Having seen many specialist/doctors I generally think they suck at diagnosing, unless its something you could literally discover yourself by a google search and access to a specific test. I don't think they are bad doctors, I just learned that sometimes diagnosis can be rather difficult.
Then, you have a computer, it can reference recent studies compare symptoms with hundreds of diseases and generate a test plan accordingly based on probability of a disease or urgency. Frankly, I am surprised not more work is being done with algorithms in medicine.
Humans are biased, egoistic and most don't like being told they are wrong.
I tend to agree, doctors is going to be the holy shit moment for society when we begin replacing them across the board. I mean we still might need 100 or so (nationwide in the usa), maybe keep the top 1000 doctors/surgeons for times when creativity and unique problem solving is needed and to 'sign off' on computers' work remotely (for awhile, till they're not needed at all anymore)...
If the only job left per se for doctors is to 'okay and sign off' on prescriptions that a computer recommends, after a computer does the entire intake/evaluation, then 1 doctor can replace 100 or 1000 for 20 second evals over skype.
I'm a doctor and a software developer. I frequently perform surgery. There's no way a computer/robot will be doing what I do as a doctor in the next 10 years.
Believe what you will. I don't really have a personal horse in this race, since I wouldn't mind if I was replaced as a doctor/surgeon. I actually took last year off medicine and focused on my tech businesses and it was great.
I have very high hopes for ai/automation. Maybe not 10 years, but can you be sure 20 years is safe? What if your kid started med school in 10 years, and by the time they get through residency they could be replaced by robots? It's a possibility...
I think specific 'parts' of medicine are more adaptable to change... PCP's could be 'mostly' automated. I'm not saying it'll be 100%, but say 20% stay around for a decade or two as management overseeing the robotic slaves -- making sure they have > levels of care and lower death rates than a human would by a magnitude big enough to avoid lawsuits.
There's not a job in sight that can't someday be done better by a robot. Even software developer. I don't know when, I just know on a long enough timeline it's almost inevitable.
I think you're looking at this all wrong. If robots replace me then that means the robots are safer than I am and do a better job. That's awesome. By the time robots have replaced me, they would have long ago replaced accountants and lawyers. We'd be living in a super efficient world. The human element is extremely important in medicine/surgery so we won't be replaced for a little longer. For a long time it'll be there to help us and make our jobs easier.
It's like another industrial revolution where we all have more time to... not work.
I am not defined by my career.
But... I don't see this happening even in the next 20 years. We can't even yet trust cars to drive forward, back, left and right safely with strict rules, markings and signage to guide them.
Not to undermine what you do, or what any doctor does, I honestly believe driving a car to be something that's tougher than surgery (for a robot). Now of course, when I say 20 years, I think by then we'll have robots who can do a number of 'routine' surgeries at least. Maybe each robot is specialized and can only do 1 surgery but there's 15 different types of robots who all have mastered 1 surgery.
When performing a surgery, I'm sure you need to think on your feet a bit, but how many scenarios can you think of that might happen in a surgery ? Knick a blood vessel, throw a blood clot, drop a tool in the patient, unforseen bleeder/hemorrhage, alergy to drug, etc... I'm sure there's quite a list...
How many scenarios can happen in a driving situation - magnitudes more than a surgery, because a surgery is in an isolated room and controlled environment. Driving is not. On the road you need to worry about:
* Pedestrians from front, and sides.
* Animals from all angles.
* Rain/Snow. Hydroplaning.
* Traffic construction.
* Other idiots on the road.
* Helicopter Crash.
* Police Chase.
* Sheep stuck in the road.
* Forest fires. (Obey the speed limit or floor it to escape?)
* Billing (If you're an uber ai car).
* Drivers that are angry, sad, happy, distracted, having a medical emergency, falling asleep.
* Natural disasters like an earthquake.
* Bad directions.
* Loss of internet/connectivity.
* Running out of gas/electric.
* Curves in the road.
* Signs that could be permanent or temporary.
* Obeying traffic laws.
* Being aggressive when needed to take a lane.
* Be courteous and move over for faster traffic passing on the left.
* Sensor failures and malfunctions.
* Tire blowout.
My point is it's a LOT more than just forward/back/left/right, if you had an enclosed building w/ a driving track, and could control who was in the room, where they were standing, what their jobs were, before the car took off, then you mitigate MANY of the issues that self-driving cars have, and the technology as it exists today is good enough for that scenario.
Personally though I see us (hopefully) moving towards a post-scarcity society, sort of like Wall-e, but hopefully without the laziness (I'm sure we can find fulfillment in other ways) -- Research for instance will be a hobby instead of career because people --especially scientists will always have curiosity pushing them forward.
