The litigation issue is going to be the big thing that keeps doctors employed. They're going to have to verify everything and sign off on it. Not like you can give a computer a DEA#

All it’ll take is one person to die or get severely injured due to some robot going full retard, that’ll be the end of that. And we know it’ll happen. There is no way there will ever be full implementation of AI to the extent where physician jobs are in jeopardy. We already have robots being used for surgery, but guess who is running the robot?

I never understood how doctors can pass through tertiary education with 51% score.

brb. kill 49/100 people who walk into my office.

Okay, as someone involved in the field i'm gonna try to explain to the OP more in simple terms, because you seem to have a clear misunderstanding of the future of A.I and how it actually works.

The programmer has no need to understand what he wants the computer to do, he simply needs to understand machine learning algorithms. There's plenty of different ones out there like Artificial neural networks, genetic algorithms etc, but they all work in a similar way.

You need a few things for machine learning to be effective; you need lots of data, you need time, and you need processing power, that's pretty much it.

What the programmer is writing, is simply an algorithm that teaches the computer HOW to learn (i.e, how to go over the data and converge on a 'solution'), which is why the programmer doesn't need any knowledge in the field to program an effective A.I using a machine learning algorithm. For any semi advanced programmer, these algorithms are really piss easy to write - i've done a few myself for simpler things (predicting weather, financial markets etc)... they barely take more than a few days. They aren't very dissimilar from how our actual brains work, and in fact as a doctor who has some understanding of how neurones work, you should know our brain is basically binary (i.e, neurones reach their threshold value and either fire.. i.e "1", or they don't reach their threshold value and don't fire, i.e "0"), our brain is basically a massive network of neurones that operate in binary, so why is it so hard to believe computers can achieve what the human brain can when it's focused on specific tasks?

The bottleneck is time, and processing power. Even when i was at university in 2012, using a clustered network of 100 home PC's to solve complex differential equations, my algorithm took a long time to solve these, we're talking a day or two of constant running.. but if you say doubled the power, it'd take half the time. Obviously big companies have access to far more power than this. I've seen people mentioning moore's law, and i'm not sure if you fully understand this, moore's law is essentially a relationship between the rate of increase in transistors (doubles every 2 years), and it's relationship with computing power. However, moore's law cannot be extrapolated indefinitely linearly, as once transistors become as small as they can be the law no longer applies (the size of atoms if i recall), and moores law is empirical and not actually scientific. However, with the birth of quantum computing it has huge potential to smash this limitation and propel machine learning forward tenfold in a very short period of time.

If you want evidence for this, all you have to do is dig. There's tons of machine learning projects out there doing magnificent things... there's machine learning algorithms that have learned to translate chinese on the fly as someones speaking, there's algorithms that can determine precusors of cancer better than doctors, there's machine learning algorithm's that can hear better than humans, see better than humans (although not both at once yet as far as i'm aware)... there's literally guys who've started medical companies and made breakthroughs, all whilst having no knowledge what-so-ever of medicine.. just data, computing power and the ability to write machine learning algorithm's. I myself, for my dissertation wrote a genetic algorithm (a form of machine learning using evolution/natural selection as it's base strategy) to read event related potentials from EEG graphs and remove the noise, and was used to predict the onset of epileptic seizures... whilst it's not EKG's it's not a million miles away and this was 8 years ago when i was a humble student.. it was very effective and i had 0 medical background apart from one neurobiology class i took in my final year at university.

When you consider we're in a very very early stage of machine learning, and when you also consider as someone mentioned above it doesn't increase linearly but more like 1, 2, ,5, 20, 100, 500 etc then it gives you an idea of just how fast progress can skyrocket once we reach certain stages, and once you throw quantum computing into this the possibilities are endless. Machine learning is so young, old forms of 'A.I' were different, but machine learning is really the future.

