Dr. Peter Andrew Jamieson — Academic in an information evolution

← Back to Meanderings, Opinions, and Articles

Learning Produces Judgment, and Other Things I Took From the Cornell Report

September 2026 · written with the help of Claude (Anthropic)

Cornell published the report of its Committee on the Future of the American University in September of 2026, and I read the whole thing. I did not read it neutrally. I went in with two questions that I wanted answered, and I want to state them up front because they shaped everything I noticed.

The first question got a better answer than I expected. The second question got an answer too, although not the one the committee intended to give. Note that I treat these questions as a way of deciding what I should be doing, and what my institution should be doing.

Learning produces judgment

The report makes a claim early and then leans on it for two hundred pages: judgment is the skill that learning produces. Most definitions of learning point at knowledge — you learn a thing and afterwards you know the thing — but this definition points somewhere else entirely. Learning is the slow, evidence-based discernment that resists the “hot take,” and you do not acquire it through simple transmission. We have known this for a long time, but I appreciate that the committee went deeper and tried to define the apex of learning as judgment.

I think the most useful way to read this is through Leslie Valiant’s work on educability. In The Importance of Being Educable (Princeton University Press, 2024), Valiant argues that what makes humans distinctive is not raw learning but a combination of three capabilities: (a) learning from our own experience, (b) acquiring theories from others through instruction, and (c) applying that knowledge, from both sources, in an integrated way to situations we have not seen before.

Judgment lives in (c). It is not the learning and it is not the being taught. It is the thing that decides which of your accumulated theories applies here, in this messy case, with these constraints, and whether the theory you reached for is actually the right one. Valiant himself is fairly blunt that humans are not naturally good at this, and that we soak up entire systems of thought without ever evaluating whether we should trust them.

This matters right now for an obvious reason. A machine can produce the knowledge, and increasingly a machine can produce the theory as well. What a machine cannot do for the student is perform (c) on the student’s behalf in a way that leaves the student able to do it next time. A student who uses the model to skip the practice has outsourced exactly the capability the degree is supposed to build.

Transformation, and how the world works

The report distinguishes a transactional education from a transformative one: the transactional model sells a credential, and the transformative model changes the person. This distinction is not new, but the report takes it further than most treatments do, and I will come back to that.

I would point my Miami colleagues at Marcia Baxter Magolda here. She is Distinguished Professor Emerita at Miami, and her longitudinal study of college students produced the theory of self-authorship, which she lays out in Making Their Own Way: Narratives for Transforming Higher Education to Promote Self-Development (Stylus, 2001; now Routledge). She frames development around three questions: “How do I know?”, “Who am I?”, and “What kind of relationships do I want with others?” In some ways, her work suggests that the value of the four years is in transforming ourselves into who we want to be in the world — and where’s the money in that?

That first question is the one I keep returning to, because it is the epistemological question, and it is not far from what I would ask as an engineer. How does the world actually work, and how would I know if I were wrong?

Baxter Magolda describes students moving through four phases: following external formulas, arriving at a crossroads, and then, if the education works, becoming the author of their own life and finally reaching internal foundations. As far as I can tell the Cornell report never cites her, which is a shame, because the report spends a chapter arguing for self-authorship without using the word.

I should also say something uncomfortable about my own institution. Miami is driving toward R1 status, and I would argue that this is Miami going all in on the transactional approach as a bet on survival (our liberal education requirements are shifting toward skills — “what do I get for what I came here for?”). That is not a criticism of anyone’s motives, because the bet may well be correct, and the demographic and financial pressures are real. Still, it is worth naming the bet for what it is, since the report I am reading argues at length that the transactional approach is ultimately self-defeating.

The transactional model eats time

Here is the part of the report I did not expect, and it is the single best argument in it.

If you optimize for return on investment, you minimize the investment, and students told the committee that they want to finish in three and a half years, or even three. That is a rational response to a real price. Tuition is the cost, so less tuition is a better deal. We can see the public and political powers embracing this transactional approach — is it wrong?

But time on campus is the mechanism. Time is where the productive failure happens, and time is where a student takes the course that does not fit the résumé and discovers something. If you cut the time, you cut the transformation, and the transformation happens to be the thing employers keep saying they actually want. The report puts it more cleanly than I can, noting that the transactional approach “optimizes for signals of competence at the direct expense of competence itself.” I will add that industry mostly draws from our transformative pipe without helping to keep it running — the short-sighted CEO can’t see past this year’s stock price to ask whether the pipe will still be running in two, four, or ten years.

