AI is Evil

[yesmaster]

Advertisement - Members don't see this ad

I was listening to podcasts this morning and thanking my lucky stars I don’t work in tech. Meta tracking its engineers keystrokes to train their AI replacement bots. Uber tracking its drivers to train self driving cars.

The narrative from CEO’s and AI bosses that those who adopt AI will have greater job security than those who don’t, while AI adoption comes apace with broad layoffs.

Horse ****.

I would be very wary of using any AI in which your usage can be used to train your replacement - especially since 99% of rad onc’s aren’t busy to begin with.

 

[OTN]

The Tower of Babel must not be reconstructed

Encyclical Letter of His Holiness Leo XIV Magnifica Humanitas (15 May 2026)

ENCYCLICAL LETTER MAGNIFICA HUMANITAS OF HIS HOLINESS POPE LEO XIV ON SAFEGUARDING THE HUMAN PERSON IN THE TIME OF ARTIFICIAL INTELLIGENCE [ Multimedia ] ___________________________

www.vatican.va

 

[Gfunk6]

I understand why the Meta layoff headlines make people nervous, but I do not think the comparison to radiation oncology is particularly useful.

A technology company reducing headcount because AI can replace or compress portions of internal technical work is not the same thing as replacing a physician in a licensed clinical specialty. Their job is not our job.

Radiation oncology is not simply generating contours or producing a treatment plan. The physician role is deciding whether radiation is appropriate at all, selecting dose and fractionation, weighing competing risks, integrating surgery and systemic therapy, counseling the patient, managing toxicity, coordinating with other physicians, and ultimately taking responsibility for the decision.

There is also a credentialing and licensure structure here that people seem to be skipping over. Medical license, residency training, board certification or eligibility, hospital privileges, payer credentialing, malpractice coverage, physics QA, peer review, radiation safety, regulatory compliance, and a legally accountable treating physician. That is not a small detail. It is the entire operating environment.

AI will almost certainly make parts of our job faster. Chart review, documentation, contouring assistance, plan generation, QA, prior authorization letters, treatment summaries — fine. Let it. Most of that is not the soul of the specialty anyway.

But improving workflow is not the same thing as replacing the radiation oncologist. The more likely future is that productive physicians become more productive, and low-value clerical/technical friction gets compressed.

So yes, AI is coming to radiation oncology. But I think the panic about wholesale physician replacement is misplaced. The realistic question is not whether AI replaces radiation oncologists. It is whether radiation oncologists learn to use AI well enough to stop wasting time on tasks that never required an MD in the first place.

 

[dozitgetchahi]

I was listening to podcasts this morning and thanking my lucky stars I don’t work in tech. Meta tracking its engineers keystrokes to train their AI replacement bots. Uber tracking its drivers to train self driving cars.

The narrative from CEO’s and AI bosses that those who adopt AI will have greater job security than those who don’t, while AI adoption comes apace with broad layoffs.

Horse ****.

I would be very wary of using any AI in which your usage can be used to train your replacement - especially since 99% of rad onc’s aren’t busy to begin with.

I don’t use any of these ambient AI services, and I doubt I ever will.

 

[drowsy12]

I don’t use any of these ambient AI services, and I doubt I ever will.

How exactly does using Ambient endanger rad onc jobs?

this part seems like fear mongering to me.

 

[medgator]

I understand why the Meta layoff headlines make people nervous, but I do not think the comparison to radiation oncology is particularly useful.

A technology company reducing headcount because AI can replace or compress portions of internal technical work is not the same thing as replacing a physician in a licensed clinical specialty. Their job is not our job.

Radiation oncology is not simply generating contours or producing a treatment plan. The physician role is deciding whether radiation is appropriate at all, selecting dose and fractionation, weighing competing risks, integrating surgery and systemic therapy, counseling the patient, managing toxicity, coordinating with other physicians, and ultimately taking responsibility for the decision.

There is also a credentialing and licensure structure here that people seem to be skipping over. Medical license, residency training, board certification or eligibility, hospital privileges, payer credentialing, malpractice coverage, physics QA, peer review, radiation safety, regulatory compliance, and a legally accountable treating physician. That is not a small detail. It is the entire operating environment.