If we survive global warming, the next 100 years is going to be unfathomably different from the last 100...
Maybe I should mention I perform surgery on awake patients. What happens when the patient gives me feedback? I have to negotiate with them and accommodate them. What about when the surgeon communicates with the assistants/nurses? What about with the anesthetist? What if the power goes out? What about the discussion with the patient/family when things don't go as planned? What about deciding with the patient if surgery is even appropriate in the first place? Will a robot be a good judge of when it's important to ignore what the patient is saying?
You know about driving because you do it. It makes sense you think medicine/surgery is simple, since you've never done it.
Also -- if something does come up, couldn't you as a master surgeon technically have 5 surgeries on huge monitors around you, and if the robot hits a snag you can swoop in and fill in? Using a microscopic tool you couid even do that remotely, just take over the robot's hands? In this scenario you replaced 5 surgeons combined w/ robots. Still a win for efficiency.
Well, that would make it a lot different than most surgeries... one of the biggest things against AI is you can't count on what others will do, and AI isn't a good judge of people/character/humanity.
A lot of the other jobs might be replaceble too like anesthetist, so in those instances each robot would talk to the next. Power outages - most hospitals have generator backups, and for long-term power issues they have contingency plans to close OR's except for emergent issues.
I think we're 40-70 years from complete AGI, meaning a robot that can do EVERYTHING you and I can do, has full human capabilities. Sometime before then we'll have lawyers, engineers, doctors and all labor/service jobs replaced by automation. (Or at least higher than 50% for every major industry).
I'm not saying I think surgery is simple, I can't do it I couldn't 'self-teach' it either like I did software. I respect the profession and time it takes to become a doctor.
I imagine surgery as having a checklist of things you go through. Sure there's instances when the plan fails but then you have at least a flow chart in your head of what next to do. The issue w/ (current ai) is it needs us to feed it steps to learn, it has to know all the possible scenarios. In an enclosed environment you can feed it more data and have a higher chance of success than in an open / unregulated environment like the great outdoors...
Imagine you took your surgery outside. Simply moving the table outside. How many new factors would that introduce? Foreign contaminants from bugs, bird shit, temperatures, wind, rain, snow.
You can account for maybe 80% or better of possible scenarios in the enclosed OR. Accounting for real world scenarios when there are millions of other people at play is nigh impossible. -- Even if you had an enclosed mall, and put the operating table in the middle of the walkway, some idiot eating a hamburger could trip and launch his burger into the patients' abdomen -- people is what you can't count for.
In the OR there's a handful of attendings/nurses/etc, but in the outside world there's millions depending on the city. I seriously hope going into a routine surgery that my chances of survival are better than driving through rush hour on Christmas eve with 14 inches of snow coming down all around me.
I know how to drive, I think it's simple, anyone can do it, but I don't trust anyone to do it. I am always looking at people for signs of trouble and correct myself to avoid them. I don't know how to do surgery, I couldn't possibly start. But my point from the start is it's not even science fiction it's been done. There have been fully un-manned robotic surgeries.
You act as if it's impossible when it's already been done. Probably with better success than self-driving cars even. I personally think it'll be 2030 before self-driving cars are street-safe if even then.
> Well, that would make it a lot different than most surgeries
And guess what. There are thousands of different types of surgeries. You act like it's the same as driving a car, where it's pretty similar all around the world. I'm not going to bother discussing this further. If you'd like to join me in surgery or a medical consult one day you're welcome, but currently it seems like you're more keen to tell me about my work than you are to learn from me.
Sorry to sidetrack but being able to see a surgery would really exciting for me. Do you think it is possible to find a surgeon to allow you to watch a surgery? That would be pretty amazing.
There may be rules around it where you are. I'm in Australia. All you'd need is to find a surgeon who's happy for you to watch. Just don't touch anything green/blue.
If your presence there is of no benefit to the procedure then it's a good idea to get consent from the patient too.
I think it will be a step-by-step progress where we create specialized tools first. For example robot that helps with precise suturing. So the surgeon will input all the parameters, and the robot does the suturing. Then we can go to more complex things.
There won't be one robot that can replace a surgeon anytime soon. Surgeons and specialized robots/tools will work side-by-side for some time before that happens.
Aren't these algorithms trained on expert-annotated data? If so how does it end up performing better than radiologists? Or are these radiologists only mediocre at detecting cancer?
I'm a doctor but not a radiologist. I imagine it would go like this:
Computer makes diagnosis on image of lesion that has already been biopsied and has a known correct diagnosis.
I don't know how it would work for human false negatives that were never biopsied. Possibly the AI could be trusted enough to make a biopsy on its recommendation.
