Advertisement - Members don't see this ad
Sure, at age 2-3. At age 6-7 it typically isn't.
Number 2 Most Resistant Job To AI
Started by Groove
Get help with your application
Use all the free resources available to you from SDN: articles, guides, expert advising, forums discussions, and school research.
Quantum computing will be used to secure those systems as well. So it will be evil quantum code fighting against good quantum code.
Are there any videos of how this product works in real practice?
As it stands now most of these AI scribe services are just not well integrated into the EMR. And it seems clunky to turn the mic on and off, and when it's on select the pt, and then turning it off, then turning it on and selecting another patient, and whether it can just read all the data in the chart too.
AI is definitely the future it still seems somewhat nascent though. The human scribes that I work with regularly do 90% of the work and all I do is dictate an MDM, given them the PE, some patient instructions, and maybe a few other things like my own reads of imaging. I work with some pretty good scribes once I train them well. So setting aside the cost of a human scribe, i would rather have a human than AI. It's just a lot of copying and pasting, and most of what's transcribed is just unneeded wordy paragraphs, and reading an AI generated note is very annoying. We need to move towards writing less rather than writing more. AI will save time but it's just gonna suck reading AI notes from a 1 week hospitalization where every hospitalist and specialist uses AI to write. It's just gonna be too much.
Most doctors want to read, or try to find, key phrases and terse, grammatically incorrect chunks. I personally don't like reading sentences. Most of the time when I read a prior note I'm scanning for particular keywords, phrases, numbers, percentages, and I'm just skipping around until I get what I want. My fear is that AI will just write MORE and I now have to scour through more words to find what I need.
Last edited:
Advertisement - Members don't see this ad
So I get referrals from a few UCs. A couple of them are excellent, primarily staffed by ED docs, super UCs. A couple are ok. A bunch are trash, as one expects.
However a few of the good ones have started sending me patients with notes from their docs that are clearly AI enhanced. Probably epic/heidi being used for scribing AND MdM enhancement / consolidating.
Regardless, now I get appropriate referrals that have a god damned 1.5 pages MDM that spends paragraphs discussing the esoteric sensitivity of physical exam for peritonitis, and a ddx longer than my arm. When the story is 30yo with TTP RLQ, concerned for appy, no access to CT.
I need a ****ing AI to condense their AI trash notes (not wrong! Just superfluous) into something I can read without wasting 5 minutes! Ugh.
However a few of the good ones have started sending me patients with notes from their docs that are clearly AI enhanced. Probably epic/heidi being used for scribing AND MdM enhancement / consolidating.
Regardless, now I get appropriate referrals that have a god damned 1.5 pages MDM that spends paragraphs discussing the esoteric sensitivity of physical exam for peritonitis, and a ddx longer than my arm. When the story is 30yo with TTP RLQ, concerned for appy, no access to CT.
I need a ****ing AI to condense their AI trash notes (not wrong! Just superfluous) into something I can read without wasting 5 minutes! Ugh.
I've had a few come in straight from clinic with AI written notes in Epic but I can't figure out why and the patient isn't sure either. The notes are no shorter than 30 bullet points of medical decision making broken down into 5+ categories.
Epic literally has a product for this they sell to Radiology that summarizes the chart.
I wish I were kidding.
We've got that turned on for everyone, I don't much use it since I'm familiar with the patient but it seems to do a decent job of a kinda birds eye view summary of everything going on for people seeing a patient for the first time.
Why is it so damn hard to just write
"34 F with severe lower abd pain, UA and UPT from urgent care negative.
Thank you for eval."
Admittedly there are complex patients these days, much more so than 30 years ago when hand written notes were prevalent.
And writing a brief note should not be hard.
