AI makes answers cheap. Questions matter more.

Dr. Paul Kim
Former CTO, Stanford Graduate School of Education
Former Chair, World Bank Committee on Educational Technology

In a world of automated answers, universities must teach students to ask better questions, says Dr. Paul Kim.

Gavin Allen: How will digital technology and AI reshape the fundamental mission of universities?

Dr. Paul Kim: I divide educational history into two eras: B.G. — Before GenAI — and A.G. — After GenAI. For nearly a thousand years, the B.G. university rested on a stable bargain. It held scarce knowledge, transmitted that knowledge through teaching, and certified those who had absorbed it. All three of its classic functions — knowledge creation, skills development, and credentialing — assumed that information was hard to obtain, and that expertise had to be poured from one mind into another.

The A.G. university cannot run on that bargain. Skills development is the first thing to erode: a motivated learner with access to an AI model can now acquire many competencies faster and more cheaply outside the institution than within it. Lifelong learning moves from peripheral service to core proposition, because a four-year intellectual inoculation no longer lasts a lifetime.

Ready for prime time?

The deepest shift, though, is in what we are preparing students for. The B.G. university prepared students for jobs that already existed. The A.G. university must prepare them to create their own jobs and invent professions that do not yet exist. That single reframing changes everything downstream — what we teach, how we teach, and how we measure success.

Gavin Allen: What are the biggest misunderstandings university leaders have about AI, and how should they be addressed?

Dr. Paul Kim: The most common misunderstanding is treating AI as a policy problem rather than a pedagogical one. Many institutions are still drafting detection regimes and integrity rules as if this were a containment exercise. It is not. Students already live in an A.G. world, and pretending otherwise simply pushes behavior underground while leaving faculty unprepared.

A second misunderstanding is that AI is a single technology with a predictable trajectory. Foundation models, agents, retrieval systems, and domain applications evolve on different timelines and demand different responses. Leaders who buy one expensive platform and declare victory will be obsolete within a year.

But the most consequential misunderstanding is the belief that AI can be layered onto the existing teaching model without disturbing it. The teaching model itself — the professor as the delivery mechanism for information — is what the A.G. era renders obsolete. The role of the faculty member has to change from teacher to coach and mentor: from someone who transmits content to someone who develops judgment, asks provocative questions, and guides students through problems that have no answer key. Leaders who grasp this will invest in faculty development at scale and in structured experimentation with real budgets behind it. Those who do not will keep optimizing a model that no longer fits the world.

Gavin Allen: How is AI transforming the behind-the-scenes university infrastructure — admissions, operations, student support?

Dr. Paul Kim: The administrative transformation is moving faster than the pedagogical one, because the use cases are clearer and the political stakes lower. In admissions, AI is being used to triage applications, detect anomalies, and surface candidates whose conventional metrics undersell them. In scheduling, facilities, IT support, and procurement are increasingly mediated by AI agents. In student support, we are seeing 24/7 advising assistants, early-warning systems that flag students at risk of attrition, and personalized academic coaching at scales no human counselor corps could match.

The largest gains will come in research administration — grant writing, compliance, data management — and in the long tail of bureaucratic friction that has always made universities slow and expensive. But automation raises hard questions about data stewardship, algorithmic fairness, and the proper role of human judgment in high-stakes decisions like admissions and discipline. A university that automates without re-examining these questions will simply scale its existing biases at machine speed.

Build or buy?

Gavin Allen: What role should universities play in building AI systems, versus buying them?

Dr. Paul Kim: Very few universities have the resources or talent density to build foundation models, and those that try will mostly waste money. The sensible rule is: buy at the foundation layer, build at the application layer, where institutional domain knowledge, data, and pedagogical commitments create genuine advantage.

What makes this practical is that the application layer is no longer the exclusive preserve of professional engineers. Educators, researchers, and administrators can “vibe code” software simply by describing what they want. That shrinks the cost of building exactly the tools an institution needs.

Partnerships with technology companies are inevitable and, on balance, productive, but they must contain three elements. First, clear data governance, including who owns student data and whether it can train future models. Second, institutional control over the user experience, so the institution’s pedagogy is not displaced by a vendor’s defaults. Third, genuine exit options, because today’s strategic partner may be tomorrow’s monopolist. In our SMILE platform and in the AI-SANA program in Kazakhstan, we have deliberately built application-layer infrastructure on top of multiple foundation model providers, precisely so that no single vendor relationship becomes a point of capture.

