Change my mind: Medicine can be automated in 10 years

[Thesimplelifeofamyloid]

Recently, there's been a lot talk about Andrew Yang and his policies regarding physicians on the Reddit form. He says that computers could easily automate all of medicine, and his supporters state all we do as Drs is memorize symptoms and do the basic procedures over again, I didn't think so at first, but then I had a friend on IT tell me like 10 reasons otherwise. Do any wise physicians/students have good counter arguments for this?

---

Members don't see this ad.

 

[Schwifty]

 

[AMEHigh]

So you’re saying robots would talk to patients and their families, discuss complex medical issues including life and death situations and help them make decisions for their care with the appropriate emotions and understanding?

 

[Thesimplelifeofamyloid]

So you’re saying robots would talk to patients and their families, discuss complex medical issues including life and death situations and help them make decisions for their care with the appropriate emotions and understanding?

I guess his idea is that you would higher mid-levels that would be able to do that part, which really grinds my gears. He thinks they're more cost effective

 

[watermelon master]

nice try Andrew Yang's campaign staff

 

[sella.turcica]

Recently, there's been a lot talk about Andrew Yang and his policies regarding physicians on the Reddit form. He says that computers could easily automate all of medicine, and his supporters state all we do as Drs is memorize symptoms and do the basic procedures over again, I didn't think so at first, but then I had a friend on IT tell me like 10 reasons otherwise. Do any wise physicians/students have good counter arguments for this?

Ok, so if you feel like ass and eventually end up in the ICU, you would appreciate it if robots treated you? For real?

Also, just for the record, only people who know nothing about medicine (or couldn't make it into med school) talk **** about doctors. Medicine is NOT memorization only. Only a fool would think that.

 

[longhaul3]

So someone is arguing that in 10 years robots will be resecting complex skull base tumors without a surgeon?

In 10 years robots will be doing cesareans on awake patients and handing off the baby to the NICU robots for resuscitation? All without a doctor present?

What am I missing?

 

[FlowRate]

Robots are good at ideal circumstances: input physical exam, HPI, imaging, labs and get a diagnosis (or a probability set of diagnoses, or maybe a set of final candidates in the last layer of a NN.)

But how are they going to do an expert physical exam, elicit subtle history details from an anxious/reluctant patient, or engage in shared decision-making? Who will do the surgery? Who will discuss the clinical context with the ordering team (radiology)?

Why AI will not replace radiologists

In late 2016 Prof Geoffrey Hinton, the godfather of neural networks, said that it’s “quite obvious that we should stop training…

towardsdatascience.com

People think AI is going to take our jobs and yet look at how much every EMR, including cutting-edge research EMR's with baked-in AI, completely suck.

 

[TakotsuboOkazaki]

I guess his idea is that you would higher mid-levels that would be able to do that part, which really grinds my gears. He thinks they're more cost effective

A higher mid-level? So like a top level? So like a physician?

We already have robots for grammar. Use them.

 

[NickNaylor]

I think that "automation" in medicine is inevitable, I just don't think it's going to happen in the near future. I think we'll simply see more and more "clinical decision support" tools that are actually useful become available.

 

[AMEHigh]

I guess his idea is that you would higher mid-levels that would be able to do that part, which really grinds my gears. He thinks they're more cost effective

Well without getting in to what a "higher mid-level" is, your friend explained exactly why all of medicine could never be automated. Robots can’t do every single job in medicine and it’s completely ridiculous to think it is possible.

Cameras are already horrible at facial recognition of non-white people, I highly doubt cameras/robots will be able to visibly recognize if a patient is ill when they’re sitting in a doctor’s office in 10 years.

I would personally love for robots to call and argue with insurance companies and pharmacies though!

 

[electropartyogram]

Computers and AI can make note-taking and charting easier, which would mean more patient time/more patients, less administration in hospitals and insurance companies. Physicians will still be in control of procedures and decision making (hence why the Air Force still trains pilots). It would hopefully mean more income to physicians since there will be less administrative overhead.