I love surgeons notes. They will take the most medically complicated patient that you spend >1 hour working on, and two days later summarize everything in 1 sentence. It's almost demoralizing on some level. Like I did all the work, have to write several paragraphs, call 4 consults, and you get to write just 1 line and fix it. 🙂
From a UM side surgical notes can be very frustrating. All they care about is getting in OR and doing the shortest note possible. Problem is when paying for said procedure is denied I need to appeal that and if there's no written justification I need to do so much more goddam work
Especially true when there ARE complications, maybe not an official "surgical complication" but someone needs to stay an extra few days for IV dilaudid q2
and their note is
patient doing well. pain controlled.
not helpful!
Especially true when there ARE complications, maybe not an official "surgical complication" but someone needs to stay an extra few days for IV dilaudid q2
and their note is
patient doing well. pain controlled.
not helpful!
Yep. There is a point to complex inpatient notes, or at least thorough ones noting comorbidities, chronic and acute diseases, and any/all reasons to consider them complex inpatients. If nothing else for insurance.
Urgent care notes? Nope. Pretty easy to hit the base requirement…
Urgent care notes? Nope. Pretty easy to hit the base requirement…
That sounds... EPIC????
so true. Their notes are useless 90% of the time.
maybe there is a happy medium somewhere.
You are so right, AI as soon as it gets to AGI will change everything in the blink of an eye. We will never be able to put it back in the box. 2-3 years from now depending on how governments handle this it will have an insane impact on us.
Yeah, I heard someone during an interview say AI was humanities last greatest invention and I kind of rolled my eyes but you know....that might not be far from the truth.
Advertisement - Members don't see this ad
I got referred to this forum from a different radiology forum and felt compelled to respond to this. I'm an attending general DR with 6 years in practice. This is so absurd it is almost certainly wrong. Most likely the AI is spitting out garbage and you cannot differentiate from the garbage it is spitting out and actual true diagnoses/interpretations. I know this because I have used many AI algorithms which perform various functions and so do my colleagues, no one has seen anything have any use whatsoever. ICH detection (Eureka) flags everything with streak artifact as potential brain bleed. Fracture detection frequently says 'positive' though its negative or 'doubt' which is basically useless (thats Gleamer). Pulmonary nodule (AIDoc lung nodule) will flag stable nodules for 10 years (but doesnt tell you whether or not its stable) but miss 3cm paramediastinal obvious lung masses.
So, if you have the deidentified images, why not post them with the AI's interpretation so we can all see the veracity of this?
Also, if you look at the article from Apple's research group, they claim the current LLM based AI's are a misnomer. There's no real 'I' there, no real intelligence. It can't solve simple puzzles that were removed from data contamination, even when the solution is directly provided to the 'AI'. People call me a skeptic but im not. I think when we see actual AI it will blow us away. It's what we are calling 'AGI' but its really what AI was always meant to be. Right now we just have advanced autocorrect. Time horizon wise I have no idea. Could be 5 years or 50 before we get AGI. But it will not be an LLM and it will not be like anything we currently have (LLMs have reached their potential).
This is what im getting so far with AI image uploads (and ive used many AI products this is just one example):
“So absurd it’s almost certainly wrong”
Perhaps absurd, but not wrong. The film showed what I thought was a benign bone tumor, i.e. bone island. I ordered an MRI as follow up. The diagnosis on MRI as read by a radiologist matched that from the AI, and matched my read. We all agreed: me, the radiologist and the AI. N of 1. I haven’t uploaded any more films to ai models before or since.
If that doesn’t match your experience, so be it. I still ordered the MRI as read by a radiologist which I consider the gold standard. And both the AI and radiology confirmed my read.
I’m not sure what’s so hard to understand about this. I did what I’d normally do; ordered an x-ray, followed up a read with a more advanced study as read by a radiologist and he agreed with my original plain film read. And an AI algorithm happened to agree, too. So what? You still got paid last week, right?
Take it for what it is. But I didn’t make it up. Why your AI can’t read a bullet on a film, I don’t know. Perhaps the AI got faked out because it's an unspent bullet (not fired by a gun) with the casing on it, that must have either been eaten by the patient or is sitting on their skin. I don’t know.
