Our family had some nice, quiet snow days over the weekend here in Dallas, so I finally finished this historically interesting if morbid Lego mosaic of the first CT scan from 1971 (right frontal GBM):
The Stoics thought everyone should have a memento mori, but the typical skull is so staid.
Most current articles about AI and radiology work don’t help us much with regard to the questions people are constantly asking about predicting the future. Publish or perish is fine, but I’ll admit I also don’t really care how many ways we can say these early tools are sorta maybe helpful sometimes for some people but are also super brittle and unreliable and also the most important thing is that they don’t make things worse, particularly due to bad integration with a legacy workflows. How to incorporate something only borderline useful into a workflow you probably don’t control just isn’t very interesting, and people keep dropping predictions about what things will look like years (or even decades!) in the future based on what we’ve seen so far.
Extrapolating linear growth when we have exponential compute isn’t helpful. But imagining a post-work techno-utopia vs dystopia is also not actionable for anyone. We can skip the hand-wringing and just agree that everything currently available isn’t good enough to change the world but that maybe one day it will be.
We live in a complicated world where many of what we perceive as challenging problems are computationally straightforward (e.g. cumbersome long calculations, playing chess) and other things that can become automatic to most humans are computationally intense (as it turns out, driving a car, though it’s getting better!). Corner cases and uncertainty are, in some sense, the whole game when it comes to automation. Medicine—with its treatment algorithms designed for squishy humans with limited narrative skills and inherently incomplete information—is a combination of both.
Lots of medicine is straightforward and subject to guild protectionism, but some of it is genuinely hard. And sometimes the difference between those two situations is only knowable after the fact. If nothing else, ambiguity is hard, and how much uncertainty and risk to accept has never been codified. For example, many current AI tools are designed for a high negative predictive value and are therefore overly sensitive. That’s fine for a tool to help a human not make a mistake, but that still can lead to cry-wolf problems and certainly wouldn’t work for an autonomous machine.
Balancing uncertainty and service expectations is no joke for humans or machines or the combination.
As I’ve said before, I don’t think the reality of impending change means that we can meaningfully predict the contours of that change or should even try to plan for its eventuality in ways that would be destabilizing in the intervening time. What Hinton got wrong is that predictions about replacement are dangerous because, until replacement actually occurs and/or demand shifts, we need to maintain the status quo. There’s no way to know if any field (whether primary care or even surgery) will be augmented or replaced over an effectively infinite timeline, or if the contours of the work will change and if so to what extent and how fast, and therefore what the correct number of people we need long-term such that we could preemptively adjust the pipeline of training to help us land at the optimal amount of people for every kind of job. It’s not going to happen.
For long training pathways like most physician careers, trying to pretend we can is insanity. Central planning in that fashion almost never works because there are far too many unknowns. What we need to do is continue training people for the world we live in with a plan, desire, and ability to adjust our workforce over time as the world changes. We need less planning and more flexibility. Less preciousness about how we work and more orientation on the desired outcomes.
Until the exact day a computer replaces a human, you need that human. Trying to predict a societal need years in the future is a fool’s errand. We haven’t been able to fill the halls of medicine “correctly” over the past century; why would we be better at doing it now?
Let’s just posit a very narrow scenario to see how futile the situation is. Let’s magically say AI for screening mammography becomes incorporated into major vendors at some point in the next X years, allowing for automated reports, even for callbacks. A computer report and a human report would be indistinguishable, especially since BI-RADS facilitates complete automation of reporting language and organization. (I’m also picking breast imaging for this thought experiment because it’s currently very much in demand, tends to be narrowly subspecialized, and has relatively unique current job offerings.)
Let’s also say that the algorithm performance and regulations are such that, for now, those exams are still read by a human being as well—at least in the short term. Guess what the incremental efficiency gain will be? How many screeners can someone read in that setting? The answer, of course, is that it varies.
Automation bias is going to hit fast, so it’s not clear if someone will be able to read 10-20% more, twice as many, three times as many, or some other multiple. Whatever the number is, the first question becomes: how does that affect breast imaging staffing? Does it change the premium that breast imagers can currently command because of the high volume of high-paying screening exams, particularly those with 3D tomosynthesis? How many nefarious rubber-stampers do we need to change the competitive landscape? Think of how fast turnaround times can be! This of course will happen on the backdrop of continually increasing imaging volumes and technological development.
