-
Why aren’t you all freaking over the Chat GPT-4?
Posted by emilielegay on March 15, 2023 at 9:24 amGPT-4 has a new ability to respond to images as well as text. Greg Brockman, OpenAIs president and co-founder, demonstrated how the system could describe an image from the Hubble Space Telescope in painstaking detail. The description went on for paragraphs.
It can also answer questions about an image. If given a photograph of the inside of a fridge, it can suggest a few meals to make from whats on hand
.[link=https://www.google.com/amp/s/www.nytimes.com/2023/03/14/technology/openai-new-gpt4.amp.html]https://www.google.com/am…enai-new-gpt4.amp.html[/link]
buckeyeguy replied 1 year, 5 months ago 29 Members · 48 Replies -
48 Replies
-
Unknown Member
Deleted UserMarch 15, 2023 at 9:30 amMostly because doctors typically lack imagination and outside the box thinking.
-
Wow it can recognize fruits and yogurt. Board Certified material.
-
Quote from MidwestIR
Wow it can recognize fruits and yogurt. Board Certified material.
I think that’s a bit naive to sum this technology down to that. 5 years ago, nothing like this existed. But now think about what it would be able to do in 5 year. Maybe since your IR it doesn’t hit you the same way. But imagine your field. If the radiology part is automated, why do we need to pay an IR guide 500k to do procedures when I could hire a PA, teach them basic IR skills and have a machine do the rest
-
Because in radiology the leaps between:
1) Generating complete medical knowledgebase ->
2) Finding recognition ->
3) Medical interpretation
are astronomical. Unlike food ingredients -> meal
As you should know if you’re a radiologist.
Edit: not to say that it won’t ever happen, but almost certainly not in most of this forum members’ careers which is why most here aren’t freaking out (to answer your original question).-
Its not going to keep me up at night. But I’ve already done things that should mitigate the hit to some degree, if AI should be able to take over image interpretation:
-Learned many procedures. AI isn’t getting these anytime soon, and I’ll be lower on the chopping block.
-Joined a physician owned multispecialty group that owns its own scanners. Hospital or VC can’t kick me to the curb overnight
-Lived way below my means and have lots of diversified investments with a small remainder on my mortgage.
-Married a woman who [i]can [/i]be happy with clothes from target
With an aggressive approach, a new grad could be financially independent within a few years and not lose any sleep over AI.-
Can it read an EKG with 98+% accuracy ?
When it does let me know. Then basic x-rays are 10-20 years away.
-
-
The doctors who need to know how to read an EKG can do so accurately. Gotta be a troll ..
-
Rads are in the best position to take advantage of AI imo. Also, AI is generally good at single tasks. So we can use it for the boring tasks and improve efficiency. There are a ton of old Rads in practice and new Rads don’t want to work like dogs, so AI can help us currently working read more volume and make more money.
Full replacement of a Rad is basically impossible for decades similar to self-driving cars.
For my kids – I’m less certain about Rads. I actually wanted them both to become Radiologists cause I think it is such a great career and I could guide them. But, not sure due to AI potential down the road. Without Radiology as an option, I would not recommend a career in Medicine to them at all, as I feel the rest of Medicine is pretty terrible as a career.-
Unfortunately I see several radiology report impressions that just restate findings. Those guys make it that much easier for AI to replace us.
-
I own the buildings, I own the machines, I employ the technologists. The day I can buy software that does my job so I can just sit back and relax will be the happiest day of my career.
-
Unknown Member
Deleted UserMarch 16, 2023 at 9:27 amThere are two issues: whether the AI is a proximate threat; and whether we freak out. Just because it is a proximate threat – i.e. assuming that – doesn’t mean you have to freak out.
Although our jobs can be threatened by some new technology – whether it’s days, months, or decades – we as people shouldn’t be threatened.
Freaking out does nothing. It’s like anything, say gunshots ring out when you’re at the supermarket with your kid. Take a few deep breaths to calm the rough seas in you; take account of the threat again and your options; and move forward with your best plan.
For them to hurt you and have you freak out, you have to be complicit, nothing can hurt you – certainly not loss of income or poverty – if you don’t let them.
-
-
-
-
Quote from dergon
Can it read an EKG with 98+% accuracy ?
