Schools Are Rushing to Adopt AI in Education — Without Evidence or Policy

The Education System Is Running an Unprepared AI Experiment

AI tools are entering classrooms at a remarkable pace, yet a troubling reality persists: most schools have launched this large-scale experiment without reliable evidence or clear policy guidance. As discussions on Hacker News have highlighted, education decision-makers are often being pulled along by technological momentum rather than making choices grounded in verified learning outcomes.

This "deploy first, ask questions later" attitude essentially turns students into beta testers. When a school decides to introduce an AI tutoring system, automated grading tool, or generative AI writing assistant, few administrators can point to solid research demonstrating that these tools actually improve learning outcomes — or at least do no harm.

The Core Risks of an Evidence-Free Approach

Teaching Effectiveness Remains Largely Unverified

The history of educational technology is littered with "silver bullet" promises that ultimately fell flat. From early interactive whiteboards to MOOC platforms, many highly anticipated technologies failed to deliver the expected leap in learning outcomes. AI tools face the same test — whether they enhance students' critical thinking or instead foster tool dependency and erode independent thought remains an open question.

More troubling still, generative AI outputs can contain misinformation (the so-called "hallucination" problem). If teachers and students lack the ability to detect these errors, the technology could actively spread incorrect knowledge. Deploying these tools at scale without systematic evaluation carries real and concrete risks.

What is "Hallucination"? Hallucination is a systemic flaw in large language models — the tendency to generate content that sounds plausible but is factually wrong, fabricated, or internally contradictory, all delivered with apparent confidence. The root cause is that these models operate by predicting the statistically most likely next word, rather than truly "understanding" or "retrieving" facts. In educational settings, this flaw is particularly dangerous: students have not yet built a robust knowledge base and struggle to evaluate the accuracy of AI outputs. When an AI produces a wrong answer in fluent, logically structured language, even teachers can be misled. Mainstream AI tools such as ChatGPT and Gemini cannot eliminate hallucinations entirely — only reduce their frequency through technical means. This means that using these tools for knowledge instruction requires verification mechanisms, rather than treating their outputs as authoritative sources.

The Policy Vacuum and Its Ripple Effects

The absence of clear policy creates another problem: a collapse of consistent standards. Which use cases are acceptable? Does submitting AI-assisted work constitute academic dishonesty? How is data privacy protected? What happens to minors' personal information once it has been collected by commercial AI products? Most schools have no clear answers to these questions.

When policy is absent, practice defaults to individual teachers' discretion — meaning that dramatically different approaches can coexist within the same school, compounding inequity and confusion.

A note on data privacy: In educational settings involving minors, data privacy intersects with multiple layers of law. In the United States, COPPA (Children's Online Privacy Protection Act) prohibits collecting personal data from children under 13 without parental consent, while FERPA (Family Educational Rights and Privacy Act) strictly limits the sharing of student education records. Yet many commercial AI education products use vague data terms that may allow user data to be used for model training or commercial analytics — and school administrators signing procurement agreements often lack the expertise to scrutinize these clauses. Similar regulatory gaps exist in China. When commercial companies continuously accumulate data on students' learning behaviors, answers, and even emotional responses, the potential risks far exceed those of typical consumer data breaches — because this data could profoundly shape long-term individual profiles and assessments.

Why Schools Struggle to Make Rational Decisions

The pressures on schools are layered. On one side, parents and society expect schools to "keep up with the times" — not using AI can appear synonymous with falling behind. On the other, edtech companies aggressively market their products, framing AI as a cure-all for teacher shortages and the promise of personalized learning. Caught between commercial salesmanship and social anxiety, school administrators find it easy to make decisions without careful evaluation.

Educational research also operates on long timelines — genuinely verifying a technology's long-term impact on learning outcomes may require years of longitudinal study. Yet the pace of technological change vastly outstrips research's ability to keep up, creating a structural contradiction: by the time evidence arrives, the tools have already been superseded.

The MOOC cautionary tale: MOOCs (Massive Open Online Courses) offer a telling precedent. Around 2012, when platforms like Coursera and edX launched, widespread predictions held that traditional higher education would be disrupted — The New York Times even declared 2012 "The Year of the MOOC." Subsequent research found, however, that completion rates consistently fell below 10%, and that the primary beneficiaries were already highly educated adult learners — not the populations most in need of educational resources. This episode reveals a recurring pattern in edtech: access to technology does not equal learning outcomes, and technical solutions that ignore pedagogy, learning motivation, and social support rarely deliver on their grand early promises. The discourse surrounding AI education tools today bears a striking structural resemblance to the MOOC hype cycle — and warrants similar caution.

A More Responsible Path Forward

None of this argues that education should reject AI. The reasonable approach is to conduct small-scale, controlled pilots with clear evaluation frameworks, collect data, and then decide whether to scale. At the same time, education authorities should move quickly to establish framework policies governing AI use — clarifying data protection, academic integrity, and the boundaries of appropriate application.

For educators, helping students develop a critical understanding of AI tools may be more important than simply teaching them to use those tools. Ensuring students understand AI's capabilities and limitations is the foundational competency for navigating this technological era.

Conclusion

Education shapes the development of the next generation — any significant change should be grounded in evidence and accountability. The current state of schools "experimenting" with AI reflects a serious disconnect between the pace of technological deployment and the readiness of institutions and research to keep up. Beyond enthusiasm for new technology, the education system urgently needs calm evaluation mechanisms and clear policy guardrails — to avoid making an entire generation the unwitting participants in an insufficiently justified experiment.