The push to integrate artificial intelligence into education is accelerating, but experts are raising a critical concern: schools are adopting AI tools faster than researchers can measure whether they actually improve learning. A convergence of new federal guidance, emerging research frameworks, and cautionary lessons from education history suggests that the next phase of AI in schools will depend less on the technology itself and more on whether educators, administrators, and policymakers ground their decisions in solid evidence.

What Does a Century of Education Technology Tell Us?

The pattern is familiar. In 1913, Thomas Edison predicted that film strips would replace textbooks within a decade. Radio broadcasters promised to bring the world's best teachers into every classroom. MOOCs (massive open online courses) were supposed to democratize higher education. None of these technologies transformed education the way their proponents claimed.

MIT education researcher Justin Reich, who interviewed 120 K-12 teachers and students across the United States about AI, explained the recurring cycle: "The first things that happen when teachers get access to new technologies is they use them to extend existing practices," he noted. "So they do whatever they were doing before but with the new technology." Teachers might project notes on a digital board instead of writing on a chalkboard, but the fundamental teaching method remains unchanged.

Reich identified three patterns that emerge consistently across a century of education technology adoption:

• Extension Over Innovation: Teachers initially use new tools to replicate old practices rather than fundamentally rethinking instruction, from chalkboards to whiteboards to digital projectors.
• Affluence Advantage: When benefits do emerge from new technologies, they disproportionately help already-educated and affluent students, widening rather than closing achievement gaps.
• Community Matters Most: Technology only improves learning when entire school communities have time to experiment, collaborate, and adapt together; isolated tool adoption rarely succeeds.

The implication is sobering: AI might modestly help some students learn some things, but it is unlikely to be transformational in the way tech evangelists describe.

What Does the Federal Government Say Schools Should Actually Do?

On August 20, 2026, the U.S. Department of Education released new guidance aimed at helping states and school districts make informed decisions about education technology, including AI tools. Rather than encouraging rapid adoption, the guidance emphasizes a measured, evidence-based approach.

The Department outlined five straightforward questions every education technology product should answer:

• Learning Problem: What specific learning problem does the tool solve?
• Timing: When should it be used in the curriculum or school day?
• Target Population: For whom should it be used, and are there students it might not help?
• Duration: For how long should students use it before outcomes are reassessed?
• Evidence: What independent evidence demonstrates that it actually improves student learning?

The guidance also stressed that schools must be willing to abandon tools that don't work. "When evidence shows a tool is not improving learning, schools must be willing to change course, and when repeated findings confirm persistent shortcomings, they should remove it altogether," the Department stated.

How Can Schools Implement AI Responsibly?

Shari Dubos, a principal education researcher at SRI International whose work focuses on improving STEM education access, described a practical framework for responsible AI integration. Rather than building yet another AI-powered tool, SRI developed what it calls the Safe and Accessible Data Interactions in Education (SADIE) project, a middleware layer that sits between students and AI-enabled learning platforms.

The SADIE approach addresses two urgent challenges: the growing data literacy gap among students and the rapid, often inaccessible introduction of generative AI into classrooms. "Rather than building just another AI-powered tool, SADIE focuses on developing a middleware layer that sits between students and AI-enabled educational technology products. The idea is to help keep all AI interactions on task, accessible, and meaningful," Dubos explained.

Dubos emphasized that AI will only advance classroom learning if educators deeply trust it. She outlined key principles for responsible AI adoption:

• Evidence-Based Selection: Schools should not use AI for its own sake, but only when rigorous research shows it creates real opportunities to tailor instruction to individual students' needs and learning trajectories.
• Teacher Support, Not Replacement: AI should reduce teacher workload in areas like lesson planning, formative assessment, and feedback, allowing educators to focus more on relationship-building and high-impact teaching.
• Accessibility and Design: AI tools must be accessible to all students, including those with disabilities, and must be thoughtfully designed with input from educators and students.
• Scaled Implementation: Evidence-based tools should be paired with strong implementation support and pathways for scaling across schools and districts.

Dubos noted that one of her most fulfilling projects involved evaluating the Kasi Learning System, an accessible, multisensory chemistry tool for blind and low-vision high school students. Using rigorous single-case design research, her team found that the tool increased student agency, independence, and sense of belonging. "That was huge, because students who are blind or have low vision often face real barriers in STEM courses like chemistry, which are so visually based," she said.

Dubos

Why Does the Gap Between Innovation and Evidence Matter Right Now?

The timing of these messages is significant. Schools are under pressure to adopt AI quickly, and vendors are eager to sell solutions. But without rigorous evaluation, schools risk wasting resources on tools that extend old practices rather than enabling new ones, or that primarily benefit already-advantaged students.

"We want everything we do to be evidence-based. We're not using AI for the sake of using AI. We're using it because we can observe how it creates real opportunities to tailor instruction to individual students' needs, interests, and learning trajectories in ways that have historically been difficult to achieve," Dubos stated.

Shari Dubos, Principal Education Researcher at SRI International

The federal guidance reinforces this message, urging states to focus on instructional value over recreational engagement and to pair innovation with educator judgment, transparency for parents, and a relentless focus on student outcomes.

For schools considering AI adoption, the lesson from a century of education technology is clear: the tool is not the transformation. The transformation comes from communities of educators, students, and families who have time to experiment, learn from evidence, and adapt together. AI can support that process, but only if schools resist the urge to deploy it first and evaluate it later.