CONCEPT
The Jagged Frontier
Ethan Mollick’s image for the invisible, irregularly shaped boundary of AI capability—inside it the machine is superhumanly competent, outside it the machine fails with fluent confident wrongness, and the border between them respects no human intuition about difficulty.
Picture a fortress wall whose towers and battlements jut irregularly into the surrounding countryside: some towers far out, some sharply receded. Inside the wall, the AI is competent, often superhumanly so. Outside it, the AI fails—sometimes obviously, more often with a fluent, confident wrongness that is far more dangerous than visible failure. The jagged frontier is
Ethan Mollick's name for this boundary, and its most important feature is its invisibility: the wall cannot be seen in advance, its shape is wildly irregular, and it does not respect human intuitions about what constitutes a difficult task. Two tasks that appear equally demanding to a practitioner may sit on opposite sides. A model that produces a sophisticated strategic analysis may stumble over a simple arithmetic constraint embedded in the same response. A system that handles legal prose with apparent fluency may invent case citations with equal fluency and equal apparent confidence. Experience with the system on one kind