Proxy discrimination and disparate impact
Unfair outcomes caused indirectly, through variables linked to protected characteristics.
Even without using race, gender or age directly, a model can discriminate through correlated features such as postcode or first name. Disparate impact means a seemingly neutral process harms a protected group disproportionately.
On the exam
Removing protected attributes doesn’t guarantee fairness. Test outcomes across groups.
Related terms
- Bias and fairnessWhether an AI system treats groups of people unfairly.
- Features and feature engineeringThe input variables a model uses, and how they are built.
- Fairness metricsMeasures for comparing a model’s outcomes across groups.
AAIA Test yourself on this with timed AAIA practice questions in the AAIA Exam Simulation Hub, or read the free AAIA exam guide.
Test yourself with today’s question of the day.
