Can an accurate model be bad?

McCradden, Melissa D; Mazwi, Mjaye L; Oakden-Rayner, Lauren · Patterns (N Y) · 2025

editorial · Level V

Where this comes from

Abstract

Outcome-prediction models can harm patients even when they have good accuracy, as shown in a recent <i>Patterns</i> paper by Van Amsterdam et al. In this preview, we consider the ethical and empirical implications of this work by highlighting the impact of reifying self-fulfilling prophecies and propose a reorientation toward actions over accuracy as a priority for AI integration.