AI will almost certainly make parts of our job faster. Chart review, documentation, contouring assistance, plan generation, QA, prior authorization letters, treatment summaries — fine. Let it. Most of that is not the soul of the specialty anyway.

But improving workflow is not the same thing as replacing the radiation oncologist. The more likely future is that productive physicians become more productive, and low-value clerical/technical friction gets compressed.

So yes, AI is coming to radiation oncology. But I think the panic about wholesale physician replacement is misplaced. The realistic question is not whether AI replaces radiation oncologists. It is whether radiation oncologists learn to use AI well enough to stop wasting time on tasks that never required an MD in the first place.

Ditto for rads. Rads gets mentioned almost every time as the specialty most likely to be replaced by AI. AI doesn't take liability or sign off on reports though 🤷

 

[OTN]

Ditto for rads. Rads gets mentioned almost every time as the specialty most likely to be replaced by AI. AI doesn't take liability or sign off on reports though 🤷

A lot of the twitter discourse, though, centers around primary care/diagnostics. Easy to understand why from a layperson's perspective: hard to get an appointment with a PCP, lots of times when you do it's with an NP or PA, and then once you do from their point of view the end result is no different from what a LLM would have done.

However, those who use twitter and AI are a very different population from The General Population, and I don't think the twitteratti understand how messy patient data/inputs actually are in the real world.

 

[grenz]

How exactly does using Ambient endanger rad onc jobs?

this part seems like fear mongering to me.

I agree with this. I use AI scribes, have read the TOS, am satisfied by the TOS, and find that they materially make my day less cognitively taxing and probably shorter.

You have to put in the work to create your custom template though. The out of the box note style is primary care, but with some custom instructions you can make it output a good rad onc note.

 

[TheWallnerus]

 

[yesmaster]

For folks who find ambient AI has a great deal of utility, go for it. If marginal, however, it seems like a no-brainer that the value from using ambient AI is not accruing to you as the clinician but to another layer of admin, no different from an insurer or UM or hospital middle manager.

There is also a credentialing and licensure structure here that people seem to be skipping over. Medical license, residency training, board certification or eligibility, hospital privileges, payer credentialing, malpractice coverage, physics QA, peer review, radiation safety, regulatory compliance, and a legally accountable treating physician.

Obviously AI taking your job is not a near-term risk. Radiologists will get replaced before rad onc’s. I’m just saying that the AI technologists and adopters — CEO’s and capital owners — are not friends of labor or w2 employees which most of us are. There are certainly ways to disintermediate physicians from the value we create and capture, without replacing physicians.

 

[Gfunk6]

Obviously AI taking your job is not a near-term risk. Radiologists will get replaced before rad onc’s. I’m just saying that the AI technologists and adopters — CEO’s and capital owners — are not friends of labor or w2 employees which most of us are. There are certainly ways to disintermediate physicians from the value we create and capture, without replacing physicians.

I actually think this is the better concern.

I do not think AI replacing radiation oncologists is a serious near-term risk. There are too many clinical, regulatory, credentialing, liability, and trust barriers for that to be realistic anytime soon.

But AI being used by hospitals, payers, vendors, and large employers to capture more of the value physicians create? Sure. That is a real concern.

That is also not unique to AI. Medicine has already been moving that way for years through consolidation, employment models, payer controls, prior auth, protocolization, RVU pressure, and administrative standardization. AI may accelerate that process, but it did not invent it.

Where I disagree is the idea that being suspicious of AI or refusing to use it somehow protects us. I think the opposite is more likely. If physicians do not understand these tools, they will be implemented by administrators, vendors, and payers anyway — just without meaningful physician input.

The issue is not whether AI exists. It is who controls it, who owns the data, who designs the workflow, and who captures the productivity gain.

If AI makes chart review, contouring, documentation, QA, treatment summaries, and prior auth less painful, great. I have no nostalgia for clerical misery. But physicians should be involved in how these tools are deployed, and the productivity gains should not simply become “same pay, more work, more oversight.”

