Every debate about AI in education arrives at the same sentence sooner or later: it’s just a tool, and everything depends on how you use it. I’ve written versions of that sentence myself.
Then I went back to Peter-Paul Verbeek’s short introduction to mediation theory, published in Interactions in 2015, and the sentence stopped working for me. Verbeek wrote this piece seven years before ChatGPT existed. It still explains our current moment better than most of what gets published about AI today, and it hands educators a vocabulary we badly need.
The Three Stories We Tell About Technology
Verbeek opens by sorting our habitual thinking about technology into three approaches. The first treats technologies as extensions, neutral instruments that help humans do what they already intended.
He cites Joe Pitt’s defence of this view, the “guns don’t kill, people kill” position, where responsibility lives entirely with the human and the artifact stays morally silent. The second approach is dialectical.
Technology appears here as an opposing force, something that alienates and overpowers, and humans must find ways to free themselves from it. Verbeek places the Marxist critique of mechanization in this camp, along with contemporary worries about what information technologies do to our cognitive skills.
Most AI-in-education commentary still runs on these two scripts. The tool people insist AI is neutral and pedagogy decides everything. The resistance people insist AI erodes thinking and must be fenced off.
Verbeek’s argument is that both scripts fail for the same reason: they place humans and technologies in separate spheres, then ask how one affects the other. His third approach, the hybrid view, starts from a different observation.
Humans and technologies aren’t pre-given poles with an interaction running between them. They emerge from the relation itself, each helping to constitute the other. Our perceptions, our practices, and our ways of living take shape in close contact with our technologies, and the technologies take their meaning from us.

Why AI Is Not a Neutral Tool
The core of the piece is technological mediation, a concept Verbeek develops out of Don Ihde’s postphenomenology. Technologies mediate the relation between people and their world. They help decide how the world shows up for us and how we act in it.
Verbeek’s examples are deliberately ordinary and deliberately profound at the same time. The reality of a star is mediated by telescopes. The health of a fetus is mediated by ultrasound, and that mediation feeds directly into the ethical decisions parents face. Printer default settings decide, without anyone noticing, how much double-sided printing happens in an office.
Ihde’s four types of human-technology relations give the framework its analytical teeth, as Verbeek presents them. We embody technologies like phones and microscopes, looking through them at the world.
An MRI scan works differently: the human reads the world off the technology’s representation, which Ihde calls a hermeneutic relation. Interaction with a quasi-other, the way we deal with an ATM or a robot, forms a third type he names alterity. And some technologies do their work from the background, structuring experience without ever being noticed themselves.
Generative AI belongs mainly in the hermeneutic column, and I find this the single most useful insight in the paper for educators. When a student asks a chatbot to explain photosynthesis or summarize a policy debate, that student is reading the world through the model’s representation of it.
The representation carries the training data’s emphases, gaps, and biases. A student in a hermeneutic relation with AI isn’t operating a calculator. They’re inheriting a worldview, one query at a time. Alterity is in the mix too, since students increasingly treat chatbots as conversational others, but the hermeneutic dimension is where the pedagogical stakes concentrate.
The Influence Grid Every Teacher Should Know
Verbeek adds a second layer that I would place in every teacher education course: the character of technological influence, drawn from the work of Nynke Tromp and colleagues. Influence varies along two dimensions, visibility and force.
Strong, visible influence is coercive, like a car that won’t start without a seatbelt. Persuasion is weaker but still visible, an energy meter giving feedback on your consumption. The hidden kinds are the ones Verbeek finds most common and most interesting.
A coffee machine placed in a hallway to spark conversation exerts seductive influence, mild and unnoticed. An apartment building with no elevator exerts what he calls decisive influence, strong, hidden, and completely effective at putting everyone on the stairs.
Now run AI in classrooms through that grid. Almost none of AI’s influence on learners is coercive. Nearly all of it is seductive or decisive. The confident, fluent tone of a chatbot seduces students into trust, which connects directly to what Lee et al. (2025) found about confidence in AI predicting reduced critical thinking among knowledge workers.
Interface defaults decide what students see first and what they never see. A model’s refusal patterns and safety rails decide which questions receive real answers. None of this announces itself as influence, and that is precisely Verbeek’s point about hidden mediations: they’re the everyday ones, and they escape scrutiny because nobody perceives them as influence at all.
What Mediation Theory Means for AI Pedagogy
The strongest lines in the paper come near the end. “Designing things is designing human existence” (p. 29), Verbeek writes, and later, “[d]esigners materialize morality” (p. 31). If that’s true of speed bumps and printer defaults, it’s true tenfold of AI systems that mediate how students read, write, and reason.
Three implications follow for education, and each one strengthens arguments I’ve been building across this blog.
The first concerns teacher agency, which expands under this framework. Teachers aren’t just users of AI tools; they design human-technology relations every time they build an AI-integrated task.
That gives a deeper justification to the critique-and-adapt practice van den Berg and du Plessis (2023) proposed for AI-generated lesson plans. A teacher who evaluates an AI output before it reaches learners is reading a mediation, and that reading is professional work of the highest order.
Transparency is the second, and mediation theory turns it into a pedagogical principle, not merely a compliance requirement. Mike Perkins (2023) located academic integrity in whether AI use is made clear.
Verbeek’s framework explains why that’s exactly the right place to locate it: hidden mediations are the ones we can’t engage critically, so disclosure is what converts a hidden influence into a discussable one.
The third implication gives AI literacy a philosophical foundation. Verbeek argues that users, designers, and policymakers should be enabled to “read, design, and implement technological mediations” so they can deal critically and creatively with powers that would otherwise stay hidden (p. 31).
Written in 2015, that sentence remains as good a definition of critical AI literacy as anything published since. It’s also the theoretical spine behind the visual AI literacy work I’ve done on teaching students to see how AI systems represent the world.
The paper’s age needs one caveat. Verbeek was writing about Google Glass and brain implants, and his newer categories, cyborg relations and immersion, read differently a decade on.
What he treated as exotic in 2015 became mundane by 2026, since ambient AI in classrooms is an immersion relation by his own definition. A framework that absorbs technologies its author never saw is showing its strength, not its age.
Verbeek closes with a line worth memorizing: “Human freedom cannot be saved by shying away from technological mediations, but only by developing free relations to them” (p. 31). For educators, the translation is direct. We can’t teach students to escape AI’s mediations. We can teach them to read them. That’s the work.
References
Ihde, D. (1990). Technology and the lifeworld: From garden to earth. Indiana University Press.
Verbeek, P.-P. (2015). Beyond interaction: A short introduction to mediation theory. Interactions, 22(3), 26-31. https://doi.org/10.1145/2751314




