Skip to content

ACCA Strategic Professional · Strategic Business Leader · Machine learning, AI and robotics

Kestrel Bank trains a credit-scoring model on ten years of lending decisions. Historically, loan officers declined applicants from certain postcodes more often, regardless of repayment ability. The model now declines such applicants at a high rate. What is the most likely cause and implication?

The cause is algorithmic bias inherited from biased historical lending decisions in the training data. The implication is ethical, legal and reputational risk, so the bank needs governance such as data audits, fairness testing and human oversight of the model's decisions.

  1. AOverfitting only, which would be solved by using fewer variables
  2. BAlgorithmic bias inherited from biased training data, creating ethical and reputational risk requiring governance and reviewCorrect
  3. CA hardware fault in the processing system, requiring greater computing capacity
  4. DReinforcement learning, which means the model is correctly rewarded for declining applicants

Explanation

Models learn patterns in historical data, including past human bias. The postcode effect reflects biased training data, so the bank faces discrimination, regulatory and reputational risk. It needs governance such as data audits, fairness testing and human oversight. Overfitting concerns poor generalisation, not inherited bias.

Did you get it right without looking?

One question tells you little. A timed set on Machine learning, AI and robotics shows your real accuracy, how long you take and where you lose marks.

Practise Machine learning, AI and roboticsFree diagnostic mock

More Machine learning, AI and robotics questions