The responsible AI credibility problem
The phrase 'responsible AI' has been used so frequently and with so little consistent meaning that it has nearly lost its utility. AI vendors use it to describe products. Governments use it to describe policy aspirations. Academic researchers use it to describe technical properties. And organisations use it to describe their governance commitments, regardless of whether those commitments have been operationalised into actual practices. The result is that 'responsible AI' functions more as a signal of good intentions than as a description of specific, verifiable practices.
The operationalisation problem is not unique to AI — every governance framework faces the challenge of translating principles into practices. But it is particularly acute for AI governance because the gap between principles and practices is so wide and so consistently unexamined. An organisation that has a responsible AI policy document and an ethics board but has never bias-tested a single AI system is not practising responsible AI regardless of what its documentation says.
Fairness in practice
Fairness as a responsible AI practice means: identifying which demographic groups the AI system affects, specifying which fairness metric is appropriate for the AI's decision context, testing the AI system against that metric before deployment, documenting the results, setting acceptable thresholds, and monitoring fairness in production. It does not mean stating that the organisation is committed to fair AI. Fairness testing methodology varies by context: demographic parity (equal positive outcome rates across groups) is appropriate for some decisions; equalised odds (equal true positive and false positive rates across groups) is appropriate for others; individual fairness (similar individuals treated similarly) is the right standard for others. The choice of fairness metric is itself a governance decision that must be made deliberately and documented.
