For Heads-of · Practitioner
Protected Characteristics
The nine characteristics the Equality Act 2010 protects from discrimination in Great Britain: age, disability, gender reassignment, marriage and civil partnership, pregnancy and maternity, race, religion or belief, sex, and sexual orientation. They are the legal anchor for AI fairness in the UK.
- Equality Act 2010
- indirect discrimination
- disparate impact
- proxy discrimination
- fairness
Direct and indirect discrimination
Direct discrimination is treating someone less favourably because of a protected characteristic. Indirect discrimination is applying a rule that looks neutral but puts people who share a characteristic at a particular disadvantage, unless it is a proportionate means of achieving a legitimate aim (section 19). The US terms disparate treatment and disparate impact describe similar ideas, but the UK tests differ.
How AI discriminates without seeing the characteristic
Removing protected characteristics from a model's inputs does not prevent discrimination. Postcode, purchasing patterns or employment history can act as proxies, and a model will use them if they predict the outcome. That is indirect discrimination in the making, and the register entry is proxy discrimination via correlated features.
What good looks like
Test outcomes across protected groups even when the model never sees the characteristic, which may mean holding that data separately for testing; record the legitimate aim and why the approach is proportionate; and keep testing after deployment. Bias testing and fairness monitoring is the control, and fairness and bias covers the wider ground.
Sources
Related Terms
Confusion Matrix
A table comparing a classifier's predictions with the actual outcomes, from which precision, recall and false positive and false negative rates are calculated.
Fairness & Bias
The principle and practice of ensuring AI systems do not produce systematically different outcomes for individuals or groups because of protected characteristics or other vulnerable attributes. Algorithmic fairness is the same idea expressed as measurable criteria.
Fairness Metrics
Quantitative tests of whether a model's outcomes or errors differ across groups. The main families are demographic parity, equalised odds and calibration, and outside special cases they cannot all be satisfied at once, so choosing which to use is a governance decision rather than a technical default.
Training Data
The dataset used to train a machine learning model, from which the model learns patterns and relationships.