 

[Thesimplelifeofamyloid]

A higher mid-level? So like a top level? So like a physician?

We already have robots for grammar. Use them.

*Hire

 

[Mass Effect]

Recently, there's been a lot talk about Andrew Yang and his policies regarding physicians on the Reddit form. He says that computers could easily automate all of medicine, and his supporters state all we do as Drs is memorize symptoms and do the basic procedures over again, I didn't think so at first, but then I had a friend on IT tell me like 10 reasons otherwise. Do any wise physicians/students have good counter arguments for this?

Anyone who believes this doesn't really understand what the job of a physician is.

 

[Mass Effect]

So someone is arguing that in 10 years robots will be resecting complex skull base tumors without a surgeon?

In 10 years robots will be doing cesareans on awake patients and handing off the baby to the NICU robots for resuscitation? All without a doctor present?

What am I missing?

Logic.

 

[jhamaican]

I would pay good money to watch an automated computer try to take care of patient with irritable bowel syndrome

 

[Chibucks15]

This website is wild. I’d say less than 5% of “for sure predictions in under 10 years” that happen on here ever actually come true. If they did, every specialty would be getting paid pennies and robots and midlevels would rule the world

 

[mmchick]

Right... so how does Yang suppose to round up all the physicians that will take part in creating these dig-their-own-grave "replacement" AI? Obviously in this dystopian idea, AI would have to be "taught" medicine or at least programmed for the "remote memorization" that takes place. Whose first to sign up for that Yang-Gang dream happen while in the same breath hearing that they're replaceable by mid-levels and robots.

 

[libertyyne]

Right... so how does Yang suppose to round up all the physicians that will take part in creating these dig-their-own-grave "replacement" AI? Obviously in this dystopian idea, AI would have to be "taught" medicine or at least programmed for the "remote memorization" that takes place. Whose first to sign up for that Yang-Gang dream happen while in the same breath hearing that they're replaceable by mid-levels and robots.

Don't need physicians for that ,just reams and reams of data. You know the kind of box clicking that physicians do.


There is going to be a need for a person to interact with patients to share options with them, create realistic plans with them.

Even if a diagnosis is made easily by AI. The cultural , economic and personal factors that go along with management are not something an AI can accomplish in an efficient fashion. Not is it going to be able to correctly answer novel questions or situations where data is sparse.
Still going to need a human to interface.

 

[Medic741]

Doctor on Video Screen Told a Man He Was Near Death, Leaving Relatives Aghast (Published 2019)

A 78-year-old man with lung disease was told by a doctor on a streaming video that he did not have long to live, a strategy that his family saw as insensitive.

www.google.com

Not gonna happen.

 

[Lucca]

I could see a near future where NLP is good enough to significantly improve efficiency/documentation in the clinic. Anyone seeing the pt wears a mic. It’s turned on during pt encounter. Human asks questions / makes observations. Mic records. Robot takes that audio data, parses for keywords and generates the note. Backend AI assists with using that data to bring up a probabilistic differential. Doc makes edits as needed/corrects any errors and uses AI info to make final decision on what to do.

I also think AI integrated with imaging technology will assist rads and path in reading slides/images. That’s already happening. In the farther future AI will be integrated with live imaging from scopes and be able to assist in pointing out features of interest on video reducing human error during procedures or in deciding treatment. This is technology that is already being developed in labs so I’m fairly confident it’ll exist in some commercial form by the time my generation is attending.

I don’t think any of it will present an existential threat to physicians but it will change the way people train and practice.

I do foresee this tech creating new ethical and organizational challenges for healthcare, for example by necessitating the creation of bodies or other technology capable of monitoring the use of AI assistants to ensure they aren’t being gamed to over-Bill or over-use resources

 

[forumname1]

Who gets sued when a robot makes a mistake?

 

[efikey]

maybe in 2050

 

[Dro133]

Who gets sued when a robot makes a mistake?

