Teach the Tool
Amplifying Human Potential with AI in Higher Education
By Dr. Matt Mooney, April 2026
The Question We Keep Getting Wrong
When artificial intelligence began reshaping how we work, learn, and communicate, higher education institutions faced a choice: engage with the technology thoughtfully or pretend it would go away. Too many chose the latter.
The right question was never whether students would use AI. They would, and they do. The right question is whether we, as educators and institutional leaders, will be deliberate about how we integrate it, model its use, and prepare students to think critically alongside it rather than in spite of it.
My thinking on this began not with a policy debate, but with a book. Joseph Aoun's Robot-Proof reframed the conversation for me: AI excels at what machines do best, searching, organizing, summarizing, and comparing vast amounts of data with speed and accuracy no human can match. What it cannot do is make human-inspired judgment. It cannot connect content to a lived life. It cannot weigh competing values. It cannot care about the outcome. That distinction is not a limitation to work around. It is the foundation of everything that follows.
A Three-Layer Framework
What I am proposing is not a single policy or a technology adoption roadmap. It is a framework built in three layers, each dependent on the one before it, and each requiring us to keep the human factor at the center.
Layer One: Embedded, Modeled AI Use in Instruction
Faculty cannot teach students to use AI well if they are not using it well themselves. The first layer of this framework asks institutions to stop treating AI as a threat to academic integrity and start treating it as a teaching opportunity.
This means incorporating AI transparently into how course content is developed. Faculty often feel pressure to know everything, and the pace of change in most fields makes that expectation unrealistic. AI can help faculty surface current examples, identify how course topics are affecting modern society in real time, and close the generational gap between instructors and their students. That is not a compromise of faculty expertise. It is an extension of it.
More importantly, this layer asks us to redesign assessments. Rather than prohibiting AI use, we should be designing prompts that use AI as a starting point and then push students directly into the work that only humans can do: forming opinions, making connections to their own lives and careers, identifying what the AI got wrong or missed, and defending a position. This approach does not make it easier to shortcut the learning. It makes the shortcut less useful than doing the work.
The calculator analogy is instructive here. For decades, mathematics education treated calculators as a threat to fundamental skill development. Eventually, the field recognized that the goal was never to calculate, it was to understand when and why to calculate, and what to do with the result. We are at that same inflection point with AI, and the institutions that recognize it earliest will have a meaningful advantage in preparing their graduates.
Layer Two: Student Ownership Through AI-Powered Learning
The second layer shifts focus from the instructor to the learner. It asks institutions to expect students to use AI tools as part of their own learning practice, the same way we expect them to use library databases, citation managers, or statistical software.
Tools like NotebookLM represent a genuine shift in what is possible for self-directed learners. A student can upload course materials, generate interactive study guides, create practice assessments, and engage with content in formats that match how their brain actually works. Many institutions already have Google Workspace enterprise licenses that make these tools available at no additional cost to students.
The accessibility dimension of this layer deserves specific attention. According to Pew Research Center's 2025 data, nearly all college-aged adults own a smartphone, with ownership among 18 to 29 year olds at 97 percent, and institution-level surveys consistently push that figure even higher. While a smartphone is not the ideal interface for every learning task, the fact that most of these tools run on mobile devices matters enormously for students who do not have reliable access to desktop hardware. Accessibility must continue to improve across the board, particularly for students with disabilities, and voice interfaces, adaptive text tools, and flexible display options should be central requirements, not afterthoughts, in any institutional AI strategy.
The deeper shift this layer requires is cultural. Students need to be expected to take personal ownership of how they learn, not just what they learn. That is a significant departure from a lecture-based, sage-on-the-stage model that has dominated higher education for generations. It will not happen without institutional will, faculty buy-in, and explicit support structures for students who have never been asked to work this way before.
Layer Three: Adaptive Personalized Learning
The third layer is the most ambitious, and the most consequential. It is also where the risks are greatest, which is why the ethical guardrails that govern it must be as carefully designed as the technology itself.
Think of it this way: imagine a learning experience designed like a guided adventure, one where a student is presented with content and assessed for understanding, and if they demonstrate mastery, they move forward. If they do not, they are not shown the same material again. They are presented with the same concept through a different lens, with different examples, grounded in something they already know and care about.