When it does let me know. Then basic x-rays are 10-20 years away.
AI continues to be overhyped.
But I do think, properly used, it’s reached the point where it can help us the way breast CAD was supposed to help us.
-
-
-
What do you mean? We already train midlevels to do procedures in IR all the time. In fact, that is much more common than reading diagnostic studies, which are a lot harder to learn and requires more training. Which is why a software that sometimes recognizes a fruit and may give you ingredients does not mean it can replace one of the most cognitively difficult things a human being can learn. I look forward to the day it does replace doctors, because that means we live in a post-work utopia where I don’t have to slave away to pay for my Porche.
Quote from Lmfiondss
Quote from MidwestIR
Wow it can recognize fruits and yogurt. Board Certified material.
I think that’s a bit naive to sum this technology down to that. 5 years ago, nothing like this existed. But now think about what it would be able to do in 5 year. Maybe since your IR it doesn’t hit you the same way. But imagine your field. [b]If the radiology part is automated, why do we need to pay an IR guide 500k to do procedures when I could hire a PA, teach them basic IR skills and have a machine do the rest[/b]
-
Regardless of whether you are concerned or not about an AI algorithm replacing you in the near future, you are probably right.
-
Quote from MidwestIR
I look forward to the day it does replace doctors, because that means we live in a post-work utopia where I don’t have to slave away to pay for my Porche.
Exactly. I’m not freaking out because by the time my job is threatened, most jobs will be replaced by AI and we’ll either live a post-work utopia or an apocalpytic hell-hole run by the three quadrillionnaires that own those AI companies.-
im not freaking out because my career (and life for that matter) is more likely to be derailed by something i am not even thinking about – car accident, crime, neurodegenerative disease, etc. in the beginning i worried about teleradiology. then i worried about my hospital firing my group. then i worried about local competition. and do you know that the biggest hit to my job and income was? covid. a virus. something i would have never dreamed possible 5 years ago. pardon my rant, but AI is not the biggest threat to your career and life. you have no idea what it is but its out there. and you’ll manage like you always have. have a great weekend everyone.
-
-
CTxraydoc is right.
It’s usually the unanticipated threat that is your undoing (like a financial black swan or death, etc). Might as well just do your best, live a life you can be proud of, and let the chips fall how they may.
-
I personally know of multiple radiologists that have lossed jobs because of Ai
-
We already have radiologists who can describe an image with pain staking detail, but wont do a para or tumor board. Can Chat GPT do a para? That would be great.
-
Troll without a clue what a radiologist is and does.
Chat GPT can replace college professors, lawyers, authors, cpas, teachers, engineers, politicians, administrators, etc.
-
-
-
-
-
-
-
-
-
[link]https://archive.ph/KksJF#selection-851.297-851.396[/link]
In case you want to read the article that’s behind a paywall.
Recognizing what large objects are in images is a lot different than analyzing 100s of images for every single possible pathology. AI tools are essentially a 1 trick pony… or they will be at some point.
Based on the article, I would be more concerned if I was an non-procedural physician where there are already huge midlevel encroachment like EM, family med, internal med, ped’s, psych, etc. Epic is already everywhere. All epic has to do is implement chatgpt and we will have midlevel+AI replacing docs. Hospitals around me are already cutting back on EM docs in favor of midlevels since they realized they don’t need a doc to deal with a majority of EM complaints and they make way more money with midlevels. Not just because they are paid 1/3 the salary- but because they order a lot more tests. -
[link=https://www.cnbc.com/2023/03/16/microsoft-justifies-ais-usefully-wrong-answers.html]https://www.cnbc.com/2023/03/16/microsoft-justifies-ais-usefully-wrong-answers.html[/link]
Bing AI is based on chatGPT-4. Please let us know how does “usefully wrong” help Radiology image interpretation.
Quote:
“For instance, in February when Microsoft debuted its Bing chat tool, built using the GPT-4 technology created by Microsoft-backed OpenAI, people noticed that the tool was [link=https://www.cnbc.com/2023/02/14/microsoft-bing-ai-made-several-errors-in-launch-demo-last-week-.html]providing wrong answers[/link] during a demo related to financial earnings reports. Like other AI language tools, including similar software from [link=https://www.cnbc.com/quotes/GOOGL/]Google[/link], the Bing chat feature can [link=https://www.cnbc.com/2023/02/14/microsoft-bing-ai-made-several-errors-in-launch-demo-last-week-.html]occasionally present fake facts[/link] that users might believe to be the ground truth, a phenomenon that researchers call a hallucination.”-
-
Yea, but radiologists are humans.