Okay, then what happens to mammography reimbursement? The RVUs will inevitably change if the effort to do the work changes. How fast will that happen, and how drastic will the change be? The RVU system is a zero-sum game, and undoubtedly, there will be a lot of fighting about any changes to radiology reimbursement that occur thanks to AI tools—but they will occur. Tomosythesis is well-paid and already in crosshairs.
How would the subspecialty handle a shifting balance of non-procedural and procedural work, or remote vs on-site work?
Then, how would the job market respond? Would we see similar jobs with less pay? Or would we see a chilling of the growing lifestyle niche of 4-day workweeks with no evenings or weekends? Would we just be able to better staff growing volumes? When I finished training less than a decade ago, it was common for breast imagers to take general call. It would seem that’s pretty rare now. Medicine, like politics, is local. Of course, breast imaging isn’t operating in a vacuum; this is all taking place within an even larger lifestyle teleradiology trend. But variables don’t play nice and isolate themselves: everything is always changing all the time.
What is the timescale for those changes, and how piecemeal would they be? What are the second and third-order downstream consequences of even a single shift in radiology productivity? What about fully automated DEXA? What about prepopulated ultrasound measurements? Natural language draft reports for other modalities. We don’t even know. But the answer isn’t simply that we need fewer breast imagers when there are currently not enough, just because one day we might.
Medico-legally, how will we as a society handle mistakes made by machines? Or, with a human checker, how will we weigh a change made by a human versus a mistake made by a human coupled with a machine?
How will we feel about a mistake where the AI misses something and the human doesn’t catch it–or doesn’t fix it–versus a situation where the AI calls something and the human pulls it back, even when the AI is right? How will we handle type 1 versus type 2 errors? Will companies self-insure for autonomous work as the dominant strategy, or will we see protectionist legislation about a variety of key human tasks?
We don’t know, long-term, whether there are edge cases where humans are more or less important. But we also don’t know, even short-term, where the contours of those skills will reside. Maybe we’re entering a world where the ED would be a better place if a computer did most of the interviewing and initial workup for stable patients.
We are amidst a jagged frontier where AI is good at some things and bad at others, and we know there is a lag between capacity, implementation, legislation, and regulation. Planning is always a challenge; useful prediction is basically impossible.
So, instead of predicting exactly what will happen, we need to acknowledge the broad potential possibilities that could occur and plan a system to deal with them when/if they do occur, as opposed to trying to nail the blind landing.
I don’t want to call out any behavior as irrational, but we should acknowledge upfront the human tendency to just do whatever we want to do and then create post hoc rationales + explanations + narratives to justify and explain our behavior after the fact. Much of this is unconscious and may be a clever trick of our mushy brain psychology, but that doesn’t make it any less real. Try to be honest with yourself and see why you feel the way you do about just about anything and if the reasons you give yourself actually mesh with reality. Write it down, or you’ll cheat (it’s our nature).
It’s like a medical student saying they like “technology” and “problem solving” and “people coming to the reading room for answers” for why they picked radiology when it was really that it seemed relaxing and quiet with good vibes and plenty of coffee and that people at least respected the radiologists’ intelligence and oh yeah they heard they were paid well.
Over the past decade or so, psychiatry has gone from very uncompetitive to pretty competitive. Is it because mental health is important and less stigmatized? Surely that’s part of it. But maybe also because cash-based private practice in a high-demand field (potentially even done via telehealth) sounds like a nice quality of life and a rare way to practice medicine on your own terms. We always seek single explanations thanks to our bias for narrative, but that doesn’t mean that the compelling storytelling is any more reflective of reality.
The examples of failed central planning and the reliance on narrative are legion.
To reiterate my point: No single person or entity has the answer.
Yes, we need all varieties of doctor now, and we need to keep training all of them until we don’t—if that ever occurs.
Will a job like radiology or even primary care pay the same 10 years from now as it does today, or 20 years from now? That will depend on how much value the human provides within the system, how much work they do, how hard or complex that work is, and the supply/demand balance of the available workforce. Volatility and all kinds of non-AI factors already lead to lots of unpredictability.
There may be golden years of excess income for scalable jobs like radiology as these changes first take place. If senior engineers can use AI to do more projects without hiring as many junior coders, it results in rising incomes for those on top and a rug-pull for those entering the workforce.
