Knowing What You Don't Know
I, Cringely, Thursday, June 4th, 2026
AI's real breakthrough requires machines that can recognize and admit the limits of their knowledge.
Robert Cringely argues that the crucial next step in AI development isn't computational power or efficiency, but rather building systems that acknowledge their own ignorance.
He concedes a reader's point about the Jevons Paradox, that making AI cheaper will only increase demand, but pivots to a more fundamental problem: current models confidently generate false information without hesitation or calibration.
He contends that enterprises won't deploy systems prone to confident errors in critical domains like healthcare, law, or finance, which explains why AI deployments stall despite technological advances.
Drawing on Socrates' insight that true expertise includes recognizing the boundaries of one's knowledge, Cringely distinguishes an AI junior analyst that answers everything from a senior one that acknowledges uncertainty.