The End of College as We Know It

“What's the matter with universities? Syracuse missed its 2026 enrollment target and may run a $30 million deficit. Minnesota is cutting its budget by $225 million over two years. Tulsa is slashing tuition by more than half. Colleges enroll 4.2% fewer undergraduates than last year. Cornell, Yale, Vanderbilt and Washington University in St. Louis put out plans -- not radical enough, if you ask me -- to reimagine the university. Wossamotta U?

 

Colleges are known for left-leaning professors, speech suppression, bizarre courses -- Occidental College taught "The Unbearable Whiteness of Barbie" -- and easy grading more than critical thinking, judgment and problem solving. It's easy to blame artificial intelligence, but problems run deeper:

 

-- Aptitude. In Griggs v. Duke Power (1971), the Supreme Court effectively banned corporations from giving intelligence tests, so companies rely on colleges to do their sorting via aptitude tests like the SAT. That led to branding and credentialism, but that game is over. A Cornell report noted employers' "jarring" skepticism of college graduates because they lack job skills such as "the ability to handle uncertainty and solve problems that do not have clear answers." A Forbes survey found almost half of corporate executives are either less likely to hire Ivy League graduates than they were five years ago or would never hire them.

 

-- Admissions. "Holistic reviews" that ignore aptitude and grades put a thumb on the scale and harm college brands. In combination with athletic scholarships and easy majors, they bring in underqualified students who are churned out as subpar employees.

 

-- Accounting. In 1945, Vannevar Bush suggested the U.S. government fund "basic research in the colleges, universities and research institutes," which became a big business. But universities got greedy and abused the system by overcharging the government for overhead.

 

That business model is now suspect. The Trump administration instituted a 15% cap on "indirect cost rates." A 2025 study found that at more than 350 institutions funded by the National Institutes of Health, the average negotiated indirect cost rate was 58% and the effective actual overhead rate averaged 42%. The Trump administration's 15% cap was struck down in most cases and the administration has backed off for now, but we've seen research funds cut off. Watch for "let's hit up alumni for more" as research overhead may no longer be an overstuffed cow to pay for campus excesses.

 

That includes professor salaries. Many eyebrows lifted in 2024 when Claudine Gay was fired as Harvard president and went back to being a professor, at a $900,000 salary. The average full professor salary at Harvard in 2026 is $293,619, up 6.4% from last year. I'd bet many of them hate capitalism but are happy to enjoy its fruits.

 

-- Affordability. Student loans lead to constant tuition hikes. Same for high professor salaries and administrative bloat. Sixteen colleges and universities cost more than $100,000 a year in tuition and expenses. Where's the value? Students could buy a decent GPU rack and start their own AI company for less.

 

-- Artificial Intelligence. Professors and students seem to be going through the motions of learning, so there's a huge structural problem. Do ChatGPT-written essays get graded by Claude? Why bother?

 

Three years ago, OpenAI's GPT-4 scored 1,410 on the SAT, maybe enough to get into Michigan. Since then, AI has passed bar exams and medical licensing exams, which says more about these outdated tests than the inadequacy of humans, perhaps a sign of an antiquated education system.

 

Even worse is a 2025 Harvard study, "AI tutoring outperforms in-class active learning." [1] Are teachers and professors becoming obsolete?

 

Universities can take advantage of AI rather than ban it. The sage-on-a-stage format is over. Learning is experiential. As corporate America rejects college branding, graduates would benefit from more hands-on experience. Universities may be uniquely positioned to implement what I call "virtual apprenticeships," real job experience while still at school.

 

I suggest all students spend their entire junior year on team projects. Assignments might include building nuclear autonomous buses or robots that cook. Cross-discipline self-organizing teams will form. You'll need engineers and coders, but also prelaw and political-science majors to clear the regulatory path, philosophers and marketing majors to write press releases, designers, nutritionists, etc.

