Human-centred AI considers the whole service: the people affected, the staff responsible, the information used and the consequences when a system is uncertain or wrong.
Understand who carries the risk
Impact mapping helps teams see who benefits, who may be excluded and who is accountable for an outcome. This is especially important where people cannot easily opt out of a public service.
Design understandable interactions
People should know when AI is involved, what information it uses and how to challenge an outcome. Clear explanations and accessible routes to human help are core service features.
Govern the full lifecycle
Models and workflows need monitoring after release. Changes in data, demand or policy can affect performance, so ownership, review points and retirement criteria should be explicit.
Key takeaways
What to carry forward
- Assess impact across all affected groups
- Provide explanation and routes to challenge
- Monitor performance throughout the lifecycle