Perhaps it was from 6 months or a year ago. AI improves every few months. Perhaps you used a weak AI model
What I’m reporting isn’t anything different than anyone else has reported. And our conversation will not change what effect AI has on Medicine.
Last edited:
Did this image come from a patient who swallowed a round? Because that looks an awful lot like an unfired round on the radiograph.
A bone island is a “difficult radiology image”? I see 30 of these a day my entire career. Your initial post gave me the impression, perhaps falsely, that you sent multiple difficult cases to AI with correct results and no prompting. I didn’t know you sent it one bone island. Yes I don’t think it proves much, I use multiple AI algorithms every day and have for years (I’m forced to) and it only slows me down and gives me extra slop to wade through.
Occasionally I may have to hedge or overcall something because AI flags something that doesn’t look like much of anything, but 1/100 chance it’s a real fracture line, and we all know how hanged I would be in medmal court if I disagreed with AI and was wrong.
I used ChatGPT last week for the bullet analysis. Yes, it’s obviously an unfired cartridge. Yes, ChatGPT is trying to convince us it’s a battery. For actual radiologists, this is what we deal with 100 times a day with AIDOC/gleamer/viz/rapid/other FDA approved product
Last edited:
Perhaps I overstated a bone island being "difficult," and contradicted myself. What I don't want, is to miss something malignant, so I don't hesitate to order confirmatory studies, especially with anxious, hypervigilant patients like this one was. Low likelihood, but severe consequences.
I don't think anyone is arguing AI, as it stands today, is replacing any of us (physicians) today. The question is, what happens if AI gets exponentially better? Nobody really knows the answer. Maybe it will get exponentially better, maybe it won't. Maybe if it does, it doesn't change the job prospects of physicians much at all. Maybe the predictions are exaggerated. They often are. Or, maybe they're not exaggerated. I don't lose sleep over it. I've heard lots of FUD (fear, uncertainty, doubt) over the years. 2026 is no different.
My opinion, admittedly with a low degree of certainty, is that AI (if it gets good enough) will function sort of like a midlevel does in other fields. You keep your job. You use AI tools to do more work. The next generation is affected more; in what ways, it's not entirely clear. I doubt Open AI wants to be on the hook for all malpractice suits in the world. They'll likely want someone else with their own insurance policy to sign off, washing them of liability.
But even that could be an exaggeration. Anesthesiologists and ER physicians fretted for years over mid-levels replacing them. It hasn't happened. More provider supply and greater efficiency just created more demand for their services. We adapted before, we'll adapt again.
Big picture: Physicians are in the 99th%ile of intelligence and productivity. Society will find a way to need us; AI or not.
Last edited:
Any introspection on why turnaround times on radiology are the rate limiting step for your institution? Try looking at the volumes you're ordering today versus 2-3 years ago or 5-6 years ago. What used to be a XR is now always XR + CT +/- US or MR. What used to be a head CT is now a CTA head/neck/perfusion + MR. And so on.
If you're making 1M as a radiologist from home your job is most certainly not chill. You're doing $40/wRVU * 25,000 wRVUs to get there. That's two FTEs of work (MGMA median rad is around 10-11k wRVU/yr). That's a busy day.
Do you ever get things wrong? Do you ever cause delays? You must be the perfect doctor.
Know what you know today, would you tell an MS3 or MS4 to go into Radiology if their goal was a long career for at least 15 years where they could make similar income to a Radiologist today?
When you've been in the field of radiology, you've heard the predictions constantly for the last 15 years, you just get numb to it I guess. I've heard we're 5 years away from radiologists being out of a job since I was in medical school. Before that people said CAD (which is not AI but just a static algorithm) was going to do it. Everythings been hype and useless tech, AI slop.
Look, and this isn't directed at you but the ER docs reading this thread, we (rads) are not your enemy. Yes I make close to 7 figs in PJs. But I work my ass off in those PJs reading studies constantly, answering questions, protocoling studies, etc. We actually share a common enemy, admin.