So the transactional model does not merely undervalue transformation. It actively consumes it. In a transactional model curiosity becomes a liability, because a course outside the major does not lead to a job in the major, and a hard course risks a lower grade. That is a strong claim, and I believe it is correct.

The paradox of high grades and AI skills

The paradox is that our grades and the capability they are supposed to measure have come apart, and AI keeps pulling them further apart. Even before AI, using grades as a public signal was the start of the problem — their original purpose was feedback to the learner. A grade has always been a proxy. We assign work, the student does the work, and the grade reports how well they did it, which we then treat as evidence of what the student can now do on their own. AI breaks the second step without touching the first. The work still arrives, and it arrives in better shape than before, so the grade goes up while the capability underneath it goes down. The better our students’ grades look, the less those grades tell us, and the less they tell the students about themselves.

And then the paradox closes on itself. Grades inflate, partly from tuition-driven customer service and partly because student evaluations sit inside faculty promotion, so rigorous grading carries a personal cost. Inflated grades stop discriminating between students. The competition does not go away when we stop measuring it, though, so it relocates to student clubs, which become ruthlessly selective. The signal migrates to a place that has nothing to do with learning at all, and we are left grading a thing we can no longer read, or even care about.

The students in the report are not cynics, and I want to be fair to them here, because almost all of them told the committee that they do not want to use AI to do their work. I would put it this way: we all play the games we are in as best we can. If wood converts to coal converts to money converts to points, and I don’t need to know how the conversion happens — I just need the points to “win” — then yes, I will work that transactional system to get the most and best points. I will also counter that AI should be used; the question (which I haven’t seen answered) is where and when.

Students should play the game, and if it is transactional, so be it. That logic is not a character flaw; everyone else in the institution similarly acts according to the game they are presented with, where winning is existential. Transactionally speaking, not using AI costs more time, more time means a lower grade somewhere else in the schedule, and a lower grade means worse odds in a competition for a shrinking number of good jobs. The honest student is the one who pays. The system quietly rewards the student who defects. And the educational system keeps band-aiding the problem when we ourselves barely understand the technology.

I read this and thought about my own courses, because I am in the same arms race for student attention. If I do not award points for showing up and doing the reading, students rationally spend their attention on the course that does.

AI literacy is a band-aid

The report describes the AI literacy module Cornell built into its existing courses, which gives students resources and structure for thinking about what these tools mean for their own learning. I looked at it, and I think it is a band-aid — not to suggest Cornell is the only one reaching for band-aids. Every university treats this as the latest technological change that must be patched so that higher ed is holding up its side of the transaction.

I do not mean that it is bad, because it is not. I mean that it is additive. A module bolted onto a course does not change what the course is for, and if the underlying problem is that our assessments can now be completed by a machine, then a module about machines does not repair the assessment. The student sits down in front of the same assignment with the same incentive.

AI literacy treats AI as a topic to be covered. The harder problem is that AI has changed what counts as evidence that a student learned anything, and I do not believe that problem is solvable by adding content, acronyms, and yet more of the bias that comes from the illusion of explanatory depth. I see undergraduate education as a space where we, the teachers, know the answers — the literal correct answer is printed in the textbook. Before, we hoped the undergraduate would get to that answer by working both forward and backward to reverse-engineer the solution. Now, generative AI will pull back the curtain and show it to them. If I read it, I’m pretty sure I learned it. It feels good, and good feelings suggest I have done good.

The recommendation that actually matters

Of all the undergraduate recommendations in the report, I think one is the important one: provide a hands-on, project-based education at a human scale.

The reasoning is the same as in the first section of this post, and it is consistent all the way through. Sound judgment develops through practice rather than transmission, so you have to put the student in a position to practice. This is Valiant’s (c), taught deliberately and repeatedly, with a human present who can tell the student when their judgment was wrong and why. For those of us in the “professional” or “doing” degrees this isn’t hard to understand, and yet we still devote only a shallow portion of four years to it. Why? It’s hard to count fuzzy-looking nuts and then convert them to letter grades. Plus, none of us are rewarded for such hard counting. Why? It’s hard to count the fuzzy nuts of counting fuzzy nuts.

This is also expensive. In my own analysis, Electrical and Computer Engineering costs more per credit hour than any other discipline at Miami, and very nearly the whole of that difference is laboratory and build content. That cost is almost always described inside the institution as a problem to be managed.