So yes, I agree the labor/capital concern is real. I just do not think Luddism is a serious defense. The better answer is physician fluency and physician governance. Use the tools, understand the incentives, and do not let people who do not treat patients redesign the specialty around us.

 

[drowsy12]

AI can reduce rad onc jobs in the arc of time

abridge is not why


The logic of yesmaster doesn’t track in regards to abridge


To be clear I don’t use any AI for note taking. It just doesn’t pass the logical smell test to not be able to discern between different AI uses.

Using chatgpt will kill my job too?

 

[WildRivers]

Short sighted, IMO. We will be able to see / treat many more patients than we can now. But, the number of patients is not going up, or at least in terms of number of patients per doctor. It's not that we will all be out of a job. It's that we will all be able to handle more work.

Back in the late 90s-00s, 5 consults a week was plenty. It was plenty of RVUs/income and the "work" it took to get the job done was significant. Now, I think many efficient docs could handle 10 consults in a 4-5 day and 40 hour work week. Many people here already are doing >10 new patients a week, maybe some see 12-15 routinely.

Imagine - the AI scribes + parsing software reading documentation seamlessly creating a note. The auto-contouring reaches a level that surpasses humans and even GTV delineation is managed by the software in many non-complicated cases. IGRT is already mostly managed by the RTTs and the auto-match, but that will get even better. If ROCR goes through and a single ICD10 code gets you full payment, so weekly management is no longer needed for all patients every week. In addition, we are getting paid less per patient, so to maintain income, those that can see more patients will try to see more patients. Tumor board? Document parsing + imaging review + histo review = treatment plan. There is a great paper on this I highlighted a while ago on the substack. There are also the "unknown unknowns" that will arise over the next 5-10 years - designer meds that pre-treat people with potential cancers (think of this single shot treatment for LDL that was published last week!) or other treatment options. There are papers highlighting AI designed medications that appear to work.

I don't think it's a chatbot replacing us. I think it's guys like OTN that will be able to routinely see 20+ consults a week and manage 65-70 patients on beam with assistance from an extender. Maybe she/he is already doing that, but now others will be able to, as well. If doing that much volume, the E&Ms become less important and a lower skilled "provider" can see the vast majority of those patients.

Not a doomer, but thinking today's technology is the limit will constrain your view of the future. In 2021, even as a techno-optimist I could not imagine what 2026 AI would look like. 2031? Who knows, but if it stays at this speed, how can one not have some level of anxiety?

Check these in 5 years 🙂

2031 Predictions
Documentation nearly entirely automated - 90%
Autocontouring of targets is nearly entirely automated - 80%
At least one of the top 5 (non skin) cancers can be pre treated curatively (prostate, breast, lung, colorectal, bladder) - 10%
Tumor board is automated - 40%
Unfilled residency spots increases to 15% - 70%

This is from Gemini - ‎Gemini - direct access to Google AI

 

[Faol]

I use Ambience daily for notes. It’s actually quite helpful for the assessment, as it consolidates it well. It also helps with the ROS and PE, as it logs those things I say to the patient in the moment, but forget by the time I get to notes at the end of the day. The doom and gloom is a bit overblown IMO. By the time AI takes physical jobs over, Skynet will have already become self aware. So bigger fish at that point.

 

[NotMattSpraker]

This is from Gemini - ‎Gemini - direct access to Google AI

The link is a fun read.

I disagree with a lot of it, but liked this one: "To justify their headcount and maintain income, radiation oncologists will have to aggressively claw back clinical territory they previously outsourced or ignored."

All those "oncologist first" rad oncs out there, your time to shine.

E/M doesnt pay a lot though, so well see how that plays out 🙂

 

[grenz]

The link is a fun read.

I disagree with a lot of it, but liked this one: "To justify their headcount and maintain income, radiation oncologists will have to aggressively claw back clinical territory they previously outsourced or ignored."

All those "oncologist first" rad oncs out there, your time to shine.