The Six C’s

Gavin Allen: How will AI force the university teaching process to evolve?

Dr. Paul Kim: The traditional classroom was built around information delivery: the lecture, the textbook, the test. With AI, that core is largely indefensible — a student can get a clearer, more personalized explanation from a well-prompted AI than from most lectures. Teaching has to move from delivering content to developing thinking, which in practice means moving from teaching to coaching and mentoring.

The shift is also a shift in what we are cultivating. For years educators have spoken of the Four C’s — critical thinking, creativity, communication, and collaboration — as the competencies that matter beyond rote knowledge. In the A.G. era I would argue for Six C’s, adding compassion and commitment. The reason is straightforward: as AI absorbs more of the analytical and productive work, the distinctively human contributions become decisive. Compassion — the capacity to understand what other people need, and to direct one’s work toward human ends — is something AI can simulate but not feel. Commitment — the persistence to stay with a hard problem, to see an invention through, to care about the outcome — is what separates a clever output from a finished contribution to society. The first four C’s make a capable graduate; the other two, Compassion and Commitment, make one who is worth following.

The single most important capacity, though, is the ability to ask foundational questions: questions that do not yet exist. AI is extraordinary at answering questions that have been asked before. It is far weaker at recognizing which question should be asked next. Yet that is precisely where learning, innovation, and invention begin: someone poses a question no one has posed, and a new field, a new product, or a new way of organizing society follows from it. A university’s central job in the A.G. era is to develop students who generate those questions rather than merely answering existing ones.

What rises in importance, then, is everything live, embodied, and human: Socratic dialogue, structured debate, oral examination, hands-on labs, project-based collaboration where students defend their reasoning in real time. What becomes obsolete is the lecture as primary mode, the take-home essay as primary assessment, and the lecture-textbook-test cycle that has dominated higher education for generations.

A concrete illustration comes from our work in Kazakhstan. When students build software through vibe coding, they no longer spend their effort fighting grammar and syntax. The barrier of technical jargon — the thing that historically gated entire fields — falls away. What is left is what actually matters: creativity, and the ability to ask good foundational questions. That is exactly the territory where a human coach adds value and where a lecture never could.

Gavin Allen: How do universities measure genuine talent, rather than AI-assisted output? And does it matter if students use AI?

Dr. Paul Kim: The calculator analogy is useful but incomplete. A calculator performs a defined operation on defined inputs. AI generates open-ended output that can substitute for the very thinking we are trying to develop. So yes, it matters whether and how students use AI — but not in the way most institutions are framing it.

The right question is not “did the student use AI?” but “did the student develop the capacities we promised to develop?” In the A.G. era, the finished artifact is cheap; the thinking behind it is what counts. So, we should assess process, not just product: oral examinations, in-class problem-solving under observation, iterative drafts with documented reasoning, live demonstrations where students defend their choices.

What we are really measuring lines up with the Six C’s. Can students think critically about an AI’s output rather than simply accepting it? Can they frame a genuinely creative, foundational question — one that has not been asked before — and pursue it? Can they communicate and collaborate well enough to turn an idea into something real? And do they bring compassion and commitment — a sense of whom the work is for, and the persistence to finish it? An AI can produce a polished essay. It cannot explain supply why the question mattered in the first place. Those are the things worth evaluating. We should expect students to use AI, just as they will throughout their careers, while being clear about which moments of learning require unaided struggle, because that is where the capacities above are actually formed.

Using mobile technology to help children’s learning

I founded Seeds of Empowerment in 2008 at Stanford as a UNESCO-supported nonprofit dedicated to bringing inquiry-based learning to children in under-resourced communities worldwide. Our flagship platform, SMILE — the Stanford Mobile Inquiry-based Learning Environment — rests on a simple conviction: children learn most deeply not when they answer questions but when they pose them.

