Skip to content

ACCA Strategic Professional · Advanced Performance Management · Technology and information systems

Marlow Insurance has introduced a machine learning model to set bonuses for claims handlers by scoring each handler's performance. The model was trained on five years of past bonus decisions. An internal review finds that handlers in one regional office are consistently scored lower despite equal measured outputs. Which is the most appropriate interpretation and response?

The model may be reproducing bias contained in its historic training data. Marlow should audit the data and the outcomes for fairness and keep human oversight of reward decisions. Automation is not automatically objective, and switching to unsupervised learning or hiding behind unexplainability would not solve the problem.

  1. AThe model is objective because it is automated, so the scores should be applied and the review ignored
  2. BThe training data may embed historic human bias, so the company should audit inputs and outcomes, and keep human oversight of the reward decisionsCorrect
  3. CThe only solution is to switch to unsupervised learning, which cannot be biased
  4. DThe scores should be kept but the regional handlers should be told the algorithm cannot be explained

Explanation

Models trained on past decisions can reproduce historic bias, so inputs, outcomes and fairness should be tested and human oversight retained, particularly where rewards and motivation are affected. Automation does not guarantee objectivity. Unsupervised learning can also be biased, and non-explainability undermines the fairness and acceptance of reward systems.

Did you get it right without looking?

One question tells you little. A timed set on Technology and information systems shows your real accuracy, how long you take and where you lose marks.

Practise Technology and information systemsFree diagnostic mock

More Technology and information systems questions