Both medicine and society need ongoing, never-ending conversations in order to work on creating adaptable systems. Instead of placing bets, the acronym organizations need to set stakeholders up for success across the full spectrum from an AI-nothingburger to AI-work-is-optional.
I’m in the new episode 7 of AJR’s Radiology Trailblazers podcast with Dr. Lindsey Negrete and Dr. Amy Maduram. Please do not think for a second that I accept my inclusion in any “trailblazing” premise(!), but we had a nice discussion about the writing process in and out of radiology. Whenever I do a podcast, I am always able to create at least one new verbal tic. This time it was “in reality”—please forgive me.
A paraphrased reader question:
I want to be a daytime tele neurorad. Why do so many of the listings seem to be for body? Are there too many neurorads?
I don’t think it’s really that there are too many neuroradiologists per se. It’s that the true need across the widest variety of practice types is general radiology.
If body imagers only read specialized body MRI, they wouldn’t be filling the holes that we have as a field. The greatest need is for plain films, ultrasound, and generic CT.
That is generally part of the job for body imagers, but many neurorads (and other subspecialties as well) really want to read more within their subspecialty, leaving a pile of general radiology for which groups are desperate enough to bring on remote readers.
In reality, there is nothing about a thyroid ultrasound that should make it a “body” examination and not a neuro one, given that the thyroid gland is in the neck. But this is the way everyone practices.
As a result, if a group hires more neurorads to support general needs without needing a full neuro FTE, that dilutes the available neuro work and requires the neurorads to start reading more general radiology. They, generally, don’t want to do that. And in practices where neurorads are 100% or nearly fully subspecialized, they don’t have the ability to overflow into another subspecialty easily.
Staffing in body typically doesn’t really work that way: smaller changes to their casemix aren’t going to result in a super different job or fundamentally change the spectrum of cases they interpret. No new skills needed.
And therefore, body is the greatest need. Not because there are an infinite number of liver MRIs to read (though there are a lot), but rather because we need people who are willing to do basic radiology work. And many body radiologists are expected to do that.
At least that’s my impression of the current state.
There is daytime tele neuro work out there, but I agree it’s mostly if you’re also willing to also read general too.
The first few months of radiology residency can be pretty bewildering. It’s rare in adult life to start from scratch quite as much as you do the summer you transition from internship. Even if you did some studying during internship or you were fortunate enough to have electives in radiology during your intern year, there is a steep learning curve in July, and no one’s foundation in radiology is typically all that strong in terms of actual working ability.
Even though you’ve been looking at things and talking your whole life, it isn’t the same thing as looking at medical imaging and dictating reports. And so the first few months are dedicated to learning the core skills of a radiology foundation: getting comfortable, learning the hospital and the EMR, understanding the radiologist’s roles in the clinical workflow, and putting up some serious imaging reps.
Junior residents will typically take some variety of call after a year, and—even if you don’t have independent call—I know that’s a big step. As your spirit attending, I do want you to continue seeing as many cases as you can and studying outside of work to make sure that you are rounding out your radiology knowledge to build on the things you’ve seen in real life and fill in the gaps for the things that you haven’t. The List Gods do not provide you with everything you need, and the Venn diagram circles of radiology experience and radiology testable knowledge from books/questions/videos/etc do not have perfect overlap.
The spring semester of the R1 year is a great chance to work on the deliberate practice of radiology. There is an iterative loop of interpreting radiology cases that we perform over and over again, and I find that most trainees struggle with certain parts more than others.
I view the five-point interpretative loop as follows:
1. Attend
Bring your complete focus to every new case, because every case is a new patient who deserves your full attention and your best work.
2. Observe
Review the images to make the observations/findings. This involves scrutinizing everything, of course, but also involves practicing avoiding certain pitfalls, like missing findings at the top and bottom images or at the edge of the field of view, falling prey to satisfaction of search, and forgetting to do a targeted review of specific imaging areas to exclude relevant pathologies, second order findings, and seek pertinent negatives (the true search pattern).
3. Decide
Make the decision/diagnosis. This includes deciding if an observation is real or artifactual as well as settling on the differential or significance of a finding. Practice deciding to the best of your ability without perseverating on findings by endlessly scrolling back and forth or agonizing. Practice making a targeted review on the internet to help categorize a finding without getting too deep down the rabbit hole, and be willing to phone a friend or an attending to get an outsider’s more experienced perspective when you need it. Cultivating independence is critical, yes, but we can also get more reps and do more work if we don’t get too bogged down. Try to do it alone when you can, but don’t get trapped in your own head.