SMILE is designed to run on low-cost mobile devices in places with limited connectivity, and it has reached millions of learners across dozens of countries — from rural South Korea and the slums of Buenos Aires to refugee camps and indigenous communities in the Amazon. What two decades of fieldwork has taught us is that the bottleneck to global learning is rarely content or even infrastructure. It is whether children are granted the dignity of being treated as thinkers. That philosophy now looks like preparation for the A.G. era, where the ability to ask good questions is the central skill — we were cultivating it in children long before GenAI made it urgent for everyone.

Gavin Allen: Will AI widen or narrow education inequality?

Dr. Paul Kim: Both, simultaneously — and which dominates is a policy choice, not a technological inevitability. AI has the latent capacity to deliver what the educational psychologist Benjamin Bloom called the two-sigma effect: the dramatic gains of personalized tutoring that the world has never been able to afford at scale. Deployed thoughtfully, it could narrow gaps that have persisted for generations.

But the default trajectory runs the other way. Well-resourced students already have AI tutors, guidance on using them well, and family contexts that reinforce productive use. Under-resourced students often get policing instead of pedagogy. Without deliberate intervention, AI will accelerate stratification rather than dissolve it.

Our AI-SANA program in Kazakhstan is, in part, a test case for the other path. We are training 600,000 college students in AI and entrepreneurship, 60% of them women. Traditionally, technical fields were gated by jargon and syntax, disproportionately excluding women and underserved learners. With AI and vibe coding, that barrier is gone. What counts now is creativity and the quality of one’s foundational questions, and those are distributed far more equally across genders and backgrounds than coding fluency ever was. Universities sit at the intersection of research, policy, and practice, and their primary role is to develop and validate equitable deployment models, train the educators who reach the next generation, and use their convening power to insist that access is addressed before technologies scale, not after.

Gavin Allen: You launched the 2012 MOOC “Designing a New Learning Environment.” Is it time to redesign another one?

Dr. Paul Kim: The 2012 course drew tens of thousands of learners worldwide and tested the premise that the design of learning environments could itself be taught at scale. Looking back, I am struck both by how prescient some choices were and by how completely the underlying assumptions need updating — it was, after all, a thoroughly B.G. course.

Yes, it is time for a new one, but not a MOOC in the original sense of recorded lectures and discussion forums. It should be an A.G.-native learning environment in which the AI is itself a co-participant — tutor, critic, research assistant, and subject of study all at once. The course should be about designing learning when the learner, the teacher, and the curriculum are all being reshaped by intelligent systems, and when the goal is no longer to prepare people for existing jobs but to help them invent new ones. I have been sketching exactly this and expect to launch a successor before long.

Designing the university of the future

Gavin Allen: If you were designing a university from scratch today, what would it look like?

Dr. Paul Kim: I would build neither a single physical campus nor a purely online institution, but a network: a small number of physical hubs distributed globally, anchored to local communities and partner institutions, connected by a shared AI-augmented learning platform.

Organizationally, I would dissolve the traditional department structure in favor of problem-centered studios where students rotate through real-world challenges in health, climate, governance, and education, guided by faculty who function as coaches and mentors rather than lecturers. Credentialing would be modular and continuous rather than degree-bound, recognizing that careers now span fifty years and several reinventions. The explicit aim would be graduates who can create their own jobs and invent professions that do not yet exist.

Technologically, the institution would treat AI as infrastructure, not an add-on. Every student would have a persistent AI learning companion calibrated to their goals and held to high standards of integrity. Every faculty member would have AI support for course design, assessment, and research. Building tools would be cheap, because domain experts could create their own software through vibe coding rather than waiting on engineering teams.

The curriculum would be organized explicitly around the Six C’s — critical thinking, creativity, communication, collaboration, compassion, and commitment — because those are the capacities that remain scarce when intelligence itself becomes abundant. And every studio, every project, every assessment would push students toward the same habit: not to retrieve the best existing answer, but to ask the question no one has asked yet. Those foundational questions are the seeds of learning, of innovation, of invention, and ultimately of change in our society. A university that produces question-askers will matter in the A.G. era. One that produces answer-givers will not.

And underneath all of it, the institution would commit to two things no technology can replace: deep mentorship between younger and older minds, and serious engagement with the question of what kind of human being a graduate should become. The A.G. era does not make those things less important. It makes them the whole point.