Which AI assurance techniques should my organisation use?

Answered

There is no silver bullet. Combine techniques across the life cycle — for example risk assessment, algorithmic impact assessment, bias audit, compliance audit, conformity assessment and formal verification — chosen for context. Low-risk cases may need fewer techniques; high-risk cases need a more robust combination.

From the guidance

Primary (how) Introduction to AI assurance

There will never be a silver bullet for AI assurance. Rather, multiple assurance techniques will need to be used in combination with one another across the lifecycle.

Section: 4.2 AI assurance mechanisms

Read this in Introduction to AI assurance (opens in new tab)

Secondary (normative) Introduction to AI assurance

However, this also allows for a proportionate approach to assurance, with low-risk use-cases able to rely on a smaller range of assurance techniques, and high-risk use-cases utilising a more robust combination of assurance techniques.

Section: 4.2 AI assurance mechanisms

Read this in Introduction to AI assurance (opens in new tab)

Related questions