There's certainly a lot of evidence that machine learning, especially with quantum computing has the potential to do most jobs in the not so distant future (next 20-50 years or so)... and not even doctors are safe, but they won't be the first to go either. I would figure there will be many years of machine learning algorithm's aiding doctors in their work, which may mean you see a decrease in doctors required, rather than doctors being removed entirely, or machine learning will start with the more menial jobs (i.e walk in GP for colds and stuff) and slowly take on more responsibility over time. Even as a software engineer i'm not safe either.... it's certainly not far fetched for machine learning algorithm's to learn to write machine learning algorithm's, and in fact is completely feasible.

Regarding litigation, i don't know or care, i'm just pointing out the technical implications.

This video may help you to understand a little more and it's fascinating, but it's already 5 years old and this stuff moves at a very fast rate, so even the stuff in this video is 'old news' in the machine learning world:


There's no need to be so insecure, everyone is losing their jobs to A.I in the future.. not just doctors and when it does society will have to go through some huge reform.. we're not there yet so relax. Just hope it doesn't happen till you're dead/retired.

IMO AI will eventually outdo humans at being a doctor. Not quite there yet, but its coming

https://science.slashdot.org/story/2...n----sometimes
https://science.slashdot.org/story/1...n-radiologists
https://science.slashdot.org/story/1...er-study-finds
https://science.slashdot.org/story/1...-a-pathologist

That’s until they can scan you with a 80+% accuracy rate and diagnose you on the spot. Then a robot will perform your surgery, prescribe your medicine and humans will just do menial work like providing comfort for the patient, until humanoid robots come around.

There is no stopping what’s coming.

I disagree; there are a lot of aspects of the job that are amenable to AI intervention. It might seem like a lot of consultants/attendings have some sort of incalcuable wisdom that guides their decision making but if you ask them to lay down their thought process a lot of it is pattern recognition. Something which AI excels at; Not to we have enormous mountains of medical data now.

There are certainly aspect of the job that are irreplacable: Patient communication, Interventions, charting and discussions with families etc but to think the field is immune to AI is arrogant. Everyone thinks their job is irreplacable and that they're special. The machines are coming; stay safe.

"They'll never be able to build a machine that can do my job as good as a human oh fuk they did."

~Like 30 different industries.

People itt who say doctors don't know what AI is/does/can do, are ironically unaware of what doctors do.

Let's start with the start. A patient enters with shoulder pain. You want to figure out whether it's a torn tendon, a fracture, a bruise, or just someone who is trying to score painkillers. How will AI figure this out? It will need information about the patient. Where will this information come from? From someone who does a physical examination. A doctor. Who is then going to input this information into the AI? The doctor, using a keyboard + mouse. I can already tell you this consultation will take 2-3x as long because while putting in all this information, it's already in the doctor's head, along with a differential diagnosis and various treatment plans, which he will talk through with the patient.
You could argue that eventually speech recognition software will make this easier. I don't see any doctor saying something out loud like 'Visually the shoulder looks perfectly okay, this and this test were positive, these tests were negative, however I think this guy is faking it to score painkillers' while the patient is sitting there.

You could argue AI will eventually be better at evaluating X-rays, CT and MRI scans, and blood samples. I can agree with that. However scans are known not to correspond to the real life situation: someone with a perfect back on a scan can score 10/10 on a pain scale, while someone with a horrible looking back (on the scan) can score 2/10.

Secondly, a doctor makes diagnoses with 98% accuracy, and the AI with 99,9% accuracy, but the AI is 10-100x as expensive because hospitals need to buy computing power and time from multinationals like Google with supercomputers and patented algorithms.

Thirdly, interestingly nobody has reacted to my points about psychiatrists with paranoid patients, and patient advocacy groups.

And lastly, while typing this, my computer crashed twice, forcing me to retype this whole post 2 times. How will we deal with technology failure? How will we identify a bug in a software that overdiagnoses or underdiagnoses certain illnesses and/or prescribes wrong information? We will have no idea how AI comes to certain conclusions, and no way to verify whether it's conclusions are correct until it's too late.

