For Whom the Bell Curves - by Hollis Robbins
Anecdotal Value
SubscribeSign in
For Whom the Bell Curves<br>AI may explode how students are evaluated
Hollis Robbins<br>Jun 26, 2026
Share
One of the biggest higher ed stories of the year is the push to reinstate the SAT in elite college admissions. Everyone is doing it. Fourteen hundred faculty at the University of California, test blind since 2020, signed a petition to require a math score from applicants to STEM majors. I’m not against; I think we should bring back the antonym section.<br>Another big news story is “grade inflation.” Harvard faculty voted to cap A grades. Now the Manhattan Institute is proposing that states regulate grading at public universities. I’m against the idea on principle, as someone who wants to keep administrators out of the classroom, but I’m also against it because it ignores the much bigger problem in public higher education, which is lack of quality. (See Todd Wallack’s Washington Post story this week about cheap online credits.)<br>What do the advocates for more testing and the advocates for fewer top grades have in common? Both camps are annoyed that a college degree no longer signals anything. If anyone can get into college and everyone can get an A, what does anything mean? The SAT camp wants a minimum number going in. The grade camp wants a disciplined number coming out. Neither is focused enough on what happens in between and neither seems to understand that everything will change in the AI era, including the role of the bell curve/normal distribution as the go-to tool for comprehension. When you can ask a pro model what the real distribution of some quantity looks like and where a particular case sits in it, you stop needing the handy two-parameter proxy.<br>If Harvard wants to limit A grades, fine, though a Harvard professor friend told me over dinner some months ago that he gives out more As than he did 20 years ago because the students are better and work harder. I am guessing he will continue to give A grades.<br>At a place like Harvard, the quality control is the faculty. Whatever anyone thinks of their politics, they are highly qualified and don’t tend to be slackers. But in too many public (and some large for-profit) universities there is little-to-no quality control, particularly when it comes to online programs. Many of the people giving out the grades are not professors. They may not even be people.<br>If a student got a 1500 (99%) on the SAT and then went to a regional public university for a degree in an online psychology program, the GPA would be the least interesting signal to an employer. I would wonder: why is a kid with a 1500 SAT doing that in the AI era? Who is advising this kid? Who is pocketing the money?<br>I understand the desire for a legible signal from any institution, familiar or unfamiliar. I get that people see a national sorting system as a kind of public good. You have your SAT before college and your non-inflated GPA after. There’s a distribution and everyone knows their place on it. Otherwise, the fear is, things are unsorted, unfitted, and bad, as the Manhattan Institute report claims:<br>Employers have very little time to evaluate individual applications and are likely to ignore most of the information provided by median grades. On the other hand, the inflation-adjusted GPA summarizes the entirety of the information provided by median grades into one number. It is easy to understand, as it follows the same scale as the traditional GPA. If employers then respond to this policy by preferentially hiring those with higher inflation-adjusted GPAs, this provides stronger incentives for students to both demand more rigorous course grading and take courses headed by professors who grade more strictly.
The report says explicitly that the “inflation adjusted GPA” is not a curve, but it might as well be, since it depends on “data on course grade distributions that most universities already collect.” Who is served by thinking in terms of where a person falls on a distribution? Busy strangers, apparently. But the report doesn’t factor in the kind of data that AI will be able to offer that will be better than any bureaucratic proxy.<br>The day of the proxy may be over. The UC faculty don’t care about the SAT except to signal baseline preparedness. AI may offer new ways to show preparedness. AI will very likely offer new ways to deliver what is quaintly called “a college degree.” I don’t pretend to have the answers but the market for new ideas is robust.<br>Cultural resistance<br>Of course many people still want the SAT. The median SAT score of test-takers nationally is approximately 1050, and the median of test-takers from households earning above two hundred thousand dollars is higher by about 200 points. High income households can buy better schools, tutoring, test prep, books, culture, and pro-model LLMs. A motivated student from a family without means can study hard and beat the household-income prediction, as the UC...