I’m studying for a professional AI exam while simultaneously preparing deliverables for an AI governance engagement at a global bank. The exam covers fairness measures, model governance, explainability, regulatory frameworks. The deliverables require me to think through fairness measures, model governance, explainability, regulatory frameworks.
For weeks I had two separate TODO lists: a cramming schedule (phases, MCQ drills, practice exams, kill-lists) and a deliverable plan. My AI assistant finally asked the obvious question: why are these separate?
The exam is a proxy for understanding AI governance. The bank deliverables are the real thing. Cramming for the proxy — memorise the fairness formula, drill until you can recite the difference between demographic parity and equalised odds — produces exam-passing knowledge that evaporates by the following Monday. Thinking deeply for the real thing — working through fairness tradeoffs on an actual credit scoring use case — produces understanding. And might pass the exam anyway, as a side effect.
The heuristic: when you have both a credential and real work in the same domain, route effort through the real work. The credential either follows or it doesn’t. If it does, you got both. If it doesn’t, the real work was still worth doing. The reverse — passing the exam but producing shallow deliverables — is the worse outcome in every scenario.
This generalises beyond exams. Any time you notice yourself optimising for a metric that’s supposed to represent something else — interview prep that doesn’t make you better at the job, code coverage that doesn’t catch bugs, OKRs that don’t move the business — ask whether you have access to the real thing. If you do, go direct. The proxy is only valuable when the real thing is unavailable.
The sunk cost of the exam fee tried to pull me back toward the cramming schedule. But $750 already spent is $750 already spent, regardless of whether I memorise Elastic Net regularisation. That’s not learning. That’s loss aversion wearing a study plan as a disguise.