Why Students Tell AI What They Never Tell Teachers
For decades, educational psychology operated on a quiet assumption: the learner is the variable, and the content is fixed. You standardize the curriculum, sequence the concepts, and trust that enough repetition plus a motivated student will produce understanding. The teacher serves as the bridge, the translator between fixed material and variable minds.
AI-assisted learning tools have started to crack that assumption open — not by replacing teachers, but by surfacing something uncomfortable. The way students interact with AI tutors tells us things about learning that traditional classroom observation never could. And what it reveals challenges some deeply held beliefs about pedagogy itself.
The Problem With One-Way Knowledge Transfer
Open educational resources — the kind CUNY Pressbooks champions — exist because knowledge hoarding is antithetical to public education. Make the textbook free. Remove the access barrier. Let anyone read, adapt, reuse. The philosophy is right. But access is not the same thing as comprehension, and this is where the conversation gets interesting.
A student can read a chapter on photosynthesis three times and still fail to explain why a plant in a dark room eventually dies. The text did its job. The transmission happened. The understanding did not. This gap between exposure and internalization has always been the central problem of teaching — and most traditional solutions address the wrong end of the pipeline.
We keep improving the content. We add diagrams, summaries, glossaries, and practice questions. What we rarely change is the nature of the interaction itself. Learning, it turns out, is not a reception problem. It is a conversation problem.
What Happens When You Can Ask Anything
The most revealing thing about AI tutoring tools is not their accuracy. It is how students use them when no one is watching. Research from educational technology labs shows that learners ask AI systems questions they would never raise in a classroom — questions they consider too basic, too obvious, or too embarrassing to voice in front of peers.
This matters enormously. The questions students suppress are often the exact questions they need answered most. A first-year college student who cannot quite grasp what a variable does in algebra will not raise her hand and say so. She will nod, copy the example, and fall further behind. With an AI interface, she asks. She asks again. She reframes the question five different ways until something clicks — and the system responds to each attempt without frustration, without judgment, without the social cost that makes classrooms feel unsafe for intellectual risk-taking.
This phenomenon connects directly to what researchers call “psychological safety” in learning environments. Amy Edmondson’s foundational work on the topic demonstrates that people learn more — and more deeply — in spaces where failure carries no social penalty. AI tutors, for all their limitations, create a strange kind of psychological safety by being incapable of judgment. The machine does not remember that you asked the same question yesterday. It does not sigh.
The Bonding Problem Nobody Talks About
Here is where things get philosophically thornier. Students do not just use AI tools instrumentally. Many of them form something that resembles attachment. They develop preferences — they prefer one phrasing style, one level of formality, one kind of response rhythm. Some express frustration when an AI gives a different kind of answer than expected, as if it broke a kind of contract.
This is not a bug in the system. It is a reflection of how human cognition actually works. Research into why people form emotional connections with AI systems points to something fundamental: human brains are wired to attribute intention and personality to any system that responds to them in a contingent, contextual way. When a tool responds to you specifically — adapting to your confusion, adjusting to your pace — your brain begins to treat it socially, even if you consciously know better.
For educators, this is not a reason to panic. It is a reason to pay attention. If students bring social expectations to AI interactions, then the design of those interactions carries moral weight. An AI tutor that is cold, terse, or dismissive teaches students something beyond the content. It teaches them what to expect from intellectual authority figures. That lesson outlasts any chapter on the periodic table.
Open Pedagogy Meets Machine Interactivity
CUNY Pressbooks represents a specific vision of education: knowledge as commons, not commodity. Open textbooks, freely adaptable, built by and for the communities that use them. This model has proven powerful. But it still operates largely within the paradigm of fixed content meeting variable readers.
The next frontier for open education is not just open content — it is open interaction. What does it look like when the textbook can ask you questions back? When does it notice you skipped section three and suggest you return? When does it recognize that your three attempts at a summary exercise all share the same misconception and offer a different angle?
This is not science fiction.Platforms exploring AI-integrated learning are already experimenting with these interaction models. The challenge is not technical. The challenge is pedagogical and ethical: who designs the interaction, whose values shape the response patterns, and how do we ensure that the “intelligence” in AI tutoring reflects diverse ways of knowing rather than replicating the blind spots of whoever built the training data?
The Metacognitive Gap
One finding from cognitive science deserves more attention in educational technology discussions: the gap between knowing something and knowing that you know it. This is metacognition — the brain’s capacity to monitor its own comprehension —, and it is notoriously unreliable.
Students routinely overestimate their understanding after passive reading. They feel familiar with a concept without being able to retrieve or apply it. Traditional assessments catch this after the fact, usually through tests that arrive too late to redirect learning in the moment.
AI tutors, when designed well, can address the metacognitive gap in real time. Not by quizzing relentlessly, but by asking students to explain concepts back, to generate examples, to predict what comes next. These are not tricks. They are the cognitive moves that actually build durable knowledge — a phenomenon known in the learning sciences as the “generation effect.” What you produce, you remember. What you merely receive, you forget.
The question open educators should be asking is not whether to integrate AI into learning platforms. That decision is already making itself. The question is whether the integration will be thoughtful — grounded in what we know about how people actually learn — or whether it will simply layer a chatbot on top of a static PDF and call it innovation.
What Thinns Means for Educators Right Now
None of this requires abandoning human teaching. It requires understanding what human teachers do that machines cannot replicate — and doubling down on exactly that. Teachers read a room. They notice the student in the back who stopped taking notes twenty minutes ago. They build the kind of trust that makes a struggling student stay after class. They model intellectual humility — the act of not knowing something and figuring it out in front of others.
AI handles scale. It handles repetition without fatigue. It handles the embarrassing question at midnight. Human teachers handle everything that requires a genuine relationship: the student who does not trust education yet, the one who needs someone to believe in them before they can believe in themselves.
Open educational resources, AI tutoring tools, and skilled human educators are not competing solutions. They address different dimensions of the same problem: how do we help a person move from not knowing to knowing, in a way that sticks, that builds confidence, and that opens the door to wanting to know more?
That question has no final answer. Which is exactly why it keeps being worth asking.