A seismic shift is underway in higher education, fueled by a massive bet on Artificial Intelligence. The core assumption? That AI-delivered instruction is a proven, effective substitute for human teaching, warranting a reallocation of resources. However, a comprehensive 2026 re-analysis of nearly the entire AI-and-learning literature delivers a starkly different verdict. After correcting for severe publication bias, the much-touted learning gains from AI shrink dramatically, potentially to zero. We are gambling on an Unknown while devaluing the most Known-Known in education's history: the human mind.

3 Actionable Steps to Cultivate "Human-Era" Learning Skills
- Seek Cognitive Dissonance: Actively engage with ideas that contradict your own. Join a debate club, read philosophy from opposing schools of thought, or have a structured debate with a colleague on a work issue. The goal is not to win, but to sit in the discomfort of incompatible meanings.
- Practice "Explain-Back" Teaching: After learning something new, explain it to a friend or colleague without preparation. Their questions and your struggle to clarify will reveal gaps in your understanding that no AI quiz can uncover.
- Build "Socratic" AI Prompts: Don't just ask AI for answers. Prompt it to: "Act as a devil's advocate against my following thesis..." or "Ask me five progressively deeper questions about this topic." Use AI to simulate the friction of human dialogue.

The Science Behind the Chaos: What the Meta-Meta-Analysis Really Shows
Researchers Bartoš and Wagenmakers found the literature on AI in learning isn't just "mixed"—it's incoherent. The variation between studies is so vast that the evidence cannot predict if a future AI tool will help or hurt learning. Why? Because most studies measure shallow knowledge recall, not the deep, transferable understanding that defines true education.
"Human cognition is the known-known... The capacity of one mind to engage another in the friction-laden, unpredictable, embodied process through which understanding actually forms."
The "friction"—the tangent, the misunderstood question, the real-time adjustment by a teacher—is not a bug in human teaching. It's the essential feature. AI optimizes for frictionless content delivery, which may inadvertently strip away the very conditions necessary for complex meaning-making.
The next time an AI generates a flawless essay for you, or your institution announces another AI-driven initiative, pause. Consider what might be lost in the pursuit of seamless efficiency. We risk outsourcing not just content delivery, but the core cognitive processes of questioning, integrating, and truly understanding that human interaction uniquely fosters. The future belongs not to those who know how to use AI, but to those who know what only humans can do.
Source & Further Reading: AI in Education Is an Unknown, Humans Are Not