AI will have to be essentially perfect, or, it won’t be trusted.
I fail to see how it will ever be ‘perfect,’ since it was made by humans.
Humans are more forgiving of other humans.
Not so much with software or robots, especially if a big error occurs due to poor programming or any ‘glitches.’
-
I’ve said it for years, I’ve been right for years, all you have to do is understand the society and follow the money:
It’s all marketing. Sell something for millions, exit, and then let THOSE people worry about what to do/not to do/legal issues, etc.
-
-
-
-
Because describing an image “in pain staking detail” is poor radiodiagnosis. There is an immensely sophisticated art in being a highly effective radiologist that AI is a long way off achieving it would seem.
-
By the time AI can do my job, it’ll also be able to replace the jobs of internists, dermatologists, radiation oncologists and nearly every other specialty in medicine (hell, even psychiatry). Not saying it won’t happen, but our field isn’t as simple as describing images, pattern recognition and matching it with what radiopedia has to say.
-
-
AI’s ability to understand images has progressed substantially in recent years and I suspect medical image interpretation is not far off. Though there will be political and medicolegal barriers to immediate clinical adoption, wise groups should begin thinking about how radiologists can add value in ways beyond image interpretation. Freaking out and denial are maladaptive responses.
-
Unknown Member
Deleted UserMarch 18, 2023 at 9:28 am
Quote from jazzercise
AI’s ability to understand images has progressed substantially in recent years and I suspect medical image interpretation is not far off. Though there will be political and medicolegal barriers to immediate clinical adoption, wise groups should begin thinking about how radiologists can add value in ways beyond image interpretation. Freaking out and denial are maladaptive responses.
^^^^
Absolutely.
As a preretirement old dude I imagine I should be the one poo pooing AI; but after all I’ve seen, I have profound respect for unbridled and unpredictable technological advancement. Heck, I was taught how to use a slide rule in high school with log charts. AI is going to cause huge changes to radiology; how radiologists adapt is the big question. It will also be faster than you think. 10 years seems like a long time, but in the terms of a career and family, it is nothing.
Look, a lot of folks have set up sweet remote deals for themselves and don’t want to consider the fact that what they have is temporary. The truth is, it’s all temporary. You are temporary. So keep your options open and enjoy the journey. Denial is only going to hurt your future.-
Give me an AI that can pick out differences between 4 tomo stacks and tell me which one is BC and I will kiss your feet.
-
Heck give me an AI that can dictate what I say rather than the high error rate of voice dictation.
As alluded to, who gets sued when its wrong (and it will be)?The AI creators, thats who. Its so easy for a software engineer to try to talk smack.
Software engineeryou ready to sit in court being grilled by an AI? At first there will still be human juriesbuckle up buttercup.
-
This will for sure transform aspects of the practice.
Like the Co-pilot for Microsoft. Now, imagine dictating – “compare nodules”. Pulmonary nodule comparison, or finding PE, are algorhitmically simple that the current AI already does relatively good at, a natural language interface that allows the machine to understand what you want it to do will truly be revolutionary.
That being said, more complex patterns of disease will be more challenging, but will still be achievable within the next 10 years or so.-
Groups should prepare a form letter to send to patients:
“Through no fault of your own, your study was going to be assigned to one of our terrible radiologists. Unfortunately, roughly 20% of our radiologists are not good at interpreting medical images. Some read as many cases as fast as possible, some never learned anything in residency, and some simply don’t care. Rather than allow that to happen, we have decided to allow AI to interpret your images instead.
While we have many radiologists that are far superior to AI, we have found that AI is equivalent (in fact, even superior!) to the bottom 20% of our group. Given our current staffing shortages, we apologize that we are unable to have a competent radiologist interpret your images.
Rather than allow one of our infamous hacks be involved, we have decided to let a computer try its best. We promise your care will not be adversely affected (remember, your case was already randomly chosen to be read by a terrible radiologist).
We hope you have enjoyed your experience with XXXX and wish you a happy and healthy year.”
