Perfect group fairness often requires compromising overall accuracy.

Get ready for the GARP Risk and AI Exam with flashcards and multiple choice questions. Each question comes with hints and explanations. Prepare for success!

The main idea here is that aiming for perfect fairness across different groups often means giving up some overall predictive accuracy. When groups differ in base rates—the frequency of the positive outcome within each group—a single model with one threshold will typically produce unequal performance across groups. To make performance equal (or to meet a strict fairness criterion like equal true positive rates), you frequently need to adjust thresholds or model behavior per group. Those adjustments can reduce the model’s overall accuracy because you’re not optimizing for the highest possible correct predictions across all data, you’re sacrificing some accuracy to equalize outcomes.

Briefly, explainability and value judgments aren’t the core dynamics at play here. Equal Opportunity is a fairness criterion that focuses on equal true positive rates, but the statement is about the broader trade-off between achieving group fairness and maintaining overall accuracy, which is captured by the idea of a performance vs. fairness tension.