What the Diploma Was Proxy For

On Thursday, Matt Barnum at Chalkbeat published a column with a question most school boards have not yet brought themselves to ask out loud.¹ Will more education still pay off? The piece sits with a finding that has been circulating quietly in labor-economics seminars for a year. In randomized experiments, when participants are asked to solve a workplace-style problem, the most-educated workers do markedly better than the least. Hand both groups a chatbot, and the gap narrows. In a paper released this spring by the National Bureau of Economic Research, AI assistance lifted scores for everyone, but it lifted them most for the people with the least education.²

The result is the kind that makes deans of admission lean back in their chairs. For a century, the case for more schooling rested on a single empirical pattern. The college-educated earned more than the non-college-educated, and the gap was wide enough to justify the years of tuition. Harvard's David Deming, who has spent a career documenting the pattern, currently puts the premium at roughly seventy percent.¹ What AI is starting to do, in the labs at least, is bring the floor up. It does it most visibly for the people who were standing closest to the floor to begin with.

The Premium Was Always a Proxy

The labor-economics literature has never quite said the college premium was about a specific set of facts a student had memorized. It has said the premium was a signal. A signal that the worker could reason. Could write a paragraph that argued something. Could read a memo and notice what was missing. Could hold a problem in her head for an hour without flinching. The diploma was, in shorthand, a proxy for the practice of thinking.

For a long time the proxy was reliable enough. The act of getting through college was, more or less, the act of practicing those skills under supervision. The credential stood in for the practice. An employer trusted the credential because there was no good way to inspect the practice itself. The diploma was the trust receipt for four years of thinking the employer never saw.

What AI has quietly changed is the inspection problem. The final piece of writing on the page now arrives looking competent whether or not the writer practiced. The memo is well-edited whether or not the writer drafted it. The proxy worked when the output was hard to fake. The output has become easy.

What the Deans Already Suspect

Earlier this spring, in an Ec 50 lecture, Dean Deming told his students that AI literacy was now an essential new skill and warned them, in the next breath, about the cost of overreliance on the tools.³ In a separate talk a week and a half earlier, he had acknowledged that AI use among Harvard students is far more pervasive than the College's honor system has been able to catch, and said he hoped to "push harder" on the question in the years ahead.⁴ The two observations point in the same direction. The institution that has built its reputation on producing thinkers is no longer sure it can tell, from the work, whether the thinking happened. The premium it sells, at three hundred thousand dollars across four years, rests on a signal that has become noisier.

This is not, mostly, an argument for less education. It is an argument for a different one. If the diploma was a proxy for the practice of thinking, and the proxy is becoming unreliable, the response is not to abandon the practice. The response is to start measuring the practice itself.

The Practice, Made Visible

A future of school that takes the Chalkbeat question seriously will have to surface the thinking the diploma used to summarize. Not the output, which AI can match. The practice that produced it. The first draft a student typed before she asked any model anything. The second sentence she rewrote three times. The paragraph she abandoned after she read it back to herself. The question she asked Aidan that turned out to be the wrong question, and the better one she asked twenty minutes later. The breakthrough at minute thirty-eight, after a long pause at minute thirty-one.

None of that is currently on the transcript. The transcript is a list of finished objects. A future transcript would be a record of the work that produced them. It would not replace the diploma. It would let the diploma point to something verifiable again. The premium would not have to rest on what the worker can produce when the model is in the room. It would rest on what she did to learn how.

The Floor and the Ceiling

If AI is bringing the floor up, the schools that still matter will be the ones that raise the ceiling. The ceiling is not, mostly, more content. It is more practice. More revision. More returning to a hard problem the day after you thought you were finished with it. The skills the economics literature has always pointed to are the ones that get harder, not easier, when the model is sitting next to you. The schools that find a way to capture and credit those skills will be the schools whose graduates still mean something at the next decade's hiring interview.

That is the future Koan has been quietly preparing the ground for. Not by replacing teachers with models or models with teachers. By making the practice of thinking, the unglamorous middle of it, visible. The diploma was a summary statistic. The next document a young worker walks into an interview with may need to be the trail.

If the value of a degree was always a proxy for the value of the thinking behind it, what happens when the thinking finally becomes something we can see?