Bringing human intelligence to AI

Consider a hiring team using AI to help review applications. One applicant is screened out because the information used to assess them is wrong. Someone needs to be able to hear that applicant, correct the record, and reconsider the decision. The usefulness of the tool depends in part on whether the organization has made that possible.

Hi on AI begins with this kind of question. Hi stands for human intelligence: the experience and judgment people bring to a decision, including their understanding of those affected. Bringing it to AI means deciding where the tool can help and what people must remain able to do.

Our published opinions and radio conversation explore how respondents can welcome useful AI while wanting a say in its use. People can value assistance with a task and still expect a person to be available when the consequences become important. Leaders need to understand those expectations before deciding how much authority to give a system.

Trust in AI research overview ↗

Start with the work people need to do.

Choose a task where the result can be checked. AI might help someone organize information or prepare a first draft. Decide what a better result would look like and who will assess it. As the consequences increase, the review needs to be more demanding.

The person doing the checking needs time and enough experience to question the answer. Employers should consider how staff will acquire that experience as AI changes the work assigned to beginners.

Make human involvement practical.

A promise of human oversight needs an explanation of what the person can do. Can they inspect the information behind a result? Can they stop the process or change the outcome? Can the person affected reach them?

These questions help an organization decide what review requires. They also help a customer or employee understand what will happen if something goes wrong.

Listen to the people living with the decision.

The same approach applies to a community considering a data center. A customer buying the service, a neighbor living beside the facility, and a ratepayer sharing infrastructure costs may have different questions. Research should examine their interests directly.

Our Detroit News column brings the AI findings into that discussion as recommendations for project leaders. The surveys describe respondents’ views on AI; local research is still needed to understand a particular community and project.

Ask what needs to be checked next.

Before expanding an AI use, name what it is meant to improve and how that improvement will be assessed. Explain who owns the decision, how people can challenge it, and what would cause the organization to change course.

Read the published pieces and listen to the radio conversation. Bring us the decision your organization is trying to make.