AI and Intuition (theheasman.com)

This piece argues that “intuition” — the tacit, experience-driven judgment people develop from routine work and failures — is underappreciated in debates about AI in the workplace. Drawing on Robin Hogarth’s Educating Intuition and everyday anecdotes (the good waiter, generational social-intelligence gaps), the author distinguishes three kinds of non‑deliberative know-how: flawed System 1 heuristics, expertise from deliberate practice, and the middle-ground intuition formed by day‑to‑day problem solving. In data work this shows up as being able to spot bad dictionaries, noisy captures, duplicate keys, or when a story “feels” wrong despite fitting formal criteria. A concrete example: using CoPilot/ChatGPT Enterprise to remove empty columns in legacy SAS tables produced a convoluted proc-means solution that errored; the human’s simple SQL-join approach (count vs. non-null counts, then drop columns) was easier to debug, explain, and surface useful dataset structure. The takeaway for the AI/ML community: large models can’t easily replicate tacit heuristics or the introspective cues experts use during iterative workflows. Builders should prioritize tools that capture and surface human heuristics, support simple, inspectable strategies (KISS), and evaluate models on collaboration, interpretability, and exploratory robustness — not just final accuracy. Educating and preserving human intuition remains essential as AI augments knowledge work.