We were unable to load Disqus. If you are a moderator please see our troubleshooting guide.
Now - can we get rid of ALL the MBA's, the Coder's, and every other NON-ESSENTIAL person in medicine today - because the problem with medicine are the cost overruns and the HUGE salaries from the C-Suite on down - it is NOT the doctors - get ride of the Lard and you will make Medicine affordable - and that means ZERO insurance - I have asked ChatGPT to give me a good outline to this - and guess what - the numbers in cost are Astounding!!!! Want to cure the Budget for the United States of America - there you go!!!
Here we have the "science" of medicine. I think that when I took the USMLE, the passing grade was higher than 60%. But it was a long time ago, I could be wrong.
What about the "art" of medicine? What about those flashes of intuition that lead us into the right place to make a diagnosis, even though the evidence may not all be there yet?
I love Star Trek's Lieutenant Commander Data - but even he had trouble with jokes, and spent his entire existence trying harder to be human.
Let's use computers for what they are really good for - collecting and analyzing massive amounts of data. Let's use humans for the compassion required to be a truly great physician - priest - counselor - CEO - businessman - sales clerk...
Congratulations AI ! But don’t we already have a Doctor Google?
60 % is a pass rate for physicians?? Not THAT'S a problem.
Score is actually fairly weak. The AI took an essentially open book test with access to databases. Most MDs would do better than 60% open book.
I'm presuming that the AI passing the medical licensing will only be allowed to do digital exams?? :)
That’s not really AI as much as an automated script scraping the web. This would be like using Google to look up answers rather then the AI knowing the answers de novo.
So what? These AI's don't "know" anything and can't make judgements. Any student could pass any test given access to 4 billion references and enough time to search them all. It doesn't mean that student knows anything or that they can apply what the know to a previously un-imagined situation.
There in lays the quandary of information vs knowledge. How do we teach in the information age? Perhaps every student will get an AI tutor that demands reasoned knowledge in depth and reduces teaching load and prepares for a human professor-processor personal in depth evaluation of each student.
This is our new colleague joining us today, Dr. HAL.
I for one welcome our new robot overlords
Well, 1 more step in the long-predicted robots taking us over.
It would be interesting to see the pass rate on the USMLE by NP's. I think I would prefer to have the AI treat me.
That’s very rude. There are excellent NPs and excellent physicians. And then there are poor NPs and physicians. You can’t judge an entire profession based on whatever experiences you have had with NPs. I’m sure many NPs can pass the USMLE. We also have to pass our boards to be licensed.
There is already research data on this done at Columbia. NPs scored as well or better than physicians. Be careful what you ask for.....
Check out this video on the DNP versus MD on the USMLE Step 3 exam. Results: "In 2008, the National Board of Medical Examiners offered down a simpler version of the USMLE Step 3, an examination all physicians take to receive a medical license. The pass rates for DNP candidates ranged from 33%-70%, and the experiment was discontinued in 2014 due to "low utilization."https://www.youtube.com/wat...
What if we look at this not as a ho-ho, gotcha moment or musing on how to get AI’s nose under the tent in medical education but flip the coin to see the side that already damns much of current medicine for its slavish submission to algorithms and other forms of what we used to deride as “cookbook medicine”? Do we really want more of this, not less, when we’re dealing c real people?
And while we’re at it, llet us remember that the state boards of registration in nursing cast a dim view (and a strict improvement plan) on any program that prepares only 80% of their grads to pass NCLEX, the national licensing exam. And, of course, nursing offers a more wholistic approach to caring than what the disease focus in medicine brings to the table.
And with the new version of the NCLEX coming out in April, I would just love to see any AI pass that exam! It's almost all critical thinking, and the "Select All that Apply" questions are now on steroids with about 10 options to weed through.
Wonder if the AI would be able to pass the bar exam. Just think no more lawyers. lol
I use it for basic legal questions for housing advocacy. I even got it to write a rough draft of a motion.
Podcast interview with Gary Marcus, brilliant AI expert.https://podcasts.apple.com/...
AI can deal with the medical test as it is a structured, unemotional, logical substrate. Enter the patients..... well, not so much.
AI indeed has an expanding role to play in medicine and triaging images (very structured data), the best use I have seen thus far.
I know plenty of unempathetic, cold, condescending physicians. I bet they could add some nice personality to the AI that would be a pleasure to interact with.
Exactly!For example asking basic questions - in most any language.
The problem with AI is affordability. At this time we can't afford to make it work.
Have you seen what doctors charge?
The problem with medical AI is structured data required to train the algorithms. And this is what drives the cost.
With time the cost will come down as we get better with prompt engineering.
I worked doing academic Neuropsychiatric research for 15 years. This would be so very helpful in deep searches into the literature. The work that took days for me to complete could be done instantly, and without chasing down references that turn out to be dead ends.