I have a genuine question: who gets sued when a physician makes a mistake? Is the hospital liable for damages or is just the physician? Or both?

 

[electropartyogram]

I have a genuine question: who gets sued when a physician makes a mistake? Is the hospital liable for damages or is just the physician? Or both?

Usually, it is the physician(s) who get sued if they are performing an operation, but it can depend. Nurses have been sued for gross negligence/intent to harm. Different states have different laws regarding the process to sue a physician (tort reform).

 

[pinkowl250]

I just finished my first week of med school and we had an intro lecture on how to take a patient history. Our lecturer mentioned that it's an art that computers will never be able to master since patients can often give illogical, irrational, and jumbled stories that would make any sort of AI explode. So while I do see AI assisting a lot in the future, especially in path and rads, it will never fully eliminate the need for physicians in terms of doing a basic patient interview. Medicine is more than just "you have symptom X so you must have diagnosis Y"

 

[sliceofbread136]

Been waiting years for the ekg reading machines to not be total crap. When that happens maybe I’ll start thinking this could happen in our lifetime

 

[Hippocrates II]

Right... so how does Yang suppose to round up all the physicians that will take part in creating these dig-their-own-grave "replacement" AI? Obviously in this dystopian idea, AI would have to be "taught" medicine or at least programmed for the "remote memorization" that takes place. Whose first to sign up for that Yang-Gang dream happen while in the same breath hearing that they're replaceable by mid-levels and robots.

Well, funny enough, you can bet on the fact that there are certainly radiologists and pathologists doing exactly that right now in AI automation over labs. Never underestimate human greed - you would be crazy to think there isn't a doctor out there that wouldn't take on a job like this, spelling out the eventual decline of his/her profession, so that they can turn their own profit in the short-term.

 

[deleted407021]

Lol this STEM bro assumes he knows something about medicine and healthcare

 

[Thesimplelifeofamyloid]

Been waiting years for the ekg reading machines to not be total crap. When that happens maybe I’ll start thinking this could happen in our lifetime

He says the physician healthcare lobbying was too strong and resists innovation and advancement.

 

[raiderette]

Ob/gyn isn't best left to mid-level and computers. When things go wrong, AI can't help. We will need surgeons, we will need people who understand the big picture.

 

[Dr G Oogle]

I conduct a lot of research on AI and the idea that medicine can be automated only demonstrates that the people who are stating do not know what they are talking about. Superficially it’s very easy to see why people are saying this as doctors we take factual information and use to synthesize a diagnosis and treatment plan sort of like 1+1=2 like intuition, past experience, what is currently on in your life to make one think of a particular treatment and diagnosis in the first place, different etiologies Of a particular group of diseases, different presentations in different ethnicities, races, etc. etc etc. in effect to automate medicine you’d have to be able to automate human intelligence in general and right now these guys are excited that AI can reliable differentiate a dog from a cat. It’s sort of like saying my 2 year old can hold a pencil pretty well maybe in 10 years she’ll be able to hold a scalpel and use it to perform a CABG.

On the other hand AI can help with some aspect of medicine. I’ll use obgyn as an example. We are currently working on a AI that will help identify whether or not patients who come in in preterm labor end up delivering prior to 37 weeks. Right now our model is about 85% accurate in predicting who won’t deliver, but you can be your behind not a single hospital or doctor will adopt this thing until it can be demonstrated to be of much higher accuracy in every conceivable combination of patient characteristics, basically it can’t just predict stuff in white people which is what most AIs right now are doing because the data they were trained on come from a predominant white population and that ain’t gonna work so well in the Bronx. So we have a decent model but I’d be it’s at least a decade before even a reasonable clinical application if not longer.

So fear not pathologists, radiologists, obstetricians, EKG machines and anyone else who spends too much time on tech blogs. Your jobs are safe.

 

[Tenk]

Will the hospitalist AI try to block the ED AI admission that was sent in by the FP AI’s office because they were at the AI golf course and didn’t have clinic that day?