You may think admin will replace us with a high functioning highly accurate AI. What is much more likely to happen is this: a billion dollar AI company gives an admin a nice kickback/shares in its company in order to hamfist some crap AI which 'replaces the radiologists and will save you millions'. They fire all their radiologists save one to read the questionable cases. Meanwhile, 70% of the CT heads you order gets an AI read of 'Possible ICH'. You call the radiologist to confirm but they don't pick up, there are hundreds of such calls every hour. You complain to admin but they tell you to use your clinical judgement to determine if the possible ICH is real. You then try to explain to an MBA grad from Devry why you can't use clinical judgement on a demented 80yo with witnessed fall by family, there either is a hemorrhage on the CT or not. And you need a radiologist to tell you that, but there are none. You will then know what it is to reap what you sow.
I would be hard pressed to tell people to go into radiology with the expectation of maintaining the same incomes we see today. But none of us have any idea what is going to happen or else we wouldn't be posting about it on SDN. There are no oracles out there when it comes to the short to medium term timeline on AI in radiology. Eventually, sure. It could be 20 years before real displacement/replacement happens and in the mean time we're just making the shortage worse by understandable fear mongering.
Last edited:
The biggest change to me when I moved across the country for my first attending job was how the Radiologist is the first and often only physician to “see” a patient in the new midlevel norm for up to 1-2 days.
Patient comes in to ER. Sees NP/PA. Gets misc imaging. Based on imaging/labs, admitted to hospitalist NP/PA with consults to Cardiology or CritCare NP/PA or surgical NP/PA.
Rounded on the next morning with NP/PA. Maybe they see a physician this day. Maybe the next.
But they’ve got imaging every day.
Many times we call a critical finding and just get silence, or worse, “what am I supposed to do with that”. It was bizarro to think we are the only MDs “touching these patients” but it’s true.
Maybe the whole army of mid levels with OpenEvidence or Perplexity will be good enough. Or maybe it won’t.
But what I can say is it’s unreal the amount of evidence we require for new clinical devices or drugs or even surgeries yet we are running a real time “experiment” with both midlevel directed care and unvalidated clinic AI “tools”.
Patient comes in to ER. Sees NP/PA. Gets misc imaging. Based on imaging/labs, admitted to hospitalist NP/PA with consults to Cardiology or CritCare NP/PA or surgical NP/PA.
Rounded on the next morning with NP/PA. Maybe they see a physician this day. Maybe the next.
But they’ve got imaging every day.
Many times we call a critical finding and just get silence, or worse, “what am I supposed to do with that”. It was bizarro to think we are the only MDs “touching these patients” but it’s true.
Maybe the whole army of mid levels with OpenEvidence or Perplexity will be good enough. Or maybe it won’t.
But what I can say is it’s unreal the amount of evidence we require for new clinical devices or drugs or even surgeries yet we are running a real time “experiment” with both midlevel directed care and unvalidated clinic AI “tools”.
It is very true that hospital administrators will pay doctors a salary of $1 if they can get away with it. But here's something to think about: Ask yourself why that "slop" AI that they build into your software, that isn't good enough to replace you, is there at all? They know it's slop, but they're paying somebody to embed it there. Why?
Are they collecting data with the hopes it will be good enough someday? Are they harvesting your interpretations, to train their models to get better? It's a question worth asking. And if the slop AI doesn't help you, you have to ask yourself, why are you signing hospital contracts potentially requiring you to train your replacement? Their plan may never work. It may never get good enough to replace you. But someone its hoping it will, trying really hard to make it so and collecting data from you to do so.
Bring this up with your physician group CEO. Next time your rads group negotiates their hospital contract, ask to have the slop AI removed from the software. It should be a very easy feature for the software people to turn off, if it's "only there to help you." Tell them it doesn't help, it only slows you down and distracts you such that patient care is at risk.
See what they say. I bet they refuse.