This idea is not new. A professor I had in a doctoral-level statistics course handed out note cards on the first day of class. He asked for our names, our areas of study, and our research interests. From that point forward, every example he used to teach statistical concepts was drawn from those responses. Social scientists got examples from political data and human development research. It was an enormous amount of work for one person to pull off, and it made all the difference. The content stopped feeling like an obstacle and started feeling relevant. That professor proved the concept by hand. What AI makes possible is doing it at scale, for every student, in every course.
Consider a single student enrolled in both chemistry and geography in the same semester. In chemistry, that student shares that they love to cook and are learning to prepare food from Morocco. Suddenly, the course has an anchor: chemical reactions explained through what happens when ingredients are combined, when food is heated, when a sauce emulsifies or breaks. In geography, that same student's Moroccan cooking interest opens a door to geopolitical history, cultural identity, trade routes, and regional conflict in North Africa. Same student, two completely different profiles, two courses that now connect to something real in that person's life. This is not customization for its own sake. It is what good instructional design has always tried to do, and it is what AI finally makes possible at scale.
The key mechanism is the learner profile, built fresh for each course at the start of the semester. What does this student already know? What are their areas of interest? What context will make the abstract concrete for them? That profile guides how AI delivers and adapts content throughout the course. And when the course ends, the profile is retired. It is not aggregated into a persistent data asset that follows the student indefinitely.
This is not the most efficient approach from a data architecture standpoint. A single persistent profile would be easier to build and cheaper to maintain. But the per-course profile is actually the better pedagogical model. A student in chemistry and geography in the same semester is, in a meaningful sense, two different learners in two different contexts. The profile that serves them in one course may not serve them in the other. Ephemeral profiles are not just more private. They are more accurate.
The Guardrails Are Not Optional
Any serious discussion of AI in higher education has to reckon with the ways it can cause harm, not just the ways it can help. I want to name three concerns directly.
Bias in AI systems is not a minor technical issue. It is a structural one. The models that power adaptive learning tools reflect the data they were trained on, and that data carries the historical inequities of who has been centered in research, whose examples have been used, and whose voices have been amplified. Addressing this requires more than better algorithms. It requires diversity in who funds AI research, who conducts it, and who holds positions of authority in the institutions that deploy it. There is no technological fix for a problem that is fundamentally about representation and power.
The risk of cookie-cutter instruction masquerading as personalization is real. An AI system that shuffles the same limited content into different formats is not personalization. It is the illusion of personalization. Genuine adaptive learning requires rich, human-authored content, diverse examples, and an underlying model sophisticated enough to recognize when a student needs something genuinely different, not just differently packaged.
And the question of who is doing the work matters. There are tools available today that will log into a learning management system as a student, review course materials, attend virtual lectures, and complete assessments with no student involvement whatsoever. This is not a gray area. A tool that removes the student from the learning experience is not an educational aid. It is a product designed to help students avoid education entirely, and institutions have both the right and the responsibility to draw that line clearly.
Instructional designers are not replaceable by AI, despite what some voices in the field have suggested. The work of designing effective instruction requires human judgment about how real learners process information, where cognitive load becomes a barrier, and how to create the conditions for genuine understanding. AI can assist with structure, accessibility checks, and content organization. It cannot replace the trained professional who knows when a beautifully organized course is still failing its students.
What We Are Building Toward
The goal is not AI-powered education. The goal is better education, and AI is one of the most powerful tools we have ever had to help us get there, if we use it well.
That means using AI for what AI does best: searching, organizing, adapting, and scaling. It means protecting the human work that no algorithm can replicate: judgment, connection, mentorship, and meaning-making. And it means being honest with our students about both, because the graduates who will thrive in the next decade are not the ones who learned to hide from AI. They are the ones who learned to think alongside it.
Teach the tool. Then teach the student what the tool cannot do.
About the Author: Dr. Matt Mooney is a retired Associate Dean and senior higher education leader with experience spanning faculty, staff, and administrative roles at R1 institutions. His work focused on online learning, instructional design, educational technology, and the intersection of innovation and student success.
A note on how this was written: I used Claude, an AI assistant from Anthropic, as a thought partner in developing this paper. That is the same kind of transparent, modeled AI use the paper argues for.