Skip to content

FRM Part II · FRM Exam Part II · The Financial Stability Implications of Artificial Intelligence

A bank's AI fraud-detection model flags a much higher share of transactions from one customer group, though fraud rates are similar across groups. Investigation shows the historical labels reflected past biased manual reviews. Which is the most accurate diagnosis?

The problem is biased training data being propagated by the model. Historical labels from biased manual reviews taught the model to over-flag one group, so it is a data quality and fairness risk that more of the same data will not fix.

  1. AOverfitting caused by too few model parameters
  2. BBiased training data being propagated by the model, a data quality and fairness riskCorrect
  3. CLiquidity mismatch in the model's inputs
  4. DPure random error that will average out with more transactions

Explanation

Labels produced by biased past decisions embed that bias, and the model learns and reproduces it. More data of the same kind would not remove it. This is a data quality and fairness issue, not underfitting, liquidity or random noise.

Did you get it right without looking?

One question tells you little. A timed set on The Financial Stability Implications of Artificial Intelligence shows your real accuracy, how long you take and where you lose marks.

Practise The Financial Stability Implications of Artificial IntelligenceFree diagnostic mock

More The Financial Stability Implications of Artificial Intelligence questions