Software can help organize choices, identify patterns and recommend actions. The larger the consequences, however, the more important it is to understand what the software can and cannot reliably determine.
Oversight can include checking the data used by a system, testing outputs, documenting who can override a recommendation and creating a route for people to challenge errors.
The U.S. National Institute of Standards and Technology treats accountability, validity, reliability and risk management as important considerations in AI development and use.
The goal is not to reject automation. It is to match a tool’s responsibilities to its demonstrated abilities and the potential cost of failure.
This article is a source-linked explainer, not a claim of firsthand interviews or exclusive reporting.
We want errors brought to our attention and corrected openly.
Corrections and feedback