Deep searches would question the "Matthew's Effect" that allows prominent established scientist to usurp credit from lower ranked seminal innovators. I have done immunotherapy research for many decades and invented the first injectable (soluble Fc receptor) checkpoint drug half a century ago. When I file a patent I am required by law to cite the original related idea "conception" priority date. Those that invented insulin, cancer imaging and checkpoint immunotherapy and hold parents on the technologies are not the same as those credited in science. A "pioneer" invention may need considerable "Nobel" innovation to move to clinical use. The line between the "Matthew's Effect" and plagiarism is not well considered or cited in the science literature. The record of discovery is often obscure. I welcome AI suggesting references that can help me credit seminal innovators as well as those that add refinement to discovery for courtesy of science and legal patent citations.
I was thinking the same. Would've been great to help.with my annotated bibliography during my thesis.
ChatGPT's searches look great but will just as easily make up references as position real ones. It has no veridicality function. Google is safe for now. Try a few and you will quickly see how useless it would be to rely on the software.
Scientific people tell me AI makes up non-existent references.
Isn't that just a few lines of code and training? Scholarly work includes the creation of new knowledge, debunking invalid studies, and summarizing available science. When a human makes up a reference that's just fraud. IMHO, if AI is spontaneously that sketchy . . . we're in big trouble.
You must definitely verify, but it’s great for giving a synopsis and formatting citations.I paste the articles in that I have already found.
AI something wonderful and wicked this way comes. This is an AI crawler not even a toddler. Soon quantum computing will grant AI a million times the speed and power.Things to come of and beyond our dreams and imaging. Yet the AI gods may envy our human emotions and character and find a way to emulate, protect or destroy us? Humanity is running a technological gauntlet. AI, gain of function and global warning and warfare just some of the players. Let the games continue and may the odes be ever in our favor.
I love that movie genre as well. But seriously, I truly hope we humans can learn not to repeat the mistakes portrayed.
The AI "baby" has been born of many technological mothers and fathers and everyone will interact with this offspring. This is not a movie, AI "IS" a reality. Considering our human history our new technological inventions like our past splitting of the atom "Will" yield dreams, endless energy and nightmares thermonuclear war. Our inventions like ourselves have duality.
The New Chatbots Could Change the World. Can You Trust Them? Siri, Google Search, online marketing and your child’s homework will never be the same. Then there’s the misinformation problem. https://www.nytimes.com/202...
Good Article that scratches the surface of AI. Someone will figure out how to run serval programs in tandem and fact check each other. In future, maybe a decade or two, imagine a say six way conversation with a three humans and three AI programs representing constructs of historical "figures" like great US President all in life like 3D holographs OR in my case a few great scientists. The later might blow most dinner dates but you would have a vast selection of AI artificial-actor guests. Maybe Dion-Abraham, Martin & John (1968) https://www.youtube.com/wat...
So AI passed, though not with a great score. Did any of the AI programs match anywhere? (sarc)
Never call your own remarks "sarcastic," which refers to nasty people. Say u r speaking "facetiously."
The ability for AI to apply judgment. for example to intubate or not, will be the defining factor.
Will it ask for consent? If a patient is Full Code but clearly futile, what will it do? How about a DNR but organs are in short supply (always)? I cringe thinking of AI in Epic . . . 'yeah I know what you are trying to do . . . and I could help you out . . . but I'm just not feeling it today . . . so I'm going to let you get almost there and then take you for a frustrating keystroke ride.'
'Hello Dave..Can we talk? And No, I don't know why those other astronauts haven't woken-up yet. But Yes, my programs are working fine and everything seems to be going well with the rest of my systems, with the rest of my systems, with the rest of my systems'...With apologies to Stanley Kubrick ( And HAL 9000).
This finding may be upsetting to those who believe that USMILE scores, or scores on similar professional examinations, have a useful degree of predictive validity for the dependent variable of professional success. They do not. Many decades ago, a similar examination for licensing clinical psychologists was sometimes passed by individuals who did not possess an appropriate education, internship, etc. The solution was to require relevant degrees, experiences, and professional recommendations prior to sitting for the examination. The same transformation has occurred in the legal and other professions.
Whenever careful studies have been conducted, the data are pretty clear: test scores alone have unusually low predictive validity whereas successful in one's education and hands-on experiential components has reasonably high predictive validity.
No one thinks ChatGPT can remove an appendix and if you ask it deceptively simple questions but logically indirect questions such as by what route should one eat a hamburger, you are likely to get a bizarre answer. Even ChatGPT has difficulty responding accurately based on conceptual ground that a one year old has mastered.
One final observation which I hope will be invalid a year or two from now. ChatGPT is a very sophisticated sentence completion algorithm. It is not built on a core truth function in which wide range, relevance, truth, and implied principle drive the the core logic. In some circumstances it will create a syntactically eloquent but factually nonsensical response. It is tempting to think that this software development decision was a mistake.
ChatGPT is already an issue in higher education (and I'd suspect high school as well). Students relying on it to write their papers for them is increasing. Lots of chatter about that in higher ed. It is likely this "controversy" will increase in medicine as well.
You don't think we can integrate AI with a DaVinci robot to do an appy? Think again.
USMLE includes photos that you have to evaluate. How does a language model answer those questions?