AI Won’t Replace Educators. But It is Changing How Students Learn

AI tutors – From a Kindergartener’s dinosaur question to universities and national education plans, AI is reshaping learning—but the evidence points to a divide between answer-giving tools that reduce retention and structured tutoring that can boost learning. The real bat

A Kindergartener climbs onto the scale. looks at the numbers. and asks a question that belongs in a dinosaur book—“what dinosaurs also weighed 50 pounds.” The moment is small. but it lands like a test of trust. I can’t answer it on my own. Then Claude helps. fast. and my son lights up as he hears that he weighs about the same as a juvenile velociraptor.

Later, the answer slips out of my memory. My son doesn’t forget it. He keeps the knowledge; I become the conduit.

That exchange—imperfect recall on one side. lasting learning on the other—is playing out in schools and colleges right now. Information is more accessible than ever, and students anywhere can ask an AI tool for an answer that sounds reasonable. It’s why predictions of the demise of traditional schools keep circulating.

Yet education has never been only about access. Students need more than instant responses. They need to assess the quality of what they’re reading and hearing. They need to recognize strong work and learn to connect ideas. They need to deal with the reality that people disagree—and that it’s normal.

That kind of learning depends on human relationships. the friction of real life that even the most “helpful” AI models can smooth over. The question now is sharper than whether AI replaces teachers. It’s whether AI supports genuine learning—or nudges students into “cognitive surrender,” accepting AI-generated answers with minimal scrutiny.

Learning results appear to turn on something students and parents rarely see: the kind of AI they’re using and how it fits inside the learning process.

When students rely on the standard, free versions of large language models, the dangers are greater. These models are designed to be helpful, so they tend to provide answers to the questions they’re asked. In that mode, brain activity and retained learning are lower when students work with AI in this way.

The picture shifts with tools built to scaffold learning and support in-person instruction. In an introductory undergraduate physics course. one study found students using a carefully designed AI tutor had twice the learning gains of those receiving active. in-person instruction. Memory and comprehension didn’t just improve because AI existed; it improved because the tutoring was designed to guide learning.

The physics study also warns that structured AI tutoring may not be appropriate for tasks requiring complex synthesis of multiple concepts and higher-order critical thinking. In other words, not every subject (and not every mental task) benefits from the same kind of AI involvement.

Some education leaders are trying to embed those distinctions into policy. Estonia’s education minister—overseeing an ambitious partnership with OpenAI to provide a custom AI platform in upper secondary schools—has described a blended model. Students use handwriting to form memories early in the learning process. then later use digital tools for feedback and AI-assisted learning. Estonia is not introducing AI in earlier grades, aiming to let students build foundational knowledge and skills first.

Even with a well-designed model, educators still have to make the difference between “learning” and “answering.” That’s where support for teachers becomes the decisive factor.

One study from Sierra Leone found that secondary school educators completed a one-day training before adding AI tools in the learning process. After that training, math learning gains were equivalent to more than a year of additional schooling. The implication is uncomfortable: AI alone doesn’t guarantee improvement. How teachers introduce it—and what they expect students to do with it—matters.

Google, OpenAI, and Anthropic all offer learning modes and other supports built around these ideas. But in practice, many features are opt-in, and they’re getting harder to find for non-enterprise users.

OpenAI, for example, launched “study mode” in July 2025 but quietly removed it from the standard ChatGPT interface this spring. The feature remains available to schools and systems with enterprise contracts.

Those contracts are expensive. but they influence what kinds of AI schools can actually use—especially the types educators want when leaders collaborate across systems and make similar asks of tech companies in procurement. The bottleneck isn’t only pedagogy. It’s purchasing power, and the terms attached to it.

That’s why the role of philanthropy keeps coming up: schools and educators should not be navigating these waters alone. Philanthropy can support training that respects teachers’ expertise. conduct independent research on what works. and strengthen advocacy that counterbalances the size of tech firms.

It can also help make enterprise contracts more affordable, and support the development of procurement standards that protect learning, student data, and educational institutions’ sovereignty over their own systems.

The idea isn’t new. Philanthropy has a history of trying to ensure the benefits of major learning shifts reach everyone. When compulsory schooling laws were passed at the turn of the 20th century. communities gained Andrew Carnegie’s 2. 509 libraries—many serving as classrooms—and Julius Rosenwald’s 5. 000 schools that educated a third of Black children in the rural South.

Zoom out further and the lesson looks even more familiar. German apprenticeship programs endure in part because, during the Industrial Revolution, German guilds adapted their models to fit an evolving economy rather than resisting change outright.

Today’s flood of information began with the printing press. It expanded access to texts and reshaped who could claim expertise. In earlier centuries, writing transformed curriculum, credentialing, and information exchange. If past transitions were survivable—and sometimes even educationally productive—there’s reason to believe this one can be too.

The facts point to a real dividing line: AI that answers can reduce retained learning, while AI that scaffolds—especially when teachers are trained and learning tasks are matched to the right tool—can deliver gains that look dramatically better.

Humans may not be as cool as velociraptors, but the stakes here aren’t about novelty. They’re about agency. Tech providers, educators, and philanthropy can shape what’s next for students—so the person who becomes the keeper of knowledge is not just an AI system, but the learner.

AI in education cognitive surrender AI tutors study mode OpenAI enterprise contracts Estonia education policy Sierra Leone teacher training undergraduate physics course learning gains educational procurement