Meanwhile the surgeon AI will install GodMode.exe and format the hard drive of the anesthesia AI if the patient moves even a nanometer.

 

[zanzizic]

If Andrew Yang really believes this then he’s way dumber that I thought he was

 

[zanzizic]

I just finished my first week of med school and we had an intro lecture on how to take a patient history. Our lecturer mentioned that it's an art that computers will never be able to master since patients can often give illogical, irrational, and jumbled stories that would make any sort of AI explode. So while I do see AI assisting a lot in the future, especially in path and rads, it will never fully eliminate the need for physicians in terms of doing a basic patient interview. Medicine is more than just "you have symptom X so you must have diagnosis Y"

First week of med school and you’re smarter than Andrew Yang

 

[JBOB]

Ask again once our new monitors don't read "VTACH" every time a pt moves, then we can talk.

 

[libertyyne]

I conduct a lot of research on AI and the idea that medicine can be automated only demonstrates that the people who are stating do not know what they are talking about. Superficially it’s very easy to see why people are saying this as doctors we take factual information and use to synthesize a diagnosis and treatment plan sort of like 1+1=2 like intuition, past experience, what is currently on in your life to make one think of a particular treatment and diagnosis in the first place, different etiologies Of a particular group of diseases, different presentations in different ethnicities, races, etc. etc etc. in effect to automate medicine you’d have to be able to automate human intelligence in general and right now these guys are excited that AI can reliable differentiate a dog from a cat. It’s sort of like saying my 2 year old can hold a pencil pretty well maybe in 10 years she’ll be able to hold a scalpel and use it to perform a CABG.

On the other hand AI can help with some aspect of medicine. I’ll use obgyn as an example. We are currently working on a AI that will help identify whether or not patients who come in in preterm labor end up delivering prior to 37 weeks. Right now our model is about 85% accurate in predicting who won’t deliver, but you can be your behind not a single hospital or doctor will adopt this thing until it can be demonstrated to be of much higher accuracy in every conceivable combination of patient characteristics, basically it can’t just predict stuff in white people which is what most AIs right now are doing because the data they were trained on come from a predominant white population and that ain’t gonna work so well in the Bronx. So we have a decent model but I’d be it’s at least a decade before even a reasonable clinical application if not longer.

So fear not pathologists, radiologists, obstetricians, EKG machines and anyone else who spends too much time on tech blogs. Your jobs are safe.

Could you share what your level of involvement is ? Are you a coder? Are you using hardware designed for AI number crunching or off the shelf ? Is your PI a computer scientist or a physican?

I ask these questions because I am quite surprised by your take. Considering AI is getting closer to doing a complex task such as driving a car in public roads under real conditions, or there have been a few papers where it has been great at small tasks like reading an x-ray for pneumonia . Even the facial recognition critic you make is not very compelling since it is a problem with data to train not a problem with the intrinsic technology.

 

[forumname1]

If Andrew Yang really believes this then he’s way dumber that I thought he was

How smart can a man be when his solution to every problem America faces is "$1,000 a month will fix that". That wouldn't even cover a garage in California while it would be adequate living in other parts of the country.

 

[libertyyne]

How smart can a man be when his solution to every problem America faces is "$1,000 a month will fix that". That wouldn't even cover a garage in California while it would be adequate living in other parts of the country.

Basic income has been successful in small scale trials. Who knows how beneficial it will be on large scale. Doesnt mean its not a an idea worthy of a trial.

 

[Dr G Oogle]

Could you share what your level of involvement is ? Are you a coder? Are you using hardware designed for AI number crunching or off the shelf ? Is your PI a computer scientist or a physican?

I ask these questions because I am quite surprised by your take. Considering AI is getting closer to doing a complex task such as driving a car in public roads under real conditions, or there have been a few papers where it has been great at small tasks like reading an x-ray for pneumonia . Even the facial recognition critic you make is not very compelling since it is a problem with data to train not a problem with the intrinsic technology.

