An observation made by neuroscientist and entrepreneur Vivienne Ming on our “Learning Machines Podcast” prompted a discussion on how AI is impacting education by changing relationship between knowledge, understanding, and intellectual effort itself.
Ming said: “What I’ve come to understand when I’m working with my graduate students, and I kind of wish more of the public and even scientists understood is my job isn’t to teach them knowledge. They’re all brilliant. They can figure it all out on their own. My job is to teach them understanding. They know everything. They understand nothing. And that is exactly what I feel like with these large language models, these agentic models.”
In this context, “knowledge” refers to the accumulation and retrieval of information, procedures, and recognizable patterns. “Understanding” concerns the capacity to navigate ambiguity when information is incomplete, contradictory, or structurally insufficient. Today, we consider how to use AI to help develop better judgment and tolerance for uncertainty in higher education.
Democratizing access to knowledge
For most of human history, education rested upon a relatively stable asymmetry. Centuries before the “sage on stage” model, the tutor had knowledge, and the pupil did not. Those who knew Latin dominated those who did not. Authority was derived not merely from institutional position, but from informational scarcity and asymmetry.
We had to go to university, in part, because knowledge was difficult to access, slow to acquire, and socially mediated through teachers, books, archives, laboratories, and institutions. Those who could access it ran the world.
In this model, universities, leaders, parents, and experts derived their power from claiming: “I know more than you”; “I can access what you cannot”; “I interpret the world for you”; and “I mediate uncertainty for you.”
READ: Sreedhar Potarazu and Carin Isabel Knoop | Friction vs fiction: AI automators, validators, and cyborgs (May 22, 2026)
Gradually, access to information and education widened; during COVID, online access expanded dramatically. Teens in Uganda could take classes from MIT or Harvard for free.
Regardless of the dramatic changes, however, educational systems continue to reward success on what Ming terms solving “well-posed problems.” Such problems possess identifiable parameters, recognizable methods, and evaluable solutions. Examinations, credentialing systems, and many professional advancement systems continue to operate according to this logic.
“Show me an exit exam,” Ming remarked, “at any education system anywhere in the world that isn’t about your ability to answer well-posed questions.” Eloquence, recall, and procedural mastery were difficult to externalize technologically. Professions like doctors and lawyers, which combined recall and intelligence, captured a lot of economic value and commanded social respect.
From “sage on stage” to “stage on screen”
While most public discussion surrounding AI in education has centered on academic dishonesty and automation, it also promises accelerated and personalized learning. One of the AI’s most remarkable features is that it provides every human with a pocket tutor and infinite library.
Technology distributes information more efficiently than universities do.
This reality unsettles educational and intellectual balances of power by introducing a third presence into the educational relationship—by providing what once only educators or physical resources could. It can summarize, explain, coach, persuade, imitate expertise, all while generating the appearance of intellectual fluency and ease for both students and faculty. Increasingly, coherent synthesis, polished prose, plausible interpretation, and even simulated reflection can be produced rapidly and at minimal cost.
Educational institutions may prove especially vulnerable to this confusion because universities have long relied upon proxies for understanding: fluency, coherence, confidence, procedural competence, and rhetorical polish. Diction could dictate success. At Harvard Business School, British accents were overrepresented among those receiving honors.
Yet, synthetic systems increasingly reproduce these signals independent of the cognitive processes they once indicated.
This problem extends beyond students. Faculty themselves increasingly integrate AI into lecture preparation, administrative writing, recommendation letters, literature reviews, grading assistance, and curricular design. Educators are subject to the same pressures toward efficiency, cognitive offloading, and energy conservation that shape behavior elsewhere in society. Universities are not observing this transition from outside; they are fully implicated within it.
Of course, educational institutions have long-term accommodated technologies that externalize portions of cognitive effort, from calculators and statistical software to search engines and spellcheck. With Gen AI, however, blends the distinction between seeming to understand versus developing understanding from effort—for both teachers and learners.
AI as instrument of mass instruction
AI reflects our desire for certainty and belonging. But it can also open the door to reintroducing intellectual resistance into systems increasingly optimized for customer satisfaction, affirmation, and smoothness. In that context, how can educational institutions continue cultivating understanding in an environment increasingly optimized for its performance? How can they cultivate forms of disciplined uncertainty: the capacity to remain engaged when answers are incomplete, conflicting, or absent altogether.
Ming’s own research offers a useful distinction here. In our last column, we described her broad modes of interaction with AI systems: “automators,” “validators,” and “cyborgs.” Automators increasingly outsource cognition reflexively. Validators retain some agency but primarily use AI to reinforce conclusions they already like. Cyborgs, by contrast, use AI neither passively nor reverentially. The machine becomes what Ming calls “the loyal opposition,” expanding hypotheses, exposing weak assumptions, and intensifying rather than reducing intellectual effort.
Ming found that the quality of these hybrid human-machine systems depended less upon model sophistication than upon distinctly human characteristics: curiosity, intellectual humility, perspective-taking, and the willingness to remain cognitively engaged under conditions of uncertainty. These are also psychologically uncomfortable.
Ming described one experimental system her team developed called “AI Socrates.” Rather than directly answering questions, the system redirected inquiry back toward the participant:
“What an interesting question. To answer that, you may wish to consider…”
Participants strongly disliked the experience. Yet this system generated the highest levels of what Ming describes as “hybrid intelligence,” because it compelled users to remain cognitively active rather than collapsing into passive consumption.
More AI Socrates
If Ming is correct and the future depends less on the capabilities of machines than on the kind of users we become, educational institutions face a choice that extends well beyond the classroom. Will AI be used primarily to remove friction, uncertainty, disagreement, and effort, or to help students engage with them more productively?
The same question increasingly confronts parents, leaders, physicians, and experts.
In a world where information is abundant and explanations are cheap, what capacities remain worth cultivating? And what forms of discomfort are we willing to preserve because they are not obstacles to learning, but part of learning itself?