Model drift

When an AI model's real-world performance degrades because the patients it now sees no longer match the data it was trained on. It is why models need monitoring and revalidation, not just a one-time launch.

A sepsis model trained before 2020 saw a different patient mix than it sees now. Change the EHR vendor, shift the patient population, update a lab assay, and the model that scored 0.90 at launch quietly starts missing cases nobody catches until an audit. This is why a model isn't "done" when it goes live. Someone owns watching its performance every month, or it decays in the dark.

Terms like this come up in real clinical scenarios across the HelloAI courses: bite-sized modules with verifiable certificates. An account takes one minute, no password needed.