I do some of the coding and my background is in computer science prior to medicine but since I am also a full time surgeon I don’t have time to code everything (truthfully most things) I work with the eecs dept at my university (a well know engineering school). I also design the dataset parameter and prospective testing. I never said anything about limitations of tech vs data, but perhaps I should have clarified it is not a tech problem but a data problem. Most of the data used to train AI is incredibly limited, there was even a price on npr about Steve Hsu (not sure if first name correct, don’t care enough to confirm) who has an AI based on loads of genetics data that he claims can predict disease risk of embryos that can guide seelection of which one to implant. Here’s the problem though the genetic database (1.1 million people) is mostly white and and woefully missing the 2 largest populations Indian and Chinese (and all of the genetic milieu within those groups) so his AI is incredibly limited. And that’s where the problem lies, as you pointed out, data. You could have the worlds greatest AI that can detect something with a high degree of certainty, but only in upper middle class white people or only in black people or only in blah blah people. Unless you create something that covers the billions of people in the world or even just the highly disparate group in your immediate hospital area where you wish to apply AI, the model won’t work. Not saying this is prohibitive, and won’t ever happen but people must realize that the data is what’s limiting us.

Your example of autonomous driving couldn’t be further from what medicine is. All you have to do is program a car with gps and teach it not to run into things and boom you’ve got a self driving car, to be frank I’m surprised it’s taken this long and I bet we are still at least 10 years away from something resembling decent market penetration (I swore not to spend my attending income frivolously but I’ll be in line for version 1.0 for sure).

I don’t think we are as close to automating medicine as we are to driving and when we as a society are able to automate what doctors do on a global or even national level , well all of humanity better figure out some hobbies because every other possible job out there will also be just as obsolete. I

 

[libertyyne]

I do some of the coding and my background is in computer science prior to medicine but since I am also a full time surgeon I don’t have time to code everything (truthfully most things) I work with the eecs dept at my university (a well know engineering school). I also design the dataset parameter and prospective testing. I never said anything about limitations of tech vs data, but perhaps I should have clarified it is not a tech problem but a data problem. Most of the data used to train AI is incredibly limited, there was even a price on npr about Steve Hsu (not sure if first name correct, don’t care enough to confirm) who has an AI based on loads of genetics data that he claims can predict disease risk of embryos that can guide seelection of which one to implant. Here’s the problem though the genetic database (1.1 million people) is mostly white and and woefully missing the 2 largest populations Indian and Chinese (and all of the genetic milieu within those groups) so his AI is incredibly limited. And that’s where the problem lies, as you pointed out, data. You could have the worlds greatest AI that can detect something with a high degree of certainty, but only in upper middle class white people or only in black people or only in blah blah people. Unless you create something that covers the billions of people in the world or even just the highly disparate group in your immediate hospital area where you wish to apply AI, the model won’t work. Not saying this is prohibitive, and won’t ever happen but people must realize that the data is what’s limiting us.

Your example of autonomous driving couldn’t be further from what medicine is. All you have to do is program a car with gps and teach it not to run into things and boom you’ve got a self driving car, to be frank I’m surprised it’s taken this long and I bet we are still at least 10 years away from something resembling decent market penetration (I swore not to spend my attending income frivolously but I’ll be in line for version 1.0 for sure).

I don’t think we are as close to automating medicine as we are to driving and when we as a society are able to automate what doctors do on a global or even national level , well all of humanity better figure out some hobbies because every other possible job out there will also be just as obsolete. I

Thanks for the clarification. I think we agree on many points. I do have somewhat optimistic view on acquisition of data. I also disagree with the assessment that sticking a GPS is the only thing that created autonomous driving, there is complex vision AI tech that went into keeping the thing in lanes, detecting street signs, avoiding obstacles , anticipating unsafe circumstances and even navigating something as simple as a 4 way stop with other human drivers. I agree that we are not as close to autonomous medicine as we are to driving. But on the flip side there is work being done in radiology, pathology an even derm which is ultimately going to redefine the role of AI in diagnostics . The question of it being a decade away or two or more is a more difficult one to answer and since I am not in alaphbet headquarters, i wont give an estimate. But I would be surprised if during my life time those fields are not drastically changed by the tech.

