Skip to content

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

Halden Retail uses a machine learning model to shortlist job applicants. The model was trained on ten years of past hiring decisions, during which managers mostly hired men for technical roles. The model now ranks female applicants lower even though gender is not an input. Which AI risk does this primarily illustrate?

The risk is algorithmic bias. The model learned from historic hiring data that favoured men, so it reproduces the same discrimination through proxy variables even when gender is excluded. Biased training data, not a technical fault such as latency or intrusion, is the cause.

  1. AAlgorithmic bias arising from unrepresentative or prejudiced training dataCorrect
  2. BModel overfitting caused by too few input variables
  3. CData latency caused by slow system processing
  4. DCyber intrusion through adversarial attack on the model

Explanation

The model has learned patterns from historic decisions that reflected prejudice, and proxies for gender can reproduce that outcome. This is algorithmic bias from training data. Overfitting concerns poor generalisation to new data, not discriminatory patterns inherited from the data itself.

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