There seems to be a trend itt that radiology, pathology, pathology, and surgery. Conveniently those are all the ‘black and white’ specialities. Again, there is a reason why the radiologist does not treat the actual patient in the ER for example.

There are a number of examples that are complex that have been mentioned itt with no responses from the ai guys.

I'm from the industrial automation sector. Your last point about your computer crashing.. you're using a commercial OS not fit for industrial use. The AI system for use by the medical practicioner can be run on a cloud resource (dedicated servers), or ruggedized locally by using IPCs (industrial PCs) with a certain SIL (safety integrity level) factor which can also be further ruggedized by linking it to a parallel redundant computer as well. The OS on these IPCs will likely be some reduced instruction set, or optimized version of Linux or Windows Embedded. Ensuring high availability of a computing resource isn't going to be what stops automation. Most of the worlds manufacturing process is already being run on some sort of PLC or IPC without problems, and its probably not even redundant. The Medical industry would just require higher availability and a company like Siemens could easily develop a product for them..

Otherwise I agree with your first paragraph. The Doctor will be present in some form or another - inputting data etc. It might just free up a lot of his time and offload a lot of mental computation that he has to do. Allowing him to just quickly scan the results of the AI system and put his stamp of approval on it.. allowing him to see more patients in a day.

The IT folks have 0 understanding of medicine. They think doctors just treat rashes and coughs in the clinic and that's the extent of their interaction with medicine. Just lol @ any complex floor patient/icu patient or surgical patient being managed by a robot.
Human error is forgiven. Technological error is not forgiven.
Idk man, maybe cause no doctor passes anything with 51%? We take several board exams that are 9 hours in length each and have to do exceptionally well just to pass. On top of several dozen exams where you need ~75% just to pass and be at the bottom of your class.
And that's after you got 90s all through hard undergrad classes for 4 years - just to get into med school.

I fully understand that about AI and I know it's not this algorithm that the programmer has to design. But time is the first issue. Just back in 2015, some silicon valley guys said AI will displace tons of jobs before 2020. Nothing happened though.

You need to first develop this technology. Then you need to implement it within the industry and show its success AND safety. Then you need to optimize its price, good luck with that. And then you need to implement it across the country, with a reasonable price, also good luck with that. And multiply that by the coefficient of human acceptance.

It took like 2 decades just to accept eating sushi and another decade to go mainstream.Lol @ the idea of someone going to the operating room because a robot is 80% sure.

Yeah but I do see robots doing basic **** like general practitioners, most non surgeons.

I’ve been misdiagnosed a lot for my pain symptoms.

I think you’re thinking of simple Diagnosis.

You'll never truly understand how disgustingly sick and unhealthy people are until you spend some time working in a hospital. I was a ED PCT for a brief period of time and it was mindblowing. We had one lady who came in, I forget what her chief complaint was (think it might have been a COPD flare-up) but she had literally been laying in her own feces for days before her husband brought her into the hospital and had feces under all her fingernails because she was cleaning herself by shoveling it out of her ass with her hands. This lady was A/O x4 also, she just legit chose not to clean herself properly.

This is my problem with free healthcare for all. While I'm sympathetic, you have a population of people who have zero regard for their own health on a bazillion different medications constantly coming into the ER who have ZERO intention of changing their behaviors. I don't want to foot the bill for these people.


Edit: And I completely agree with the OP, we need humans practicing medicine, AI won't suffice.

this not sushi. the need for medical innovation in america is overdetermined by exorbitant medical costs, accelerating tech advancement, and political relevance.

to believe that the clusterfuk that is the medical industry will continue in this way is tantamount to a rejection of capitalist progress.

Going to the moon was literal science-fiction until we fukking actually did it.

Your cellphone has enough power to almost run a third world country. Cellphones 20 years ago were actual bricks. Just imagine the amount of progress we will make in 20 years. Your cellphone already has AI, and can navigate certain tasks. Apply that AI to a mult-million dollar medical surgical machine and it will eventually learn what to do.