Last edited:
Being a luddite isn't going to accomplish much. If AI is going to be better than us and is going to be a new profitable business then it will be found to be so regardless of what your individual hospital does. In the mean time hospitals are free to waste their money on whatever hype is sold to them by the cute AI sales girl they met at the hospital admin management conference. Very few if any of these products have true ROIs for the person paying the bill for the service.
Most of the time they buy it because of the CMS NTAP payments. The other, more insidious reason is that they are being paid for training, which they would never admit.
The only useful AI tools I’ve seen
1. Powerscribe smart impression / RadAI Omni impression
2. Riverrrain Clearread
3. AIDOC triage tools (I particularly like the incidental PE for outpatient scans)
There’s a dime a dozen reporting platforms (see Radpair) that all use the same garbage web.asm openwhisper text to speech which is god awful at medical transcription for some reason. But their report generation can be decent if you are doing ER type work. These models tend to get confused if doing outpatient specialty work like oncology. I assume that’s due to context exhaustion.
D
deleted1223140
Of course I'm not perfect, but I shouldn't be getting paid half of what they're getting paid.
What should a radiologist working at 95th percentile productivity get paid when they're doing the work of two radiologists? Median rad probably makes much closer to what you make (MGMA 50th percentile is 580k). The guys making 1M love to brag but they're outliers and are working extremely hard.
Advertisement - Members don't see this ad
Radiology is much harder cognitively than EM, there's no question.
The shortage is artificial and fear-mongering won't change residency fill rates. All it will do is change the type of candidates that get into US Radiology programs. Instead of being a competitive MD-only field, it'll get filled with mid-low tier DOs, Caribbean students (US citizen) and non-US-citizen IMGs.
Also AI companies don't need US-trained radiologists to help train their models. You can go to outlier.ai or any of these other data-gathering websites and earn $100-150 per hour doing physician-level labeling/checking of studies, cases, and other important work that AI needs to dial in its outputs.
This pay rate is more than many countries pay their radiologists per hour, so you can bet that the Egyptian and Syrian radiologists are cashing in on that revenue stream.
And then of course if the foreign-trained radiologist dam breaks, all bets are off.
Couple years ago I read a post from a radiologist that a radiology shift was like taking a step exam the entire time. True that. Probably the most cognitively demanding/exhausting specialty.
I talk to a radiologist almost every shift. Sometimes they're calling me on something critical or particularly confusing and they want to clarify what I'm worried about. Sometimes I don't understand the read because it's something I've never heard of (this happens less as I'm now old). Sometimes it's to rapidly coordinate care. Sometimes it's because they missed something important (the ureteral stone or subtle fracture) because they're reading fast and short staffed. I have the central office number memorized and the rads are never unhappy to speak with us. We're all in the service industry section of medicine, after all.
Housfield said it best but you know the answer. Admin controls these decisions a staff radiologist has no control over this whatsoever. The CEOs or administrators buying these products are likely getting a kickback in the form of shares, board seats, cute sales girls that take them to dinner, or whatever other perks are associated with the AI company. At the very least, they are basically selling hospital data to the AI companies for free, and they will definitely get something for that.
Because CEO's/admin is choosing to purchase these crap models, even though we request them not to or question why, these models must be good? Is that the argument? What about the crap AI models in clinical medicine that flag for 'sepsis' or 'dvt risk' or whatever. I'm sure you guys have your fair share of AI slop as well. Why don't you petition it be removed? If they refuse, does it mean its useful?
I'm not going to antagonize someone higher up for some AI slop that I can easily ignore. I'll just ignore it and move on. They are paying me 7 figs in my PJs after all.
I don't think my job is harder than EM. I kind of cringe at the comparisons between the two. They're both hard jobs.
Pay gaps between fields aren't about fairness or who does a harder job. They're due to price setting by the federal government and are entirely disconnected from the marketplace of supply and demand. Peds makes a fraction of EM which makes a fraction of ortho spine or Mohs surgery. None of it is about utility to society or the intrinsic economic value of the service you provide. Some pencil pusher from Harvard decided it should be so, and that's the system we have.