Watson was mostly a marketing jargon built by IBM. They marketed and sold the product before the technology was ready.
Google, Microsoft, Facebook, (and probably soon Amazon) have advanced the AI industry further than IBM at this point. You can predict this by the number of papers released by each company. Stats from 2017: https://medium.com/@chuvpilo/whos-ahead-in-ai-research-insig...
Seriously? Build AI hospital, charge 1/2 the cost across the board, begin replacing doctors/surgeons worldwide w/ google doctor. If google could take home 1/3rd of all GDP spending on medical that would make them the top company in the world.
Are not they searching for new area to expand their business?
Their turnover is around $150b, if they go in healthcare they could reach $1 trillion.
Which other business path can safely offer similar potential?
Having seen many specialist/doctors I generally think they suck at diagnosing, unless its something you could literally discover yourself by a google search and access to a specific test. I don't think they are bad doctors, I just learned that sometimes diagnosis can be rather difficult.
Then, you have a computer, it can reference recent studies compare symptoms with hundreds of diseases and generate a test plan accordingly based on probability of a disease or urgency. Frankly, I am surprised not more work is being done with algorithms in medicine.
Humans are biased, egoistic and most don't like being told they are wrong.
I tend to agree, doctors is going to be the holy shit moment for society when we begin replacing them across the board. I mean we still might need 100 or so (nationwide in the usa), maybe keep the top 1000 doctors/surgeons for times when creativity and unique problem solving is needed and to 'sign off' on computers' work remotely (for awhile, till they're not needed at all anymore)...
If the only job left per se for doctors is to 'okay and sign off' on prescriptions that a computer recommends, after a computer does the entire intake/evaluation, then 1 doctor can replace 100 or 1000 for 20 second evals over skype.
I don't think surgery will be done without human hand for a very long time
If you consider 2006 a very long time 'ago', then you're absolutely right... https://www.engadget.com/2006/05/19/robot-surgeon-performs-w...
Then there's a more modern article at: https://curiosity.com/topics/a-robot-performed-soft-tissue-s...
So, it's definitely possible. Doctors will be replaced probably before software devs, so I think my job is safe(r).
I'm a doctor and a software developer. I frequently perform surgery. There's no way a computer/robot will be doing what I do as a doctor in the next 10 years.
Believe what you will. I don't really have a personal horse in this race, since I wouldn't mind if I was replaced as a doctor/surgeon. I actually took last year off medicine and focused on my tech businesses and it was great.
I have very high hopes for ai/automation. Maybe not 10 years, but can you be sure 20 years is safe? What if your kid started med school in 10 years, and by the time they get through residency they could be replaced by robots? It's a possibility...
I think specific 'parts' of medicine are more adaptable to change... PCP's could be 'mostly' automated. I'm not saying it'll be 100%, but say 20% stay around for a decade or two as management overseeing the robotic slaves -- making sure they have > levels of care and lower death rates than a human would by a magnitude big enough to avoid lawsuits.
There's not a job in sight that can't someday be done better by a robot. Even software developer. I don't know when, I just know on a long enough timeline it's almost inevitable.
> can you be sure 20 years is safe?
I think you're looking at this all wrong. If robots replace me then that means the robots are safer than I am and do a better job. That's awesome. By the time robots have replaced me, they would have long ago replaced accountants and lawyers. We'd be living in a super efficient world. The human element is extremely important in medicine/surgery so we won't be replaced for a little longer. For a long time it'll be there to help us and make our jobs easier.
It's like another industrial revolution where we all have more time to... not work.
I am not defined by my career.
But... I don't see this happening even in the next 20 years. We can't even yet trust cars to drive forward, back, left and right safely with strict rules, markings and signage to guide them.
Not to undermine what you do, or what any doctor does, I honestly believe driving a car to be something that's tougher than surgery (for a robot). Now of course, when I say 20 years, I think by then we'll have robots who can do a number of 'routine' surgeries at least. Maybe each robot is specialized and can only do 1 surgery but there's 15 different types of robots who all have mastered 1 surgery.
When performing a surgery, I'm sure you need to think on your feet a bit, but how many scenarios can you think of that might happen in a surgery ? Knick a blood vessel, throw a blood clot, drop a tool in the patient, unforseen bleeder/hemorrhage, alergy to drug, etc... I'm sure there's quite a list...
How many scenarios can happen in a driving situation - magnitudes more than a surgery, because a surgery is in an isolated room and controlled environment. Driving is not. On the road you need to worry about:
* Pedestrians from front, and sides.
* Animals from all angles.
* Rain/Snow. Hydroplaning.
* Traffic construction.
