Which of the rationales I outlined last Tuesday for traditional higher education still hold up against AI? As I noted in a later post, the answer depends on the college-major pair. A finance degree from Wharton and a psychology degree from a commuter college are different products, so AI will affect them in very different ways, and we need to always think at the margin. The twelve rationales fall into three categories: 🅐 Mostly resilient to AI: signaling, networking, and cultural capital at highly selective institutions, commitment for time-inconsistent students, the hold-out period for traditional-age students, proximity to the research frontier at research universities, and physical infrastructure. 🅑 Highly vulnerable: skill acquisition, topic curation, and assessment. 🅒 Entirely dependent on the college-major pair: credentialing (robust where statutory, fragile where normative), peer effects, and cultural capital (robust at selective institutions, negligible elsewhere). Let me go through them one by one. ① Signaling. Robust at the top, irrelevant at the bottom. The signaling value of a STEM degree from an Ivy is untouched by AI, because the signal comes from admission and from the ability to complete the degree. By contrast, a degree in humanities from a non-selective institution is a weak signal, because both admission and completion are easy. AI opens the door to alternative ways of assessing competence (for example, personalized evaluations) that may soon compete with degrees from weaker institutions. So the student who was already indifferent between attending a mid-ranked institution and entering the labor market now has a third option.