You're not. If you're an average EM doc you're probably getting paid about 20% less than an average radiologist at this time based on MGMA data. Rads is a field where the fast people can far outstrip the slow ones in pay. Many jobs offer pay per rvu (pay per click) arrangements where you get paid based on volume. So the top 10% of readers in terms of speed are going to make a LOT more. Since everything is shift work in EM (and also still in some jobs in rads) its harder for the top 10% to make much more.
Knowing what you know today, would you tell an MS3 or MS4 to go into Radiology if their goal was a long career for at least 15 years where they could make similar income to a Radiologist today?
I’m not the person you asked but I’d reframe the question entirely:
Would you tell anyone to go into any white collar profession with those same parameters? I dont think any medical specialty is necessarily safe given those constraints.
If the person is already in medical school and had an aptitude for radiology, I’d tell them to guide their career into the technical/business side as soon as possible.
This is, of course, a natural conclusion of all of this. The problem is that there is very little room at the top. AI seems to be a winner-take-most type of tool, so I can't imagine the thousands of newly minted radiologists 5-10 years from now will all have a reasonable shot at any career if the singularity/specialty-ending technological jump happens.
And to this the reply tends to be, "but if we get to that point, everybody would be screwed." Indeed, indeed.
Maybe I should spend my time diving down into the prepper rabbit hole and start acquiring real assets and real skills.
I personally don't know the first thing about butchering a cow, or growing my own carrots.
I hear you can be solar-powered generators at Costco now for about $1000.
That's a good start.
Not only would I say that yes I would tell them to go into it, but I currently do talk to medical students and tell them that.
I think one thing is for sure: It would be very hard for us to Freaky Friday our jobs unless you give me "ureteral stone only without any significant incidental findings" body CT day or maybe "wrist X-ray" day. I'll let you pick what sort of patient presentation you'd like to do but would strongly recommend "ear pain for 20 minutes" or "work note" day. It's no coincidence that these are often the same day.
top 10 listicle writer was NOT on the 10 ten list.
So, I want you rads guys to take a look at that chart. It may mean nothing to you and that's fine but that's the "The Last Ones" security AI test build by the UK AI Security Institute to test AI cyber capabilities. It's a 32 step simulated corporate network red-team penetration testing attack chain. It takes the absolute best of the best humans in the world about 20 hours of focused work to complete it. GPT-4o could only get 1.7 steps. Opus 4.6 got around 10 steps. Claude Mythos preview is the first model to ever complete it. Look at those massive generational leaps in capability. We're talking mid-20025 models to one year later. If you guys think these models can't and won't overtake your ability to safely, efficiently and accurately diagnose radiology imaging, you are crazy. And it won't take 15 years either. That's just the scary reality of a medical specialty that has little to no patient interaction. It's pure cognitive work. (Diagnostics) ANY job centered about human cognitive work WILL absolutely be overtaken by AI. It's inevitable. Does that mean you'll be without a job in a few years? Of course not. What it means is that the AI that you dismiss as (AI slop) will become good enough that it starts to impress you and then impressed you even more and then you will find yourself relying on it more and more. That's the key to keeping your jobs. Leveraging it to make you MORE productive. Who knows, you might find that you finish the day feeling less stressed than you do now because you could get so much done with AI and the reads have lower medicolegal risk because it's getting checked over and over by the systems QA.
The absolute worst thing any of us can do is dismiss the technology which is the human thing to do when we all feel like our job is threatened. (Nobody can do it as well as ME, etc..) What's important is for all of us to jump on board leveraging this technology. I think what most of us will find is that we are probably as physicians one of the last human workers to be replaced by AI because as some of you have said, the human interaction is important. Discussing the read with the surgeon who's planning a surgical approach. The ER doc that wants to talk with you about the clinical presentation and ask you a few questions about the read. That kind of stuff can't readily be replaced by AI (Yet) and I think that's going to be true for quite a long time so I don't think any of us need to worry about that.
