G11 SUPER GIO
123
03
CHALLENGE MISSIONtraining data, prediction, bias, and accountability

training data, prediction, bias, and accountability

INDEPENDENT DEFENSEINDEPENDENT TRANSFER

Human review limit

A benefits model flags fraud with 85% recall but 40% precision. Automatic denial would wrongly deny 60 of every 100 flagged cases.

  1. Recall and precision answer different questions.
  2. Low precision creates high human cost.
  3. A flag can triage without deciding.

Which deployment is defensible?

MODELModel Responsibility Card
DATA
PROVEChoose with evidence
CLUE

Pause. Find what matters. Then make your move.