Back to blog

Why scan-reading artificial intelligence is bad news for radiologists

See blog

Readers' comments

Reader comments are listed below. Comments are currently closed and new comments are no longer being accepted.



A very good article. The enormous number of CT images is itself part of the wider problem of medical productivity: expensive machines produce scads of data but often fail to improve outcomes. Performing even more scans is not desirable; CT scans, especially, expose patients to non-trivial amounts of ionizing radiation, and more images often means more false positives. They're also even theoretically useful only if they enable more effective treatment, which often isn't the case.
Anyway, it's not just radiologists who are at risk. AIs will be significantly better at most of the things physicians do, and the case for them gets more compelling all the time: clinical judgement isn't really particularly good, in general; physician wages have grown dramatically in recent decades; and medicine in many countries is notably corrupt. In some cases, the "AI" need not even be particularly fancy. Something that applies good decision rules consistently will usually outperform a human doctor.
The broader threat to physicians, really, is a rationalization of the whole practice of medicine. Major reforms have historically been rare. Changing how jobs work, how care is delivered, and how people are trained is very hard; professional organizations resist, and the political urgency is rarely there. But if it can be shown that AIs will save lives while also slashing health care budgets -- which are spiraling, especially in ageing countries -- then the political will to change the system will materialize quite suddenly. And replacing people with machines is actually quite a lot easier than changing their behaviour.


AI should very easily replace hospital administrators, who are redundant anyway and do not display any evidence of intelligence. The American way requiring a bloated bureaucracy of business majors to run anything is delusional. As a physician working in a hospital I have to answer to 12 agencies employing hundreds of people who are not even able to understand patients, doctors, medical disorders or medicine. You do not learn any of this in business school.

Ed Zimmer

And here's an article - - from the current IEEE-USA Insight in which John Havens, the Executive Director of The IEEE Global Initiative on Ethics of Autonomous and Intelligent Systems takes a balanced look at what he sees as the social/economic implications of this evolving technology. We may finally be starting to see a fruitful conversation on this issue between the economics & engineering camps.


This analysis is too abstract and is not based on real knowledge of the forces shaping salaries in Health Care. The "radiologists" in the article are abstract variables. In real life, radiologists have many lucrative subspecialties (procedures, for example). Also, demand leads to increased salaries only up to a point, and that point comes very quickly. In many hospitals, staff shortages just lead to the closing of units, departments, and services. We just had another "Cabinet Meeting" in our department this morning, where huge staff shortages were again announced for the 10th year, and it only leads to curtailing of services. OTOH, availability of staff reduces salaries only up to a point, so the correlation between salaries and staff availability/shortages is weak. There are other, very complicated factors involved, like the frustration with time wasted on Electronic Medical Records, the likelihood of lawsuits, etc., etc.


Would it not be true that anyone loosing a job to AI has no one but themselves to blame for not doing the obvious self-upgrading homework ?
Yes? ~ No?


The author presents no evidence for the claim that AI can eventually replace all the functions now performed by a human doctor. Why would that necessarily be the case? AI's 'reasoning' power is, no doubt, growing and accelerating. By the same token, we do not know what are the inherent limitations of this unprecedented advance, and it would be naive to think it is unlimited, as the author suggests.

Ed Zimmer

Thank you, thank you, R.A. You're the first economist (or economics journalist) to truly capture the threat from digital technology! Now that you've accepted the likelihood (or even the possibility) of a widespread shrinking job market, what are your potential solutions (to avoid starving citizens and/or revolution) and how do we get other economics-knowledgeable people on board themselves thinking about such solutions?


Gee, who will take the blame when things go wrong? No human to sue. In reality, AI can assist in the final product that radiologists produce but is not in and of itself an end to end solution.

What you say may be true for the next 5 yrs... maybe 8.
In the next 10 yrs, definitely not. Healthcare costs in US are out of control and disproportionate compared to rest of the world. Congress cant control it, but technology will. Just like taxi unions are now a sliver of their former selves.

Ed Zimmer in reply to Allan Edie

It does not need to turn out badly - what's needed is for the macroeconomists to face up to the potential threats posed by the technology and shed many of their last-century beliefs, eg, that economic growth is fundamentally necessary, that currency has more than fiat value (and so requires the issuance of debt), that people must be PROVIDED work for society to function (because people can't find/develop productive work on their own), etc. Once they shed those blinders, an economic system can evolve (and be explained to politicians) that promises a much more utopian society than any we're ever seen (or even dreamed of).

Allan Edie

"Count me among those who reckon this approach is a bit too dismissive of the threat."

Me too. There are very, very few professions or jobs left that are not threatened by AI and other forms of recent technological change. And what makes this time different is that most jobs, in every field of enterprise, are all threatened simultaneously. I suspect that this is not going to turn out well.

CT Bushong

"The latter places an imperceptibly small amount of downward pressure on radiologists' wages. The latter makes all of them—every single one working in every clinic and hospital everywhere—unnecessary, all at a stroke."

I think you accidentally dropped a "former".


"The latter places an imperceptibly small amount of downward pressure on radiologists' wages". ... The former...?

There's a lot of money in radiology- I wonder how much political pressure will be exerted in trying to delay the introduction of AI into the field...?


AI- They have been trying to teach computer assisted detection for years for mammograms. Alas, doesn't work too well. These are programmed by computer people using radiologists to help but they can not replace gut intution as well as excellent visual acuity and there is some subtle aspects of a patients history that the AI won't know.

Better to use AI where it is productive - replace lawyers and politicians, then the country would progress forward at an alarming rate.


Up to now every technological change brought about an increase in the productivity and wage growth. It is not clear why in this case it should be different. Of course, it can happen that many radiologists will need to adjust or even search for a new job. But this is no reason to stop development and usage of capable AI.


No worry. AI won't replace radiologists. AI will help radiologists to better interpret x-rays.