I have come around to thinking that the cost is the product. It is the one thing we deliver that a model cannot deliver for free, and it is, at the same time, the thing we are constantly being asked to make cheaper.

Designed spaces and artifacts

A smaller recommendation, and one I liked a great deal, is that a transformative education needs designed spaces and designed artifacts.

The report means classrooms, studios, maker spaces, and residential space arranged to scaffold social interaction, and it goes as far as calling for a commitment to tech-free educational spaces, which is a braver line than it looks on first reading. It also argues that universities should invest in building their own technologies rather than adopting whatever happens to be commercially available.

I build things, so I take this one personally, and positively. In general, I believe the tools we hand our students are curriculum whether or not we choose to admit it. My greatest complaint about my current institution is the lack of interesting discussion and of a community of thinkers and doers. It’s not that we don’t do; we just do it in silos, mostly as individuals. My hope is that the undergraduates will fill this gap, but as I walk around campus I see only small pockets of engagement. And I’m looking for engaged groups doing anything at all — pickleball, music, academics, building — and we have only little pockets, most of them hidden from my view.

The incentives do not line up

The report is direct about faculty incentives, and I appreciated the honesty of it.

Research is rewarded in promotion and in the disciplinary prestige circuit, while teaching is rewarded much less, and after tenure there is very little structural pressure on a faculty member to be excellent in the classroom. The report notes that at Cornell, in the fall of 2025, non-tenure-track faculty taught 49 percent of undergraduate courses.

So the committee asks for a hands-on, human-scale, judgment-forming education, and the reward structure of the institution points in a different direction. The report does not resolve this tension, and I am not convinced anyone has (or will — I can’t count fuzzy nuts 😉).

Miami is not in the same admissions space

I should say this plainly, because it changes which recommendations travel and which do not.

Cornell is highly selective and Miami is not selective in the same way; our College of Engineering and Computing admitted roughly three quarters of its applicants in the most recent cycle. Cornell can write recommendations aimed at the pathology of hyper-competition. Our problem is a different one: a shrinking pool of college-age students in the Midwest, and a competition between departments for a share of that pool that is not growing.

The transactional erosion still reaches us, but it comes through a different door. Most of the report’s diagnosis transfers to Miami. A good portion of the prescription does not.

The research chapter has no economics, which answers my second question

This is where my second question got its answer, although not in the way I wanted.

The research chapter is careful and well argued about federal funding, about instability, and about the value of the research university to the country. What it does not do anywhere is connect research to the economics of the institution that performs it. There is an implied idea running underneath the whole chapter — that there seems to be less money now and things are going to be harder — and that idea is never made explicit enough to argue with.

That is not an economic argument. If research pays for part of the institution, I would like to know by how much, and if it costs the institution money, I would like to know that instead, because the two cases lead to completely different strategies.

To be fair to the committee, they tell us why this is hard. Elsewhere in the report they admit that a question as simple as “does the university make or lose money if it enrolls more students?” is, in their words, “currently unanswerable.” I believe them, and I think that admission is the most important sentence in the document. It also explains why the research chapter reads the way it does, because you cannot write the economics of research if you cannot write the economics of anything.

So my second question has an answer. The link between the economics of the institution and its mission is implied behind the scenes, and it is implied because the institution genuinely does not know what the link is.

The master’s degree problem

The report raises the cost of professional and research master’s degrees and asks whether those programs deliver adequate return, noting correctly that aggregate earnings figures conceal enormous variation between fields and institutions.

I would go considerably further. The master’s degree has become a revenue instrument; it is the easiest money a university can raise, and the institutional pressure to raise it is enormous. The report cites a Chronicle piece on the subject — Kevin Carey’s “The Great Master’s Degree Swindle” — and then moves along politely.

I do not think it deserves politeness. If the transactional model is self-defeating for undergraduates, and the report spends a chapter arguing that it is, then a master’s degree sold primarily as a credential is the same failure repeated at a higher price and with less time in which to do any transforming.

Conclusion

If anything, I commend Cornell’s effort in this document. As a failed R1 professor (and, I think, a good teacher-scholar), I find it fascinating to read the “big” thinkers deliberate and solve. Some of the ideas in this report are excellent, and the biggest framing I have pulled into my own mental models is that transactional higher ed is the risk — and probably the reason for the stress I see in the students I teach. Unfortunately, the recommendations leave me in a space with little I can put to practical use.

References

This post was written by me, with the help of Claude (Anthropic) for editing and for tracking down the citations.