AI won't replace radiologists, but will change their jobs

Geoffrey Hinton's 2016 prediction that computers would replace radiologists within five years has not come to pass, and the field is actually growing, with practitioner numbers expected to rise by 26% or more over the next three decades. However, his prescience about AI's role is evident: radiology is now the medical specialty with the most AI-enabled devices, with about three-quarters of the 1,400 FDA-cleared AI devices focused on it.

The core challenge is not whether AI can outperform humans on average—it can on some tasks—but how to integrate machine precision with human judgment. Radiologists must evaluate each AI decision, but neural networks are 'black box' systems that don't reveal their reasoning, making it harder to know when to overrule them. The article highlights the risks of automation bias (over-trusting AI) and automation complacency (accepting false negatives), and notes that even experienced radiologists can see drops in accuracy when influenced by incorrect AI predictions.

Experts like Nina Kottler suggest monitoring radiologist-AI agreement rates and providing training so physicians understand when AI might be wrong. Yet a 2026 AMA survey shows training is lacking. As Curtis Langlotz puts it, radiologists who use AI will replace those who don't—but the optimal human-machine team is still a work in progress.