Addressing the Future Impact of AI in Radiology: Emphasizing Planning Over Panic

Suppose you are shooting a game of pool. You are doing exceptionally well. Your opponent has gotten just about nothing done and you have one last ball to sink. It is sitting right in front of the pocket, and the cue ball is perfectly positioned to send it home.

You could certainly screw it up if you tried. Of course, you don’t but you also don’t need to invest all your powers in that final shot. You can hit the ball with 1 or even 5 percent deviation from what would have been a perfect angle, and it still gives you an obliging kerplunk.

Now suppose your shot spans the length of the table. It is no longer such a sure thing. That added distance of a few feet makes even a couple degrees of angulation more relevant. Change the venue to a golf course with hundreds of yards at stake, and they matter even more. Try to plan a mission to Mars and even a fraction of a degree’s error can be catastrophic.

In other words, the potential impact of an action increases exponentially with the amount of space (or time) that follows it. It is a lever of causation.

It strikes me as unfortunately ironic that an appreciation of this comes with experience and aging. The more you understand its power, the older you have gotten, and the less time you have left ahead of you for the leverage to work its wonders.

Physicians, especially those with longer post-grad stints like radiologists, tend to have a greater understanding of this. They recognize the long-term value of spending their youthful years in school and training, even as they envy non-medical friends who are enjoying their 20s with kinder schedules and healthier paychecks for the time being.

Our respect for future consequences might endure a little too strongly. In social media, I have noticed that the doom-n-gloom crowd (everything is going to be ruined by AI, governmental mismanagement of health care, corporate medicine, etc.) is disproportionately populated by elder rads. Those with fewer years ahead of them excessively worry about what those years might bring.

If I am planning on retiring tomorrow, my career isn’t going to be impacted by anything that happens. My pool ball is right in front of the pocket and I am sinking it no matter what. I might lament what will befall younger rads or patients needing radiological services, but my work has been completed, and I have no reason to worry about it.

The more of a career I have left in front of me, the more I should think about what is coming down the pike and try to plan for it. Note that I said “plan,” not “panic” (although if you get your dopamine fix by posting on social media that the sky is falling, you do you).

Here is an exercise in contrasts. A non-medical friend of mine has held a wordsmithing job in the tech field for the long haul. AI is knocking increasingly loudly at her door. She used to have four coworkers, but now she is the only remaining protoplasm. Machines produce what the humans used to, and her role has been recast as proofreading and editing their output. As the last one there, she is not waiting to be handed her walking papers. She has already set about making contingencies for herself.

Meanwhile, I don’t know a single radiologist who’s been replaced by AI. I have heard from some who worked with AI companies in the name of theoretically training our machine replacements. Leaving out those who have financial interests in that happening, not a single rad I have heard from thinks the machines are anywhere near ready to replace us.

(As a sidenote, we have noticed most of the “want ads” from AI companies seeking rads to train their products pay a pittance compared to what we earn from regular rad work. The only folks we imagine would take that pittance are rads who can’t cut the diagnostic mustard, or even people claiming to be rads but really aren’t. Either way, this is not exactly the cohort you would imagine capable of training an AI to competently do radiology.)

Still, let us suppose AI is foisted on radiological prime time next year. Do you think we would all be summarily cut loose? Would the health-care system, government watchdogs, and patient population blithely hand the diagnostic keys over to what many consider an unproven technology?

I think it far more likely that things would be gradually phased in, maybe slowly compensating for the increasingly rampant radiologist shortage. As widespread disasters fail to occur, it might gradually be turned up, perhaps conforming to the drop-off curve of rads retiring from the field as they normally would.

Even if it were more abrupt, there would have to be some sort of safety net. Suppose AI does all radiology interpretations as of 2027. We human rads cancel our licenses, stop MOC, etc. Now, something goes horribly wrong with the AI. A well-publicized series of catastrophic misdiagnoses, perhaps, or a terrorist sets off an electromagnetic pulse (EMP) that trashes the computer servers.

Society would want — even demand — that a human radiologist workforce remain ready to take over again. The only way that could happen would be if a healthy number of us were kept in the loop with viable work to maintain our capabilities.

I think that takes care of those of us who are already well into our careers. How about the younger rads with more years of career in front of them. What about prospective future rads who haven’t even applied for residency yet? The leverage of consequences can angle much more drastically for them. How might they future proof themselves, short of avoiding radiology altogether?

A common theme out there in the “who won’t be replaced by AI” chatter is that the more you physically do things, the safer you are at least until robots enter the picture. For rads, I would take that to mean staying sharp with your hands-on skills: angios and other interventions, biopsies, etc.

Another approach is to develop sub-subspecialty expertise in a couple of areas. Assuming future AI is anything like what we have now, its strength will be greatest in superficial, general knowledge stuff. If you have ever tried to do a “deep dive” on some particular tidbit of info, however, you might have noticed that AI gets sketchier and sometimes even makes things up.

So, for instance, when referrers find themselves dissatisfied by an AI that gives them a namby-pamby bundle of words about indeterminate liver lesions, if you have honed your hepato-radiology skills, they will want to keep you around. Maintain an awareness of what the technology isn’t doing well, position yourself as a more reliable alternative, and you will remain a step ahead of everyone who doesn’t.