All medical innovation within this context in the past 20 years has slowed down doctors and increased costs and led to us seeing fewer patients.Once planes were invented, going to the moon was not science fiction. We also had an understanding of the solar system to the level of warranting travel. Your notion of scanning the body in 2020 is indeed pure science fiction.

funny how medicine people have 0 understanding of IT. My understanding of the complexities of medicine > your understanding of the complexities of AI. Medicine is complex. Plenty of things are complex. AI can account for all the factors of complexity you're concerned aboutgreat. so you acknowledge that it's possible, you just think that it will take longer than i do. I can end the argument at that. Based on knowledge of the AI industry and following advancements made over the last decade, I think AI will be capable of replacing more or less everything within my lifetime. You don't. That's fine.


and i'd guess that the AI will be 99%+ sure for it to be actually used

Lol at thinking AI can’t think outside the box.

Primary care DRs are the only ones that really need to worry. Outside of that i doubt specialized DRs need to worry. No way in hell would I let a robot do a c section on my wife or complexed procedure on my child.

Edit: also I’m pretty sure something’s gonna ruin western society before AI really becomes a threat to these jobs.

I actually beg to differ on this as well. I seriously question whether AI can do many of the components of good medical practice.

Beneficence, maleficence, justice, ethics, consent, autonomy, costs, confidentially, contraception for a 15 year old etc etc.

treating patients as a whole - presenting complaint, home situation, work life, psychological state, physical issues, polypharmacy, their families, palliative care.

I sure as hell know that youre just stressed and thats why you have headaches for example. But you need to reassure the patient and all that ‘human’ stuff.

Maybe im wrong but who knows

That's hardly an argument. There would have to be some real world application/ evidence, some kind of progress for anyone to believe that we are going to see this within the century. Lol@ the guy somewhere above stating how machine learning is "phasing out" radiologists. Not for 3 decades atleast, even then radiologist's training will be altered to let them optimally utilize these techniques instead of 'phasing them out'. It needs a master, always will.

Besides, computers or IT has contributed maybe 1% to any advances/ improvements in surgery in the past century, almost negligible. A whipple's pancreaticoduodenectomy is still done practically the exact same way as in 1935. FFS even a leg amputation. Let that sink in.

Sorry to burst your bubbles, not happening in this lifetime boyos. Not in the real world atleast.

There's a lot more to medicine than procedures and primary care clinic lmao.There is no radiology AI on the market. Even path doesn't have that

I’d rather have a toaster diagnose me than op

What a deluded insufferable kunt

they won't get rid of human doctors, but that field will be cut in half at the very least in the next 25 years when AI is good enough to diagnose 75% of issues. think about it, AI can compare, reference, and examine millions of case studies, symptoms, and medication interactions within seconds. they would only need the human to approve the diagnosis and they can do that remotely.

Deluded insufferable kunts make the best doctors though. The most heroic and dexterous surgeons I saw in my hospital would usually be described this way by everyone else. But when **** hit the fan, these were the people taking care of business while others were literally cowering under tables.'Diagnosis' is overrated. It's not exactly hard to do any of this right now. For example, when medications interact the wrong way, its hardly ever because the physician did not consider it, its mostly because the patient didn't mention it, wasn't complying with it, was self-medicating for years or some such.

Strong reading comprehension.

The cutoff to get rid of the human factor when diagnosing will most likely be above 80% accuracy.

Please tell me how doctors diagnose with almost 100% accuracy today.

Like in any profession there’s the good and the bad, and that oh boy have I seen sorry excuses for a doctor in my lifetime. Patients often walk in to get seen by a phycisian that rarely palpates them, assess them physically and is quick to write a generic cookie cutter prescription, all in under 5 mins.

A machine can and will do a better job, is nothing personal.

There is a lot that has to take place in order to phase out the entire medical field and you are lucky that more than likely you will not live to see it, but it will happen, because the technology will be there and $$$ rules.

Bumping this. AI imo would need to be completely human to deal with the mentioned above, which is is very very far from doing at this time. Brb diagnose an xray better than a human sure but this chit above? No chance