee can anticipate them all from a conference room.
At ASU Prep, a network of physical, digital and hybrid schools, early AI experimentation yielded insights that the leadership team would never have gleaned from the sidelines. When they invited staff to identify internal challenges AI might address, ideas flooded in.
For example, teachers had to consult multiple curricular documents, calendars and testing schedules to determine what students needed to learn, when to teach it and how each skill connected to the next. ASU Prep’s Strategic Solutions Team, composed of technology, engineering, curriculum, instruction and design specialists, used AI to organize that information into a single planning resource.
The resulting instructional guides show the sequence of skills and common points of struggle to look for in the classroom. Teachers, instructional coaches and school leaders can now work from the same information to identify gaps, discuss progress and target support.
In just one year, the team reviewed 22 staff-generated ideas, moved six into use and evaluated more than 10 commercial AI tools with employees. But it could not sustain that demand. ASU Prep is now creating departmental AI task forces so employees closest to the work can identify and test promising uses, while a central review process safeguards privacy, security and quality.
Experiences like these led ASU Prep to adopt a three-part test for every proposed AI tool or project: What does it make easier? What does it improve? What might it weaken or damage?
The answers can be two-sided. AI may make an assignment easier while weakening productive intellectual struggle. It could deepen personalized learning but expose sensitive student data. Overreliance on AI feedback could displace vital human interaction.
Thoughtful instructional design can help. Khan World School, an online honors high school operated by ASU Prep, uses a Socratic model in which students explore, present and defend positions. They may use AI in their research, but live discussions challenge their assumptions and send them back to reconsider the evidence. AI becomes a companion to student thinking rather than a replacement.
Such tradeoffs become visible only in practice.
AI is also challenging longstanding assumptions about how schools are organized. When students can receive patient, personalized explanations and practice at any hour, grouping them strictly by age and moving them through the same schedule may make less sense.
The teacher’s role may place less emphasis on delivering information and more on cultivating curiosity, judgment and independent thinking — especially helping students understand how human reasoning, shaped by experience, values and responsibility, differs from pattern-based AI outputs.
For these reasons and more, AI literacy cannot be reduced to great prompt writing or prohibited behaviors. Students need training to ask better questions: Is this true? What assumptions are embedded here? Who built this tool? What incentives shaped it? Am I using AI to extend my thinking or letting it think for me?
Teachers who have never seriously used these tools cannot truly help students answer such questions.
None of this argues for unfiltered embrace. Limited pilot programs should be supported by educators and operations staff to monitor data security, set age-appropriate expectations and identify what is strengthened or diminished. Pilots cannot anticipate every consequence, but they can help schools develop the habits and processes needed to detect problems early, respond quickly and revise their approach as the technology changes.