E/M doesnt pay a lot though, so well see how that plays out 🙂

It’s fun to read stuff like this, but if you start pushing back on certain assertions, you will see that the LLM parrots what it thinks you want to hear.

 

[grenz]

I use Ambience daily for notes. It’s actually quite helpful for the assessment, as it consolidates it well. It also helps with the ROS and PE, as it logs those things I say to the patient in the moment, but forget by the time I get to notes at the end of the day. The doom and gloom is a bit overblown IMO. By the time AI takes physical jobs over, Skynet will have already become self aware. So bigger fish at that point.

How do you feel about the assessment and plan? I feel like most rad oncs have long, flowery A/P that are kind of like a calling card to the referring, but in my experience the AI A/P hits all the high notes but produces a much more surgeon-like A/P

 

[NotMattSpraker]

It’s fun to read stuff like this, but if you start pushing back on certain assertions, you will see that the LLM parrots what it thinks you want to hear.

People should always assume it's trying to please you and should make a habit of pushing back. It gives a lot of interesting information if you argue with it!

Also, that LLM read out sounds specifically like someone that knows a lot about medicine from reading books/papers, but doesn't spend much time working in a clinic face to face with patients. There are a lot of those folks out there.

 

[grenz]

People should always assume it's trying to please you and should make a habit of pushing back. It gives a lot of interesting information if you argue with it!

Also, that LLM read out sounds specifically like someone that knows a lot about medicine from reading books/papers, but doesn't spend much time working in a clinic face to face with patients. There are a lot of those folks out there.

It’s also worth imagining what training data it’s inferencing off of when you ask it to speculate on the future of radiation oncology in the context of flat demand and rising labor supply. There’s a heck of a lot of human-generated training data about that topic on SDN!

 

[Palex80]

I am confident that all of you will stop using AI to autocontour OARs after reading this thread...

 

[HolySmokes]

I am confident that all of you will stop using AI to autocontour OARs after reading this thread...

Jokes on you, my hospital still hasn't even purchased any autocontouring software!

 

[Palex80]

Jokes on you, my hospital still hasn't even purchased any autocontouring software!

 

[WildRivers]

Agree with all this - it’s not for sure, it’s not all doom and gloom. It’s one perspective and it’s from my account and of course I have my own biases and the way I ask questions may lead it in a certain direction.

 

[Lamount]

My concerns about increased efficiency from AI harming the future job market are somewhat tempered by the fact that medicine -as a whole- is seeing a mass exodus. This will only get worse if the changes to federal student loan limits calcify... as most folks won't be able to afford med school without selling their souls to private student loan sharks

 

[yesmaster]

The thing is most rad onc’s aren’t limited by hours in the day so even if AI boosted efficiency by 500% with some improvements in radiation quality the latter of which is doubtful, even the top 1% of rad onc’s are not suddenly going to see 50 new patients a week and have 150+ on treat. There’s other reasons why the average rad onc is at 250 patients a year.

Gemini’s race to the bottom example of cost cutting and bargain bin radiation is dumb for everyone involved.

90% of my sentiment is pessimism about AI in general, not just AI in medicine or rad onc.

I don’t think a 10x or 100x increase in compute does much of anything for rad onc since we are structurally bounded by the scenarios in which radiation would benefit patients plus those for which insurers will pay. Something like an adaptive unlock would be helpful but my understanding is that’s a billing problem rather than a technical or compute problem.

 

[yesmaster]

This will only get worse if the changes to federal student loan limits calcify... as most folks won't be able to afford med school without selling their souls to private student loan sharks

It does feel like something is wrong with the economy and the price of everything

 

[Faol]

How do you feel about the assessment and plan? I feel like most rad oncs have long, flowery A/P that are kind of like a calling card to the referring, but in my experience the AI A/P hits all the high notes but produces a much more surgeon-like A/P

I actually like it a lot. I have my templates, very long with studies. I still include this, but added a plan section with the bullet pointed AI summary and a one line CT setup. So, best of both worlds.