If you DO get replaced, it's more than likely going to be a new radiology grad, nowhere near your level of experience or expertise that jumped on the AI bandwagon early and can knock out reports in half the time that you can leading to them being overwhelmingly productive and your employer deciding to let you go and keep the young guy.
As for the rest of the discussion in here, I hope none of the rad guys see us as enemies because I was honestly a bit surprised to see some of the snarky comments in here. There's no way any of us could do our jobs without you. 100% of management and disposition depends on DIAGNOSTICs and you guys are the gatekeepers. If you feel overwhelmed with the amount of reads, don't project that onto us please. We are slammed each and every day and forced to see more and more people in les and less time in a growing medicolegal environment and diagnostic imaging is just part of that beast. You guys are being forced to read more and more reads when you probably need MORE people and that's just admin screwing both of us but we truly are not enemies and I hope we can always remember that.
You’re putting the cart before the horse here. The assumption, for some reason, seems to be that exponential improvement will happen for AI in every aspect continually, but I don’t see convincing evidence that that’s the case and there is evidence that AI does have limitations that are harder to overcome than previously thought. See autonomous driving for example. Tesla FSD has been at level 2 for many many years, while the most functional autonomous cars are stuck at level 4 in very defined geographical areas. Level 5 remains theoretical. The vast, vast majority of driving is still done by humans and commercial autonomous driving requires constant human attention. What’s to say that this won’t be the case with radiology where very specific, narrow use may affect a small amount of imaging without overall making a large impact on the specialty in general? What evidence do you have outside of AI specifically improving rapidly in a few specialized aspects like coding?
I’m all in on embracing and mastering AI as it comes as I am going into radiology, but most of what we have is trash. Yes, there will be improvements but I see no indication that the trajectory will be an exponential curve. The people who are bought in on the idea seem very assured, and I just don’t see why they are given lack of evidence of dramatically improved performance in the real world. Studies can be a different story because of the artificial conditions they create. You can make studies say anything, so I’m not convinced by these companies who publish their studies then fail to perform when the rubber hits the road.
Advertisement - Members don't see this ad
Interesting... a very different answer than your colleague gave above. Lends credence to the fact that "nobody knows the future"
I think Radiology is the best specialty in medicine. But I also think it will change a lot. The biggest “problem” with radiology is that physicians largely got out of the technical operations after the 2007 cuts. The technical side has the capital to make AI/software investments. That’s why I said I’d recommend residents target that area going forward.
But I don’t think I have the foresight to know where things will be in 15 years for any specialty much less any white collar job.
It’s trendy to say “go be an electrician or hvac” but I grew up in construction. That market is boom/bust. It SUCKED from 2008 to about 2018 and new construction sucks again.
You don't follow this space like I do and that's totally fine. You're talking about Moravec's paradox. Easy for humans, difficult for machines. AI operating in an unpredictable physical world with concerns for safety, regulatory approval, massive liability exposure, etc.. Radiology diagnostics and TLO are the complete opposite. Abundant training data. Verifiable outcomes. No real time physical risk during inference. Cloud scale compute. FSD is the absolute worst comparison. This whole principle is why robotics is nowhere near advancing as fast as AI. The cognitive/agentic areas are where these models thrive and is where the absolute greatest advancements will be observed first.
I leave you with the classic Will Smith eating spaghetti evolution within 3 years. But we don't have anything to worry about because these models move so slow, right?
Like I'm so tired of these posts. I'm tired of replying to them to be honest. Alpha Fold anyone? You used x-ray crystallography and cryo-EM over 60 years to produce only 200K proteins and AI+deep machine learning enabled us to produce over 200+ million in a single year. How long would that have taken us otherwise? That's miraculous. AND that's a system that's a few years old at this point! You guys need to stop with these statements. It's ridiculous in 2026.
I truly hope the mass AI skepticism and distrust continues for at least a few more years so all of us early adopters can continue to leap frog everyone else by adopting and leveraging it early. I feel like that guy from inception every time I try to convince people.
