Fairness
The property of an AI system treating individuals and groups equitably and without unjust discrimination.
Full Definition
Fairness in AI refers to ensuring that model outputs, performance, and impacts are equitable across demographic groups defined by race, gender, age, nationality, religion, disability, and other attributes. Fairness is not a single concept: demographic parity (equal positive prediction rates), equalised odds (equal true positive and false positive rates), and individual fairness (similar individuals treated similarly) can be mathematically incompatible. This means choices about which fairness criterion to optimise are inherently value-laden. Fairness auditing involves evaluating model performance on representative demographic subgroups, documenting disparities in model cards, and implementing targeted mitigation for the most harmful disparities.
Examples
Auditing a loan approval model and finding it denies credit at twice the rate for Black applicants with identical financial profiles as white applicants.
A resume screening model flagging that it recommends female candidates for 'support roles' and male candidates for 'leadership roles' at significantly different rates.
Apply this in your prompts
PromptITIN automatically uses techniques like Fairness to build better prompts for you.