An Air Force order studying "calibrated reliance" on AI (DCS Corporation) — a federal contract (USAspending)

"Calibrated reliance" describes the idea that a person using AI or machine learning (systems in which a computer learns patterns from data to make judgments or predictions) should keep their trust in the output at a level matched to how reliable that output actually is. Over-trusting AI means following flawed output uncritically; under-trusting it means failing to make use of genuinely helpful support. Avoiding both extremes and adjusting the degree of trust to the situation is regarded as a central challenge in human-AI collaboration. The description of this contract names exactly this theme.

This matters because correctly judging how reliable AI is can be the key to using it safely and effectively. AI is not infallible: there are situations it handles well and situations it does not. Only when the people using it understand where that line falls — and can separate the moments to trust it from the moments to be cautious — does AI become a tool one can rely on with confidence. The fact that the buyer is the Department of Defense and the Air Force signals that this theme is being explored in a domain where the accuracy of judgments carries particular weight. The original record does not state the specific research results or deliverables, and none are assumed here.

Viewed more broadly, this contract is one example of the federal government funding not only "how to build AI" but also the human side — "how people should trust and use AI." As AI becomes more widespread, designing the trust between people and AI grows in importance alongside the raw performance of the technology itself. Reading across contracts like this one published on USAspending offers a way to see which stages of putting AI into practice the public sector is emphasizing.