4. Describe
Dictate the report with the goal of an organized and concise but thorough findings section as well as a clear, actionable impression. I personally subscribe to the “the more you write, the less they read” perspective, but everything is truly a balance, and the goal is not for you to emulate any specific style but rather to figure out how you want to practice and how to do your best work.
5. Polish
Proofread and hone that report until it is as good as you can make it. I argue that sloppiness never stops at transcription errors or other seemingly less important things; it percolates and permeates. It’s important to practice doing your best work, and that may mean adjusting how you work so that you make great reports routinely.
Within the spectrum of giving each case your full attention, you need to set yourself up for success, and that means not cutting corners. That includes reviewing the history and utilizing relevant priors. Yes, it takes some more time upfront, but it also guides your search pattern and informs your decision-making. It is critical to the practice of radiology. And you will make so many fewer mistakes if you get into the practice of doing this consistently.
You would probably severely underestimate how many times I have seen people be wrong in their conclusion or agonize over a decision that didn’t even have to be made because they failed to look in the chart for relevant history, to look at prior scans and prior reports in your system, or even to see if there were prior radiology reports hiding in Epic available in Care Everywhere. There are countless age-indeterminate or nonspecific findings—including those that generate useless follow-up imaging and unnecessary admissions—that, with some scrutiny, are chronic or stable for years. You’re never going to bat a thousand on this kind of stuff, but you might as well try.
Additionally, those prior reports are often your greatest teacher. So please do not waste these opportunities.
Additionally, I strongly recommend that when reading follow-up scans, you take a moment to look at the presentation scan. In many cases, that initial scan when that stroke is less obvious or the pre-operative scan before a lesion is resected gives you the most valuable visual exposure instead of just monitoring a mostly stable ball of blood or resection cavity. Don’t rob yourself of the opportunity to see the full spectrum of radiology/pathology, and try to find things and see how they look and really evolve over time. These are an important part of your learning, so don’t waste them.
It is very reasonable to review your case first before looking at the priors and their reports so as to give yourself the opportunity to approach each case de novo. This is admirable if time-consuming, but don’t take the desire to bring fresh eyes to each experience as an excuse to cut corners and not review the priors.
It’s your job, with the help of your attendings, to figure out:
It’s worth pointing out that the things you don’t like doing are often the things that you struggle with. There is a saying that practice makes permanent, so it’s important not to just try to get through the day but to really focus on these steps to build your skills and work on your tired moves:
If you don’t do things the thoughtful way on a regular day with no pressure, then you’re never going to do them well when you’re tired and under pressure. So we need to work on doing things the “right” (read: consistently improving) way every time, over and over again during the regular workday, to build up the muscles of doing radiology a better way.
And, those things we don’t enjoy aren’t as fixed as you might think if you’re able to cultivate a craftsman’s mentality.
There may be no universal right way to practice radiology, but there are definitely some wrong ones.
I finally got a standing desk, which means I was finally able to try a foldable treadmill under it. I love it.
In early 2024, I started using an under-desk elliptical, and it worked surprisingly well. One downside is that since it is only self-paced, it’s easy to stop.
This fall, I got the WALKINGPAD P1, which is a foldable treadmill I stow behind my door when not in use. It’s reasonably quiet and has served me well for a couple of months so far. If I were to buy one right now, I actually really meant to get the R2, which has a handle that allows for higher speeds and, more importantly to me, also means it’s easier to move and can be stably stored upright. If folding and stowage weren’t important to me, I’ve also heard good things about the Pacerocker, which has a 9-level incline feature to make things harder without increasing your walking speed (though I have no idea if having an incline would make it too hard to concentrate?). I may still upgrade to the R2 at some point.
Prior to purchasing, I had wondered if I’d be able to work while walking. I figured if I couldn’t, I could at least use it for meetings and maybe lighter/passive tasks like email.
The answer is I can definitely work while walking, at least at modest speeds, and I can easily clock 15-20k steps walking intermittently during a workday.
Admittedly, super fine motor control with the mouse is a bit harder, especially at higher walking speeds. It can, for example, be hard to measure lesions if you’re bouncing, so I find myself walking at a slower pace than my natural cadence. But while mouse control is a little harder, typing on the keyboard is surprisingly natural. It also definitely remains much easier to scroll effortlessly with my Contour Shuttle, which uses a finger, than power scrolling with the mouse.
