Signify Premium Insight: Google’s Generative AI and the Problems with Predictions

“We should stop training radiologists now. It’s just completely obvious that within five years, deep learning is going to do better than radiologists.”

This, now infamous, quote from Turing award-winning academic, former Google employee, and AI pioneer Geoffrey Hinton in 2016, should serve as a reminder when assessing a vendor’s promise or the potential of a new technology. And, while we should not judge Hinton too harshly, and he isn’t the first, nor will he be the last to overestimate the pace of a technological change; it is  important to remember that possibility and reality are frequently uneasy bedfellows. Hinton, after all, was not entirely incorrect; there are, after all, real-world studies that have been published showing that for radiological tasks (e.g., triage, detection), in real-world settings, algorithms can match the performance of radiologists. However, the combined value of radiologists with AI algorithms has been shown to outweigh the performance of either one independently. Further, image analysis is only one small piece of a very complex puzzle, most pieces of which will still need to be provided by radiologists for years to come.