Research program
What earns trust when AI becomes part of a decision?
A research frame for learning where people welcome assistance and where they want a human. It also asks what accountability requires.
Trust is not one question
People may trust AI to condense information and reject it as the final judge of a high-stakes choice. They may care less about whether a system is called AI than about who selected it and what it can see. Responsibility when it is wrong is a separate test.
Research should separate comfort with the task from trust in the institution using the tool.
What we would measure
A useful study would vary the task and stakes. It would also vary human review and the source of the data. The explanation given to the person affected becomes another part of the test. Interviews can reveal the language people use for control and accountability; choice exercises can show which protections change acceptance.
- Where people draw the line between assistance and authority.
- Which explanations improve understanding and which sound like evasion.
- How human review and a right to appeal affect willingness to rely on the system. The ability to correct a decision belongs in the test too.
- Whether trust changes across the different groups affected by the decision.
What this note does not claim
Attitudes about AI change with the use case and the institution. A general trust score cannot substitute for research tied to the real decision and population, then judged against the consequences.
Choices this research can support
Leaders can use the work to decide where AI belongs in a process and what requires human judgment. The research can test disclosure and the safeguards that affect trust. It can also show how to explain the change without overstating certainty.
Who this helps
An organization introducing AI into a decision about customers or employees can use this frame. The same applies when donors or students are affected, or when the public depends on the outcome.