Educational Realism in the Age of AI

Princeton just abandoned a 133-year-old tradition. Ohio State spent the same year building something new. The contrast is worth thinking about.

In May, Princeton’s faculty voted to end an honor code that had survived since 1893. Professors will now stay in the room during exams. The vote wasn’t close. Out of more than three hundred faculty, one voted to keep the old system.

The reason was AI. Roughly thirty percent of seniors admitted to cheating last year, and take-home assessments had become, in the words of faculty, “frequently worthless.” Let’s be honest about what that means. Asking ChatGPT to do your homework or take your test is cheating, lazy, and unethical, and Princeton was right to take it seriously. The question isn’t whether the behavior was a problem. The question is whether Princeton diagnosed the right one. Treating AI as an integrity threat to be policed produced one set of answers. Treating it as a transformative tool that requires a curricular response would have produced another.

Princeton didn’t formally vote that AI is cheating. They didn’t need to. Their institutional response treated AI primarily as an enforcement and detection problem, which is functionally equivalent to treating it as a cheating problem. The proctors are the visible consequence of that framing. The framing itself is what set everything in motion.

Almost exactly a year earlier, Ohio State went a different direction. Their AI Fluency initiative ensures that every undergraduate, starting with the class of 2029, graduates fluent in AI and its application to their specific field. Not as a computer science elective. As a core graduation requirement woven through health care, agriculture, the humanities, and every discipline. The provost called it making students “bilingual,” fluent in both their major and the application of AI to that major.

Ohio State isn’t alone. UC Irvine’s humanities school, Notre Dame, Arizona State, and a growing list of institutions are building variations of the same idea. None of them are pretending the developmental questions are settled. They’ve just decided the work is worth doing.

Two institutions, same situation, opposite diagnoses. Princeton saw an integrity crisis. Ohio State saw a curriculum problem. The middle ground, the place where you take cheating seriously without abandoning the work of teaching students how to use the most powerful tool of their working lives, is where the real conversation should be happening.

Who are we helping with the old model?

Strip the romance from the honor code, and you’re looking at an assessment model built around memory, recall, and processing speed under controlled conditions. Bloom’s taxonomy, the framework most education programs still teach, put recall at the bottom of the cognitive hierarchy in 1956. Seventy years later, the assessments that determine who gets into law school, medical school, and the consulting class still lean heavily on the bottom of the pyramid.

Computers have outperformed humans at recall and processing speed for forty years. The honor code survived all of that because the artifacts students produced still looked like evidence of human thinking, even when calculators and search engines were doing more of the cognitive lifting than anyone acknowledged. AI just made the gap impossible to ignore.

Speed and access have changed. Has education?

The World Economic Forum’s 2025 Future of Jobs report surveyed more than a thousand employers representing fourteen million workers across fifty-five economies. AI and big data top the list of fastest-growing skills. Thirty-nine percent of core skills will change by 2030. Critical thinking, creativity, and adaptability rank right behind the technical skills, because leverage compounds on foundation and the workforce knows it.

Students who graduate unable to use AI fluently will walk into a workforce that demands it. That isn’t a forecast. It’s already happening. The question is whether the credential they earn prepares them for the world they’re actually entering or the one their grandparents entered.

Cheating is real. The diagnosis still has to be right.

Math educators fought a version of this fight from roughly 1975 to 1995. Calculators were going to ruin a generation. The resolution took twenty years and came from doing developmental work, not enforcement work. Calculators stayed out of early arithmetic, where children build number sense. They moved in gradually through middle school. By the time students reached calculus, calculators were required equipment. Math research kept producing breakthroughs that would have been impossible without computation. The math department did not get diminished. The math department got bigger.

The institutions that came out of that fight stronger are the ones that figured out where the foundation had to be built by hand and where the leverage had to be taught explicitly. The institutions that just banned calculators in 1985 are not remembered as defenders of rigor. They’re not really remembered at all.

The same research that worries about AI eroding critical thinking also names the buffer that prevents it. Information literacy, critical evaluation skills, and explicit instruction in cross-validation are what separate students who use AI well from those who use it poorly, and the cognitive decline appears only in the second group. Read honestly, the research is an argument for teaching AI literacy as a core requirement, not for treating the technology as the enemy.

How do we build the human skills in an AI-saturated world?

This is where I’d love to hear from people who actually run things. The friction is in the developmental sequence. A freshman writing their first analytical essay still needs to sit with confusion long enough to build an argument from nothing. Skip that work, and you don’t graduate a more efficient thinker. You graduate as a fluent prompter with no internal compass for whether the output is any good.

A graduate student doing a literature review lives in a different reality. The foundation is mostly built. The job is synthesis at a scale where AI is the new floor of competent work, not a luxury. Banning it in that context isn’t conservatism. It’s malpractice.

Designing an assessment that respects this developmental arc is hard. It requires oral defenses, process portfolios, in-class drafting where the AI use itself becomes part of the assessment, and capstone work where the entire grade rests on the student’s discernment with AI assumed as a starting tool. This is the work Ohio State and the others are starting. It’s the middle ground Princeton declined to occupy when they reached for proctors instead.

I’ll be honest about my own bias. Princeton’s response looks to me like an institution holding on to a tradition that no longer serves its students, at considerable cost to those students’ readiness for the world they’re entering. But the question is bigger than Princeton, and I’d rather hear how other people are thinking about it than win the argument.

If you’re a faculty member, an administrator, an employer hiring graduates, or a parent watching this unfold, I’d genuinely like to know: how do we build the developmental sequence right? Where should friction stay, and where should leverage take over? And what does an honest, modern credential actually look like?