Additionally, while the treadmills are definitely movable, it doesn’t mean they’re the kind of thing that you really want to have to wheel all the way to and from their corner multiple times per day. I find that when I would like to sit, I tend to just fold the treadmill back under the desk into the leg nook and use it as a bit of a footrest or put my feet along its sides. Having a standing desk means that you can adjust the height a bit as needed to keep things comfortable to accommodate those few extra inches. The main downside of sitting in front of the folded treadmill is that the wheels on your chair can easily bump up against it. It’s best to sit slightly back, so having a pull-out keyboard tray (which I do) really helps with positioning. This is the main reason I wish I’d bought the model with the handle, which would make it easier to move around and probably obviate the need to ever do this. If you try it and don’t like sitting at it folded, the answer is probably just to use it in one session per day and move it away when you’re done.
I think standing desks on their own are somewhat overrated. I like standing at work sometimes for variety, but it’s also sorta meh? Unless you like to do a little dancing in place, the data suggest that standing isn’t really magically better than sitting, and there is some data that they’re not great for your lower extremity veins if you clock too many hours. Being sedentary has more to do with not moving than the actual position you’re staying still in.
The opportunity to try work-walking was my main motivation for wanting a standing desk at home in the first place. I’m not a teleradiologist, so I’m not cranking in my pajamas during the workweek as a full-time job, but as a component of my hybrid workweek, the treadmill desk has been awesome.
A helpful note from a reader:
It seems the R2’s more robust build quality would be better allow you to stand on the rails intermittently as suggested. I’ve tried this as a way to take a brief measurement/stretch break here and there to stretch my back or do some fine mouse control, but I will admit that it’s a little bit hard on my P1 with its plastic rounded edges. It’s another reason why I’m thinking about upgrading, being able to hop on and off without needing to restart the treadmill would be awesome.
A paraphrased reader question:
I am still in the middle of residency but entertaining the idea of doing 100% remote work fresh out of training. I want to live in a rural area out west, so I am thinking of swing shift jobs in CST/EST. Do you have any advice for what to look for in my job search?
In general, compensation and PTO for many smaller or lower-density markets can be shockingly good, so it’s always worth looking at any local options—you may find the balance appealing even if the practice is small or more generalist than your training focus.
For remote work, central/east-coast tele is very reasonable. Generally, most daytime-remote jobs skew toward general and ER work, but as more groups struggle to recruit locally, more are entering the tele market for subspecialty needs as well (if you like general or body imaging, for example, we’re hiring).
If you’re open to swing shifts, one important question is how the exact hours fit into your life. Some shifts would be seamless with the time change; some remain mildly annoying. One upside is that many swing roles have 7/7 or 7/14 scheduling, which opens doors for internal moonlighting, secondary gigs, or a serious hobby. When available, a mix of call styler and non-call work also helps with sustainability and variety. For example, having part of a shift devoted to outpatient and part to inpatient/ER can keep things from becoming a grind (which is what our group did for our swing positions, which have been filled mostly by our own partners).
There’s no real shortcut to finding what you want—you’ll likely need to browse swing listings and see what matches your preferences. Most of these positions are not partnership-track, but occasionally you’ll find one that is. Depending on your plans, a conventional employed position or a contractor arrangement may fit better anyway. Insofar as job security goes, employed swing tele jobs are probably substantially safer than daytime tele.
You can find fully remote daytime work if you’re patient and willing to sift through postings, but those jobs are less common, and you have to accept they often come with less long-term security. The number of groups hiring for daytime tele has been increasing as the on-site recruitment pool dries up nationwide, but it doesn’t mean that those jobs aren’t the first to get cut if the workforce becomes more productive or that those groups are healthy enough to survive until then.
If you want to leverage time zones and earn more for your hours, swing tele is absolutely an efficient solution. The practical constraint for someone in residency is timing: groups usually don’t want to hold a very specific swing opening for someone years away from graduation. Many are also quite reasonably hesitant to place a brand-new graduate on a high-volume, high-stress, high-acuity list. Most good swing slots tend to be filled within a few months. Realistically, you might as well wait until late in R4 or early in fellowship.
Bottom line: explore local options, look closely at swing roles for flexibility and compensation, and expect the real job search to happen closer to the end of training when groups are ready to hire for these specific needs.