We cant automate all of medicine since medicine in its essence is a human endevour. But that doesnt mean that we automation of large scale pattern recognition tasks and sub tasks in medicine is not going to occur. IMO.

 

[Dr. Faust]

I don't think there will be a replacement in 10 years, but the potential good that can come of AI would be wonderful. Imagine scanning through every PubMed article and medical text book ever published in a second. I think that AI can help physicians make percentile diagnoses based upon the data that they enter during a patient assessment. For example, they enter all of the data and laboratory values from the patient encounter and then a list of possible diagnoses is generated. The physician then combs through the list and enters additional differential information to get an accurate diagnosis. If it is the wrong diagnosis, the AI has the ability to correct itself. Also, all of the computerized data can be linked on a network and improve its accuracy based on outcomes. What if this technology merged with an advanced DaVinci robot?

 

[Gurby]

In the farther future AI will be integrated with live imaging from scopes and be able to assist in pointing out features of interest on video reducing human error during procedures or in deciding treatment. This is technology that is already being developed in labs so I’m fairly confident it’ll exist in some commercial form by the time my generation is attending.

Don't know if you are directly referring to this study, but that technology has already been tested in a prospective RCT in China, and been shown to be safe and efficacious. Really cool stuff!

https://thehealthcareblog.com/blog/2019/06/07/the-best-medical-ai-research-that-you-probably-havent-heard-of/

Real-time automatic detection system increases colonoscopic polyp and adenoma detection rates: a prospective randomised controlled study | Gut

 

[Goro]

Recently, there's been a lot talk about Andrew Yang and his policies regarding physicians on the Reddit form. He says that computers could easily automate all of medicine, and his supporters state all we do as Drs is memorize symptoms and do the basic procedures over again, I didn't think so at first, but then I had a friend on IT tell me like 10 reasons otherwise. Do any wise physicians/students have good counter arguments for this?

I've been hearing this since the 1980, right down to the "it will happen in 10 years".
Not holding my breath.

 

[Dr G Oogle]

Thanks for the clarification. I think we agree on many points. I do have somewhat optimistic view on acquisition of data. I also disagree with the assessment that sticking a GPS is the only thing that created autonomous driving, there is complex vision AI tech that went into keeping the thing in lanes, detecting street signs, avoiding obstacles , anticipating unsafe circumstances and even navigating something as simple as a 4 way stop with other human drivers. I agree that we are not as close to autonomous medicine as we are to driving. But on the flip side there is work being done in radiology, pathology an even derm which is ultimately going to redefine the role of AI in diagnostics . The question of it being a decade away or two or more is a more difficult one to answer and since I am not in alaphbet headquarters, i wont give an estimate. But I would be surprised if during my life time those fields are not drastically changed by the tech.

We cant automate all of medicine since medicine in its essence is a human endevour. But that doesnt mean that we automation of large scale pattern recognition tasks and sub tasks in medicine is not going to occur. IMO.

I don’t disagree with you at all, but was only making the point that it is going to be a good long time before any doctor is rendered absolute by AI.

My general opinion is AI will be pretty good at doing things that humans are already good at; driving, working assembly lines, even some farming, predefined tasks etc. but will be pretty terrible at things we are not good or don’t have a complete understanding of like identifying elephant sounds in the forest, figuring out black hole physics or doctoring.

I definitely think tough it can drastically augment our practice.

 

[nimbus]

I've been hearing this since the 1980, right down to the "it will happen in 10 years".
Not holding my breath.

It’ll happen at the same time we stop relying on fossil fuels.

 

[ConfusedAboutEverything]

Recently, there's been a lot talk about Andrew Yang and his policies regarding physicians on the Reddit form. He says that computers could easily automate all of medicine, and his supporters state all we do as Drs is memorize symptoms and do the basic procedures over again, I didn't think so at first, but then I had a friend on IT tell me like 10 reasons otherwise. Do any wise physicians/students have good counter arguments for this?

