A city AI strategy office is under pressure to launch an automated benefits-eligibility screening tool by a publicized deadline, but disparate-impact fairness testing across demographic groups has not yet been completed. What should the office do?
Select an answer to reveal the explanation.
Short Explanation
Think about it like a bridge inspector refusing to open a new overpass on schedule because the load test isn't done yet: the deadline pressure doesn't change what's underneath the surface. A benefits-eligibility tool that hasn't been checked for disparate impact could quietly steer real hardship toward already-vulnerable residents, so responsible AI leadership means the fairness check comes before the launch, not after it.
Full Explanation
Responsible AI leadership requires prioritizing fairness validation ahead of schedule pressure whenever an AI system makes or heavily influences high-impact decisions affecting residents, because an eligibility tool that systematically disadvantages a demographic group causes real harm the moment it goes live, not after someone notices. Launching first and testing in parallel treats fairness as a bug to patch rather than a precondition, which means residents could be wrongly denied benefits during the exact window the office is supposed to be validating the system. Launching a limited version without disclosing the disparate-impact analysis compounds the problem by adding a transparency failure on top of an unvalidated model, since residents and oversight bodies have no way to know the tool hasn't been checked. Reframing the public messaging addresses optics rather than the actual risk, and does nothing to protect the residents the fairness testing exists to serve. A relevant scope caveat: not every AI system warrants this level of delay tolerance, low-stakes internal tools may reasonably launch with fairness monitoring built in after go-live, but eligibility determinations for public benefits sit at the high end of impact. A concrete operational check: require sign-off from the governance board confirming disparate-impact testing results across protected demographic groups before any go-live date is finalized.