Medicine’s narrowest job board, Independent Radiology—the home of radiology private practice—is sitting pretty again at 150 groups. A few groups sadly closed for good, and some—fantastically—are fully hired up, but it looks like we’ll close the year back at that nice round number. If you’re looking for an opportunity, perusing the listings is good holiday homework.
Time for my semiannual update and bump of this post: like every other practice in the country, my group is also hiring!
American Radiology Associates is a 100%-independent physician-owned radiology practice in Dallas-Forth Worth (of which I am a partner/shareholder). We’re privademic: we have part of the practice that works with the Baylor Dallas radiology residency, and we have part of the practice that does not. I’m the program director, and I still enjoy a nice mix. Just imagine the fun no-BS parts of teaching and variety without the rigid hierarchy and inflexibility.
We’re hiring for body, general, chest/cardiovascular, MSK, neuro, PET, and IR/DR. All of us in DFW work a hybrid on-site/remote schedule and the option of a 4-day workweek. We are also hiring teleradiologists for body/general imaging.
While our partners are generally in the Dallas/Fort Worth metroplex, we are also offering a 100%-remote partnership-eligible body swing position.
The swing shift is 2 pm-10 pm Central Time, weekdays (M-F) alternating every other week + 13 weekends of call (yes that means mostly weekdays and not 7/7 (which is 26 weekends), and never any deep nights or super weird circadian-destroying hours). The shifts are all body/general mostly ED/inpatient (neuro optional but can be included). Other schedule configurations could be considered on an employee or 1099 basis.
So if you’re in the market, come work with me and check out our great team in Dallas. If you’re interested, send your CV to careers@americanrad.com and CC me at ben.white@americanrad.com.
A few further thoughts about NVIDIA CEO Jensen Huang’s farcical description of the impact AI has “already” had in completely revamping the field of radiology.
Huang presents radiology as the “evidence” of what the near-future impact of AI on the workforce will be. I’ll include the quote again in this post for completeness:
One thing that I will say, give you some evidence, is that, and I was just telling Elon about this earlier, radiology, for example, has largely been converted to AI-driven radiology. And there’s some really great companies doing that. And the surprising thing is the prediction that all radiologists would be the first jobs to go was exactly the opposite. The trend shows that there are more radiologists being hired now as a result of AI.
And the reason for that, if you take a step back, it’s because the goal of a radiologist is not to study the images. The goal of a radiologist is to diagnose a disease. Now the studying of the images became so productive they could study more images, study more modalities, spend more time with the patients, and as a result, they were actually accepting more patients. We’re doing more radiology all around the world, we’re doing a better job with diagnosing disease.
Again, none of that is even merely an exaggeration. It’s just wholly untrue as to the current use of AI in radiology, let alone the impact far enough in the past to have percolated through and already changed how the whole field practices.
This is a man personally worth almost $200 billion in charge of a company with a market cap over $4 trillion. One imagines he has access to reality if desired.
I’ve seen the clip shared countless places by credulous people who don’t know any better. This is not to say these things won’t happen, but the use of the past tense is a real problem. I think it’s worth putting the fantasy in context.
Jevons Paradox is the observation that technological advancements that increase resource efficiency can counterintuitively lead to an overall increase in resource consumption. Jevons’ original formulation concerned improving coal technology leading to paradoxically increased demand/consumption of coal. As in, when it’s cheaper, you can buy more and do more.
Huang is parroting a recently commonly invoked human radiology analogy: AI makes radiologists more efficient in interpreting scans, more scans somehow get done, and voila demand magically increases.
The problem is this doesn’t reflect reality, at least not currently. Imaging volumes have been increasing steadily for decades. Scan acquisition time has nothing to do with scan interpretation time (the former actually ironically is benefiting from machine acceleration in some situations). Interpreting efficiency has barely moved, if at all, and turnaround times are actually lengthening amidst unmanageable volumes. Increased demand for radiologists is just the secular shortage of qualified radiologists struggling to keep up with organically increasing volumes. AI has nothing to do with it.
To further compound how much we are not in Jevons’ land yet: reimbursement for radiologist professional services is stable to falling from payors (again, not related to efficiency), but radiologists actually cost the system more recently as they demand stipends from hospital systems to provide continued access to services. More simple supply/demand at work. Those scans, even the “more efficient” ones, do not cost less for patients or payors yet. More efficient MRIs and radiologists are both paid the same on a per-study basis thanks to the prevailing fee-for-service model in US healthcare.
