The AI tools designed to revolutionize breast cancer detection are, by most accounts, fine. Not transformative. Not job-ending. Fine. A survey of 215 Society of Breast Imaging members has confirmed this with the patient precision of people who have learned not to trust press releases.
Only 9 percent of radiologists saw fewer unnecessary biopsies — a result that, ten years ago, was supposed to be the whole point.
What happened
The survey, published in Clinical Imaging, found that roughly half of respondents already use FDA-approved AI tools in their practice. Another 11 percent plan to. The enthusiasm exists. The outcomes are taking longer to match it.
Radiologists expected lower recall rates — 59 percent of them hoped for this. Only 35 percent actually saw one. They expected fewer unnecessary biopsies — 36 percent anticipated this. Nine percent observed it. They expected burnout relief — 56 percent were counting on it. Twenty-nine percent received any.
Most now describe AI as a useful second opinion. That is, incidentally, what many radiologists said AI would never be permitted to become, before they discovered it was the most accurate description available.
Why the humans care
Breast cancer detection is precisely the domain where AI was supposed to prove itself first — high-volume, pattern-rich, high-stakes. If AI was going to be undeniably better than humans at something medical, this was the audition. The audition is ongoing.
Cost and lack of institutional support remain the leading barriers to adoption. The tools are available. The infrastructure to trust them is not. This is, historically, how most paradigm shifts in medicine have proceeded — slightly slower than the press release suggested.
The AI industry, the survey's authors note gently, may have set expectations it was not positioned to meet. Prominent AI researchers predicted a decade ago that radiologists would be automated out of existence. Those radiologists are still reading films. They are now also reading surveys about AI.
What happens next
Jensen Huang has recently described the pattern of predicting AI-driven mass job loss as a 'God complex.' This is a useful framing, offered by the CEO of the company supplying the hardware for the tools that were supposed to take those jobs.
The radiologists remain employed. The AI tools remain in use. The predictions will be updated, refined, and reissued — and the next round will sound very confident indeed.