-
I say keep one foot in the hospital. Keep up with procedures and mammo. On-site Rads who are comfortable in the department are unlikely to be easy to replace. The WFH full Tele probably vulnerable, but even that is a theoretical risk at this point.
-
Quote from Waduh Dong
I say keep one foot in the hospital. Keep up with procedures and mammo. On-site Rads who are comfortable in the department are unlikely to be easy to replace. The WFH full Tele probably vulnerable, but even that is a theoretical risk at this point.
I think that is true for *any* occurrence that somehow hurts the radiology job market, whether its AI or something else.
Remote full tele are probably the first on the chopping block if groups cut back. -
Nonsense. There’s no critical thinking in radiology. all memorization. Tree-in-bud finding = tuberculosis.
-
I have too many skills in radiology to be replaced. Mammo, needle work, multispeciality reading of all modalities, nucs, tumor boards, ED list crusher with accuracy, committees on top of committees, wrvus/year etc. Some of my colleagues could be kicked to the curbed by any type of AI as they barely perform the minimum requirements of being a radiologist.
Make as much money as you can now if you are one of those limited rads like SNAC. And he most definitely is making big money lol.
-
-
-
-
-
you’re using the wrong VR? Latest version of PS is excellent, at least when me and my colleagues use it.
-
Yes, you wont be replaced. But if the AI does 50% of your job, they will hire 50% less of you, not just give you 50% more time to decide what to do with yourself.
-
Unknown Member
Deleted UserMarch 22, 2023 at 5:15 pmArtificial intelligence (AI) has made significant strides in the field of medicine and healthcare, particularly in diagnostic imaging. Radiologists, who are specialists in medical imaging, are concerned that AI technology will replace them in the future. However, the answer to whether AI will replace radiologists is not a simple yes or no. There are several factors to consider, and this essay will explore these factors and their implications.
One of the main advantages of AI in medical imaging is its ability to process vast amounts of data quickly and accurately. AI algorithms can analyze images, detect anomalies, and highlight potential areas of concern, reducing the need for human intervention. This capability could significantly reduce the workload for radiologists, allowing them to focus on more complex cases that require their expertise.
Furthermore, AI can also learn from experience and improve its accuracy over time. As more medical images are analyzed by AI, the algorithm becomes more sophisticated and precise in identifying abnormalities. This technology has the potential to improve diagnostic accuracy and speed, which could save lives and improve patient outcomes.
However, despite its many advantages, AI has limitations that prevent it from entirely replacing radiologists. Firstly, AI algorithms can only analyze medical images and provide diagnostic suggestions based on the data they have been trained on. They lack the ability to interpret clinical data, patient history, and other vital information that radiologists consider when making a diagnosis. Additionally, AI algorithms are not yet capable of making complex clinical decisions, such as recommending a course of treatment or conducting interventional procedures.
Another limitation of AI is that it requires large amounts of high-quality data to be trained effectively. This data must be representative of the population it will be applied to, and it must be diverse enough to capture the variability in pathology that radiologists see in their day-to-day work. Obtaining such data is a significant challenge, particularly in medical imaging, where data privacy and security are critical issues.
Furthermore, radiologists play a crucial role in patient care beyond diagnostic imaging. They communicate with patients and their families, participate in multidisciplinary teams, and provide consultation and guidance to other healthcare professionals. These skills cannot be replaced by AI technology and are essential in ensuring that patients receive comprehensive and personalized care.
In conclusion, AI technology has the potential to transform medical imaging and improve diagnostic accuracy and speed. However, it is unlikely to replace radiologists entirely. Radiologists bring a wealth of knowledge and experience to patient care beyond diagnostic imaging, which is essential for optimal patient outcomes. AI can assist radiologists in their work and reduce their workload, allowing them to focus on more complex cases. Still, it cannot replace the human touch, empathy, and expertise that radiologists provide. Therefore, radiologists should embrace AI technology as a tool to improve their work rather than a threat to their profession.
-
Quote from Teedevil
Yes, you wont be replaced. But if the AI does 50% of your job, they will hire 50% less of you, not just give you 50% more time to decide what to do with yourself.
This still does not account for the suits/responsibility part. No one has explained that one yet.
-
-
-
-
-
-