What would change if the public disagreed?

Public engagement creates legitimacy only when people can influence the purpose, boundaries or conditions of an AI system before the important decisions are settled.

In today’s governance note, I want you to ask a direct question before an AI project is approved: what would change if the public disagreed with your proposed approach?

Organisations often engage people after the purpose has been agreed, the technology selected, and the implementation plan largely settled. At that stage, engagement becomes an exercise in explaining the decision clearly and managing the response.

That may be useful communication. It is not meaningful participation, as public legitimacy does not come from informing people that an institution has considered their interests. It comes from showing that those interests had a place in the decision. This question should therefore arise while the organisation can still change the system’s purpose, design, boundaries or conditions of use. It is especially important where AI affects access to services, employment, education, public information or the distribution of opportunities.

Begin by identifying who will experience the system rather than only who will operate it. The people most affected may not be the direct users. A caseworker may interact with the interface while a family experiences the recommendation it produces. An employee may never see a workforce system that influences how their performance is interpreted. A member of the public may encounter an automated service without knowing which institutional choices shaped it.

You should then decide what remains genuinely open to influence. Can participants challenge the problem definition? Can they identify circumstances that the proposed data will not represent? Can they recommend limits on use, additional safeguards or a non-digital route? Could their evidence lead you to delay or decline the deployment? If the answer to all of these questions is no, you are not asking people to participate in the decision. You are asking them to receive it.

This does not mean that every view must determine the outcome. Governance requires institutions to weigh competing interests, legal duties, operational constraints and public value. But you should be able to explain what you heard, what changed, what did not change and why. I would establish that participation mandate before inviting anyone into the process. Define the decisions that remain open, the groups whose experience is needed, the evidence you are seeking and the authority responsible for responding. Then record the resulting changes alongside the project’s technical and risk decisions.

UNESCO’s Recommendation on the Ethics of Artificial Intelligence calls for inclusive participation throughout the AI lifecycle. Its Ethical Impact Assessment translates that principle into practice by asking teams to identify relevant stakeholders, map positive and negative effects and engage people early, including during research, design and procurement. UNESCO Recommendation on the Ethics of AI, UNESCO Ethical Impact Assessment

This guidance matters because social consequences cannot be assessed entirely from inside the institution. Your project team knows how the system is intended to work. The people affected can tell you how it may be encountered, misunderstood, avoided or experienced alongside existing inequalities. Engagement should not be the final stage at which you introduce the public to your AI system, but rather it should be moved to one of the earlier stages at which the public can still help you decide what that system is permitted to become.