Radiologists have not disappeared despite a prominent 2016 prediction that computers would replace them within five years. Their numbers are expected to grow by at least 26 percent over the next three decades, even as imaging becomes medicine’s largest testing ground for artificial intelligence.
By early 2026, roughly three-quarters of the 1,400 AI-enabled medical devices cleared by the US Food and Drug Administration were designed for radiology. Some draft reports or prioritize urgent scans. Others can detect patterns that are difficult for people to see; an analysis of 43 clinical trials, for example, found that AI-assisted colonoscopies identified more polyps than conventional procedures.
The harder problem is oversight. A system that performs better on average will still make some errors a radiologist would avoid, while neural networks often cannot explain a decision in familiar clinical terms. Doctors must recognize the unusual case where a generally reliable tool is wrong without becoming numb to its repeated correct suggestions.
That makes radiology a practical lesson for other expert professions. Better benchmark accuracy alone does not define a safe workflow; hospitals also need clear responsibility, monitoring and a way to combine the different strengths and failure modes of people and machines.