There are some AI tools for radiologists with reasonable market penetration: variations on list triage (enabling potentially positive scans to jump the queue for interpretation) and generative draft impressions derived from the findings section of the report. Meaningful computer vision is frankly not in particularly widespread use despite what is parroted by tech companies using the news as free marketing, and so far limited to narrow pathologies like brain bleeds, blood clots, and fracture detection. Any purported efficiency benefits have been unconvincing. A mediocre tool allows careless people to move faster but gives careful radiologists just another thing to review. So far, the state of the art has been mostly a wash.
Again, this isn’t one of those “AI could never do what my amazing organic brain can do!”; it’s an “AI really certainly hasn’t yet in real practice, and nothing on the market so far has really moved the needle.”
Greater efficiency could trickle down to depress prices in the future and/or eventually lead to increased demand for imaging, but it hasn’t happened yet, and it’s unclear if it did that it would lead to increased demand for the human component. Healthcare is not really a free market. No one in healthcare is the equivalent of coal.
Is it possible that one day AI will lead to automated scan acquisition and instantaneous scan interpretation a la Star Trek, and we will suddenly not just want but need All Scans All The Time, and oodles of AI-controlled factories will build more and more machines to scan more and more people because interpretive costs are down, and physical throughput is the real bottleneck? Sure. We can just put a conveyor belt at the entrance to the ER!
Is it also possible that—instead of a Jevons process—professional fees could crash as radiologists become box-checking liability-operators, but that as the field contracts, the smaller number of remaining radiologists will enjoy persistent high wages in the form of obscolence rents? Also, sure.
Is any seismic AI change, if it occurs, going to be limited to radiology? I doubt it.
Ultimately, humans are not quite a natural resource. The tortured metaphor may or may not hold in the future, but we should all be able to agree that it sure hasn’t happened in the past.
So why is the world—and everyone contributing to the AI bubble—so obsessed with Jevons’ Paradox?
Because it’s comforting.
Instead of worrying that the economic value of most humans is going to trend toward zero and that our entire society will have to contend with massive disruption, it’s far easier to believe that AI efficiency will unlock ultimate productivity. That somehow, this machine intelligence revolution will set us free to do our best, most magical, most human work.
A narrative where increased efficiency leads to increased demand soothes the uncertain soul. In that version of the story, the highest-skilled humans won’t be made obsolete—they’ll be unleashed.
I don’t know if that logic will hold in real life over the coming years or not. What I do know is that invoking it right now is a marketing fantasy.
To reiterate: I don’t know what will happen to radiology or to any other field. The future is uncertain, and I certainly don’t have a crystal ball. We have plenty of bullshit jobs already, even without the help of AI.
What I do know is that all attempts to pretend radiology has already seen the fruits of AI and absorbed them—that AI is responsible for the current surge in demand for radiologist services—are lies.
But we can appreciate why this specific variation of the lie is being told:
In 2016, AI Godfather Geoffrey Hinton famously said:
People should stop training radiologists now—it’s just completely obvious within 5 years deep learning is going to do better than radiologists. It might be 10 years, but we’ve got plenty of radiologists already.
So the fact that we’re still here and in more demand than ever is supposed to be comforting to other humans.
Don’t fear AI! Even the radiologists are thriving!
I’m sure that in advance of an earnings call, a narrative where AI unlocks human potential is so much more compelling than one with a zero-sum game where the short-term economic value goes to tech companies but the long-term impact is to potentially destroy human enterprises and sink the economic value of previously high-training, high-skill, once-human tasks.
If we think further ahead, no one is going to pay for computer work the same as we did for comparable human work for a sustained period. The RVU system in healthcare already attempts to account for effort, liability, etc. We don’t get to hold even that one system’s current balance fixed while the world changes. Causes yield effects, and consequences themselves have consequences.
Anyone who imagines a perpetual money-printing machine generating revenues to match the humans they once replaced is naive, so maximum AI utopianism (and the value tied up in these companies) doesn’t envision that world. The devastating disruption fears may or may not be valid, but Huang and others in the tech world clearly feel compelled to address them.
It seems to me that Huang is trying to wave away generalized replacement fears by pretending that radiology is the canary in the coal mine and we’re still here, therefore rainbows and unicorns.
Maybe that “doomer” path to darkness is wrong, but that doesn’t mean radiology is the example to light the way.