Assume worst case scenario happens: Even if it happens, so what? We're all ****ed.

There are no guarantees in life. Your future will have a recession just as many happened before, your future will involve social changes in new generations that you despise just like generations before you,, your future will have several pains and sufferings and it is possible that current government is creating a future dystopia. Get over it. Life is hard and uncertain. I don't buy this BS that somehow you are going to have a carefree life, with everything going just splendid and everything will be exactly as you expect it. I guarantee you, it wont.

If you are really that worried, go into psychiatry. That will at least give you more time before that gets automated. Don't go into radiology or surgery. Do something that requires human warmth and compassion.

 

[Lucca]

Don't know if you are directly referring to this study, but that technology has already been tested in a prospective RCT in China, and been shown to be safe and efficacious. Really cool stuff!

https://thehealthcareblog.com/blog/2019/06/07/the-best-medical-ai-research-that-you-probably-havent-heard-of/

Real-time automatic detection system increases colonoscopic polyp and adenoma detection rates: a prospective randomised controlled study | Gut

Quite a few groups doing this in the US and EU as well!

 

[kb1900]

Thanks for the clarification. I think we agree on many points. I do have somewhat optimistic view on acquisition of data. I also disagree with the assessment that sticking a GPS is the only thing that created autonomous driving, there is complex vision AI tech that went into keeping the thing in lanes, detecting street signs, avoiding obstacles , anticipating unsafe circumstances and even navigating something as simple as a 4 way stop with other human drivers. I agree that we are not as close to autonomous medicine as we are to driving. But on the flip side there is work being done in radiology, pathology an even derm which is ultimately going to redefine the role of AI in diagnostics . The question of it being a decade away or two or more is a more difficult one to answer and since I am not in alaphbet headquarters, i wont give an estimate. But I would be surprised if during my life time those fields are not drastically changed by the tech.

We cant automate all of medicine since medicine in its essence is a human endevour. But that doesnt mean that we automation of large scale pattern recognition tasks and sub tasks in medicine is not going to occur. IMO.

I think you are overestimating the technology underlying driving ai. Just to clarify what is going on:
- object detection. This is the first endeavor in AI that really took off and now anyone with a RPI2 can have a camera that draws boxes around different objects and name them in close to real time. Self-driving cars further optimize already created models based on the new data they constantly collect and their predictions get filtered/optimized for pedestrians, bikes, cars, curbs, lane markings etc.
- prediction of velocity of objects/their future position (is this bike static or moving into the car’s path)
- integration of other data that a human may not pick up on (weather, wind, prediction of a car 2 cars ahead that the driver can’t even see etc.
- +/- hierarchical rules like following street signs, lights etc.

It’s not that these things aren’t incredibly complex in practice and at the front end, but that the underlying tech is not that crazy at all. Self-driving cars are one of the first consumer level applications for
1) simplified predictions needed (object classifier and velocity/position which is vector/matrix math that computers excel at)
2) ease of data creation. Any car with sensors literally generates perfect data 24/7 while driving
3) limited outputs based on predictions (steering, gas, break, lights)
4) immediately improves on the current situation even when unoptomized


Medicine has both 1+2+3 in very few fields with the biggest issue with 2. In fact to solve 2, there are AI based deidentification projects going on right now to scrub PPI from nOtes using NLP which could be a HUGE win for the future of AI in predictive aspects of medicine.


In the computer vision world, classifier data is rectangular bounding boxes around labeled objects in images. Im hoping for the day an AI could reliably trace the outline of perceived significant findings and landmarks and spit out calculated sizes, delta from previous studies etc in real time US, CXR etc. before focusing on dxing pneumonia or cancer

 

[Thesimplelifeofamyloid]

Also, no one brought this up, but what about the security threats with a system like this? Imagine if someone could hack the AI, and make it give a different result then what's actually the case. Could be used as a weapon of biowarfare