 

[xrt123]

What data is there that

My concerns about increased efficiency from AI harming the future job market are somewhat tempered by the fact that medicine -as a whole- is seeing a mass exodus. This will only get worse if the changes to federal student loan limits calcify... as most folks won't be able to afford med school without selling their souls to private student loan sharks

I think there are some legitimate concerns in that statement, but the overall conclusion doesn’t really match what’s actually happening in medicine right now. Burnout is real, and there is clearly dissatisfaction in the profession, but that is not translating into a flooded or collapsing job market. If anything, the opposite is true. Demand for physicians continues to outpace supply across most specialties, including oncology, driven by an aging population, rising cancer incidence, and increasing complexity of care. Even with some attrition, the system is still facing a structural shortage of physicians, not a surplus.

The AI argument also tends to be overstated. We’ve already seen this play out in adjacent fields like radiology, where predictions of job loss were made years ago and haven’t materialized. Instead, what happens is that efficiency improves, throughput increases, and demand expands. Physicians end up doing more meaningful, higher-level work while health systems actually hire more people to meet the increased capacity. Radiation oncology is even less susceptible to replacement because it requires clinical judgment, multidisciplinary coordination, procedures, and longitudinal patient management. AI is far more likely to reduce low-value tasks like contouring and documentation than replace the physician role itself.

The strongest part of the concern is the financial barrier to entering medicine, but even there the implications are a bit different than what’s being suggested. Medical education costs have clearly skyrocketed over time, far outpacing general inflation, similar to trends seen in healthcare costs more broadly. Federal loan policies have played a role in that by expanding borrowing access, which in turn allows institutions to raise tuition—creating a feedback loop that drives prices higher. If anything, tightening federal loan availability could restrict access to medical education and reduce the pipeline of future physicians. That may be bad for equity and access, but from a labor market perspective it actually reinforces physician demand rather than weakening it. If someone actually care about problems in the economy I would please ask you to review costs over time of various industries. EDUCATION AND MEDICINE (followed by housing) outpace everything when looking at inflation adjusted cost to the USA. Clearly government policies are the cause of this (federal student loans/governement regulation).

 

[temujim]

What data is there that

I think there are some legitimate concerns in that statement, but the overall conclusion doesn’t really match what’s actually happening in medicine right now. Burnout is real, and there is clearly dissatisfaction in the profession, but that is not translating into a flooded or collapsing job market. If anything, the opposite is true. Demand for physicians continues to outpace supply across most specialties, including oncology, driven by an aging population, rising cancer incidence, and increasing complexity of care. Even with some attrition, the system is still facing a structural shortage of physicians, not a surplus.

The AI argument also tends to be overstated. We’ve already seen this play out in adjacent fields like radiology, where predictions of job loss were made years ago and haven’t materialized. Instead, what happens is that efficiency improves, throughput increases, and demand expands. Physicians end up doing more meaningful, higher-level work while health systems actually hire more people to meet the increased capacity. Radiation oncology is even less susceptible to replacement because it requires clinical judgment, multidisciplinary coordination, procedures, and longitudinal patient management. AI is far more likely to reduce low-value tasks like contouring and documentation than replace the physician role itself.

The strongest part of the concern is the financial barrier to entering medicine, but even there the implications are a bit different than what’s being suggested. Medical education costs have clearly skyrocketed over time, far outpacing general inflation, similar to trends seen in healthcare costs more broadly. Federal loan policies have played a role in that by expanding borrowing access, which in turn allows institutions to raise tuition—creating a feedback loop that drives prices higher. If anything, tightening federal loan availability could restrict access to medical education and reduce the pipeline of future physicians. That may be bad for equity and access, but from a labor market perspective it actually reinforces physician demand rather than weakening it. If someone actually care about problems in the economy I would please ask you to review costs over time of various industries. EDUCATION AND MEDICINE (followed by housing) outpace everything when looking at inflation adjusted cost to the USA. Clearly government policies are the cause of this (federal student loans/governement regulation).

Agree with this. AI far more likely to hollow out APPs than MDs. For rad onc, dosimetry may get hit, but won't go away. As always, physicists are the smartest and most secure.