In Seth Godin’s podcast Akimbo, he discusses the concept of being supple, or being resilient in a changing world. In the subsequent episode, he answers listener questions about the previous week. A filmmaker asks what he should do, given that much of the creation of pictures and videos has shifted away from experienced professionals to amateurs armed with newer technology. He asks “What’s next?” 

Seth responds that what’s next is what’s always next: we redefine what “better” means. There is always a path toward improvement, but the way improvement manifests changes over time. To illustrate this, Seth even briefly mentions radiologists and AI.

As I’ve written before, if we ask whether or not AI will replace radiologists, we are asking the wrong question. “What does better mean?” is a perfectly good alternative question. Despite all the claims that a computer can read a chest x-ray, or a CT for pulmonary nodules, or a bone age exam “better” than a radiologist, we still don’t really know what that “better” looks like. We don’t know what “better” means for us in anything but broad terms.

Let’s presume that we have an algorithm that can find pulmonary nodules on a CT more reliably than a human – it’s not hard to imagine, and may already the case. But that’s just one tiny part of the overall imaging interpretation process and it creates an incomplete vision of the future. How does this fit into the entirety of performing and interpreting an exam? What does “better” look like when a human and computer interact? Does the computer suggest and the human confirms? Perhaps the algorithm should follow the radiologist after the report and point out possible misses? What about findings that are too subtle for the human to independently verify? Now add another algorithm, and another, and another. How we do coordinate all this AI work and integrate it with the radiologist in a way that makes the combination better than either one alone? AI may promise a more precise, reliable, or accurate process, but it’s not enough to just give vague descriptions of computers taking over, or that they will “enhance but not replace” people. The hard work is in the details. The specific dance between the human and the computer remains unchoreographed.

The important realization is that this situation is nothing new. Radiologists have been adapting to and incorporating new technology for a long time. Digital imaging, PACS, advanced imaging techniques, speech recognition…new tools become available, and the better radiologist finds ways to incorporate them into practice in a way that improves care and increases value to the patient, the referring physician, and the healthcare system. The soon-to-be-extinct radiologist complains about the change, digs in their heels, and hopes that it all goes away.

AI feels more challenging because it comes closer to the core of what some radiologists see as their added value – finding and characterizing abnormalities. But limiting a radiologist to the task of finding and describing is a narrow view. If you, as a radiologist, define yourself as your ability to find small white dots on a black background, then yes, the ability for a computer to do it better represents an existential threat. But if see your job as serving the patient and adding value to the healthcare process, then nothing has changed. You continue your work to incorporate what is new, create a better product, and in so doing redefine what “better” means.