Seth Chandler (University of Houston) posted “Let AI Curve Your Grades” at his “AI for Legal Education” site. He writes:

Professors often face a narrow set of constraints: maintain a target mean GPA, keep distributions within strict percentage bands, and do so without assigning a lower grade to a higher-performing student. The reality is that these constraints are often challenging to satisfy given the real score distribution of a class. The result is a familiar and time-consuming process of hand-tuning cutoffs, verifying compliance, and repeating the cycle until the curve is “close enough.”

AI to the rescue once again! With the help of AI, I have developed a free app GradeCurve Pro to make that process fast, transparent, and defensible. It takes raw scores and institutional grading constraints and produces multiple compliant grading scenarios that respect the curve, and preserve (weak) monotonicity.

A couple of comments appear below the jump.

One, on computer geekery and teaching (law): In a separate post, Professor Chandler describes at great length his deep dive into Linux: the what, the why, the how. “Why I Learned Linux (and What It Taught Me About Systems, Teaching, and Control)” No law professor needs to do this (learn Linux), but I know more than a few colleagues who code, either because they learned before becoming lawyers or, in a few cases, picked up some skills during/after their professional careers began. I tried myself to learn Linux, about 15/20 years ago, building on mostly forgotten primitive programming instincts acquired during the mid-1970s. I gave up my modern quest, but I was motivated by a modest version of what Professor Chandler describes:

When you use a modern commercial operating system, you are essentially a tenant. You live in a beautiful, high-rent district designed by Apple or Microsoft. They choose the colors, they decide when the locks get changed, and they can move your furniture around whenever they please via a mandatory update. If you don’t like the new layout, your only real option is to move out, which is a massive pain.

Linux, and specifically the reproducible model of NixOS, changes you from a tenant to an owner. I don’t just use the systems I am now using; I define them. If I want a workspace that is entirely keyboard-driven and looks like a 1980s terminal, I can have it. If I want a sleek, modern interface that mimics the best of macOS but without the telemetry back to Mountain View, Redmond or Cupertino (or Beijing for that matter) , I can have that too. That autonomy—the ability to turn the computer back into a personal tool rather than a corporate portal—is deeply satisfying.

Teaching is a craft. Coding is a craft. One craft leads to another. All craft is good craft, to borrow a phrase.

Students know that, of course. Law students know it most of all, which is why this time of year (early January, after Fall grades have been released and before the Spring semester begins) is filled with public and private declarations of law student angst. The angst comes most visibly from first year students, who are partly angry and partly mystified by their results. There is a “standard” law professor response to student complaints, generally speaking: suck it up, Buttercup. Sure, law school grading can seem (and sometimes is) error-filled and arbitrary, but law school isn’t Lake Wobegon (where all students are above average), the mandatory curve is the mandatory curve (because the school says so), professors are human (especially professors who are not Excel-literate), and “this is the way” (cf. The Mandalorian) that “it” has always been done. Law school grades rely on an interface that is deeply alienating. On purpose. For both students and faculty.

In that context, Professor Chandler’s AI-for-grading intervention is, ironically, grounded in a spirit of making the computer system more “authentic” as an instrument of the individual teacher (less alienating!) but has the practical effect of reifying the alienating interface that constitutes the default law school assessment system (more alienating!).

I doubt that there is a right or best solution, if this is a problem. I do want to suggest, however, that law teachers have more and different sorts of assessment agency than they sometimes imagine, if they want to toggle the character of the interface – and if their law schools’ rules permit relevant degrees of freedom. Many don’t want to; some do. I do. I am fortunate that my own law school uses grading curves that have a lot of “play” in them. In my intellectual property law courses, I have never used a multiple choice assessment format; I have not used a comprehensive end-of-semester exam as a sole or primary driver of student assessment in more than two decades. That is also true for the large mandatory courses that I used to teach.

The one-exam-to-rule-the-course approach is used as a default today partly because of path dependence (they go back at least to Christopher Columbus Langdell), partly because professional imaginations and mentorship run only in that direction (we learn to teach by emulating how we were taught, and as seniors, we pay that advice forward), and mostly the one-exam style is a massive time- and labor-saving device for professors. For all of its abundant flaws, today’s law school grading interface is, in a word, efficient. (And AI might make it more so.)

Importantly, there is a strong affinity between the essay format of the bar exam and the format of end-of-semester law school exams. We test in certain ways in law school in order to prepare our graduates to succeed on the bar. Essay exams in law schools came first, if I recall correctly, but it may be fairer to note a process of co-evolution.

So I took note of the recent ABA Journal story about possible upcoming changes to bar exams across several states (“A Dozen Ways That Bar Licensure Could Change in 2026“), reported here at TaxProfBlog by Austen Parrish. Is there reason to hope that changes by bar examiners might motivate law schools and law professors to modify their assessment habits to make the interface more … humane?