 

Self-paced AI-tutored instruction can help students learn what they need as problems arise. A smart entrepreneur might even create a student transfer portal to recruit talented students from other universities.

 

According to Gallup, in 2013, 70% of Americans said college was very important. Today it's only 31%. If universities don't change, more than a few will implode, while others will wither away over decades. AI and alternative forms of learning are already rising. Something radical is needed to stop the decay of our once great universities. Who will step up?” [2]

 

1. The study AI tutoring outperforms in-class active learning: an RCT introducing a novel research-based design in an authentic educational setting did not use a commercial or off-the-shelf tutoring service with a retail price tag; instead, it used a custom-built, research-designed AI tutor (referred to in project materials as PS2 Pal) developed specifically by the Harvard researchers for the experiment.

Study and Service Details

•           Service Name: A custom, purpose-built AI tutor engineered by the research team using generative AI, structured pedagogical scaffolding, guided problem-solving, and tailored multi-step prompting rather than an open or unguided chatbot.

•           Price: Free / Not Applicable (Experimental Prototype). The tool was created internally by Greg Kestin, Kelly Miller, and colleagues at Harvard University for 194 introductory physics students and had no commercial price or consumer subscription cost attached to it.

 

The 2025 Harvard study published in Scientific Reports demonstrated that a pedagogically structured AI tutor ("PS2 Pal," built on GPT-4) yielded more than double the learning gains in introductory physics compared to traditional in-class active learning—all while taking less time.

 

The key to this study's success was not just raw AI capability, but pedagogical prompt engineering: the AI was explicitly barred from giving away answers, relying instead on step-by-step verified solutions, adaptive scaffolding, and guiding students to figure out problems dynamically.

Several commercially available AI tutoring systems mimic this structured, scaffolding-based approach rather than acting as mere "answer engines":

Commercially Available AI Tutors with Structured Scaffolding

 

•           Khanmigo (by Khan Academy): Developed in close partnership with OpenAI, Khanmigo is arguably the closest commercial equivalent to the Harvard model. It is intentionally programmed as a Socratic guide. When a student struggles with a math or science problem, it refuses to provide the direct answer. Instead, it asks probing questions, identifies the specific step where the student made a mistake, and scaffolds the path to the solution.

 

•           Synthea (by Alpha School): Referencing the exact methodologies highlighted by the Harvard research, Synthea and Alpha School's underlying software focus on adaptive digital mastery. The platform guides students through rigorous core curriculums using AI feedback loops that dynamically adjust problem difficulty based on real-time performance without bypassing critical thinking steps.

•           Q-Chat (by Quizlet): Built using OpenAI’s GPT models, Q-Chat acts as a personal tutor that applies conversational scaffolding. It uses adaptive learning techniques like deepening comprehension through targeted follow-up questions, testing a user's knowledge from multiple angles, and utilizing a chat-based interface to keep the user actively engaged rather than passively reading material.

•           Century Tech (Century): Primarily utilized across institutional and school settings, this platform uses AI and cognitive neuroscience to create "nuggets" of structured micro-learning. It constantly diagnoses a student's knowledge gaps and adapts the pacing and structural guidance of lessons, ensuring students actively work through conceptual bottlenecks.

Core Comparison: Standard AI vs. Structured AI Tutors

Feature          Standard AI (e.g., vanilla ChatGPT / Gemini)      Structured AI Tutors (e.g., Harvard Study Model, Khanmigo)

Primary Objective    Fast output generation and direct answers.        Conceptual mastery and cognitive growth.

Pedagogical Strategy          Tells the user the solution immediately.    Employs Socratic questioning and sequential scaffolding.

Mistake Handling     Corrects the error for the user.       Prompts the user to locate and fix their own logical missteps.

Pacing            User-driven entirely. Adaptive data tracking that scales difficulty dynamically.

 

2. Inside View: The End of College as We Know It. Kessler, Andy.  Wall Street Journal, Eastern edition; New York, N.Y.. 05 Oct 2026: A15.