* Other idiots on the road.
* Helicopter Crash.
* Police Chase.
* Sheep stuck in the road.
* Forest fires. (Obey the speed limit or floor it to escape?)
* Billing (If you're an uber ai car).
* Drivers that are angry, sad, happy, distracted, having a medical emergency, falling asleep.
* Natural disasters like an earthquake.
* Bad directions.
* Loss of internet/connectivity.
* Running out of gas/electric.
* Curves in the road.
* Signs that could be permanent or temporary.
* Obeying traffic laws.
* Being aggressive when needed to take a lane.
* Be courteous and move over for faster traffic passing on the left.
* Sensor failures and malfunctions.
* Tire blowout.
My point is it's a LOT more than just forward/back/left/right, if you had an enclosed building w/ a driving track, and could control who was in the room, where they were standing, what their jobs were, before the car took off, then you mitigate MANY of the issues that self-driving cars have, and the technology as it exists today is good enough for that scenario.
Personally though I see us (hopefully) moving towards a post-scarcity society, sort of like Wall-e, but hopefully without the laziness (I'm sure we can find fulfillment in other ways) -- Research for instance will be a hobby instead of career because people --especially scientists will always have curiosity pushing them forward.
If we survive global warming, the next 100 years is going to be unfathomably different from the last 100...
Maybe I should mention I perform surgery on awake patients. What happens when the patient gives me feedback? I have to negotiate with them and accommodate them. What about when the surgeon communicates with the assistants/nurses? What about with the anesthetist? What if the power goes out? What about the discussion with the patient/family when things don't go as planned? What about deciding with the patient if surgery is even appropriate in the first place? Will a robot be a good judge of when it's important to ignore what the patient is saying?
You know about driving because you do it. It makes sense you think medicine/surgery is simple, since you've never done it.
Also -- if something does come up, couldn't you as a master surgeon technically have 5 surgeries on huge monitors around you, and if the robot hits a snag you can swoop in and fill in? Using a microscopic tool you couid even do that remotely, just take over the robot's hands? In this scenario you replaced 5 surgeons combined w/ robots. Still a win for efficiency.
Yes. And? Are there not millions of people in your country who need surgical procedures but aren't getting them because they cannot afford them?
Well, that would make it a lot different than most surgeries... one of the biggest things against AI is you can't count on what others will do, and AI isn't a good judge of people/character/humanity.
A lot of the other jobs might be replaceble too like anesthetist, so in those instances each robot would talk to the next. Power outages - most hospitals have generator backups, and for long-term power issues they have contingency plans to close OR's except for emergent issues.
I think we're 40-70 years from complete AGI, meaning a robot that can do EVERYTHING you and I can do, has full human capabilities. Sometime before then we'll have lawyers, engineers, doctors and all labor/service jobs replaced by automation. (Or at least higher than 50% for every major industry).
I'm not saying I think surgery is simple, I can't do it I couldn't 'self-teach' it either like I did software. I respect the profession and time it takes to become a doctor.
I imagine surgery as having a checklist of things you go through. Sure there's instances when the plan fails but then you have at least a flow chart in your head of what next to do. The issue w/ (current ai) is it needs us to feed it steps to learn, it has to know all the possible scenarios. In an enclosed environment you can feed it more data and have a higher chance of success than in an open / unregulated environment like the great outdoors...
Imagine you took your surgery outside. Simply moving the table outside. How many new factors would that introduce? Foreign contaminants from bugs, bird shit, temperatures, wind, rain, snow.
You can account for maybe 80% or better of possible scenarios in the enclosed OR. Accounting for real world scenarios when there are millions of other people at play is nigh impossible. -- Even if you had an enclosed mall, and put the operating table in the middle of the walkway, some idiot eating a hamburger could trip and launch his burger into the patients' abdomen -- people is what you can't count for.
In the OR there's a handful of attendings/nurses/etc, but in the outside world there's millions depending on the city. I seriously hope going into a routine surgery that my chances of survival are better than driving through rush hour on Christmas eve with 14 inches of snow coming down all around me.
I know how to drive, I think it's simple, anyone can do it, but I don't trust anyone to do it. I am always looking at people for signs of trouble and correct myself to avoid them. I don't know how to do surgery, I couldn't possibly start. But my point from the start is it's not even science fiction it's been done. There have been fully un-manned robotic surgeries.
You act as if it's impossible when it's already been done. Probably with better success than self-driving cars even. I personally think it'll be 2030 before self-driving cars are street-safe if even then.
> Well, that would make it a lot different than most surgeries
And guess what. There are thousands of different types of surgeries. You act like it's the same as driving a car, where it's pretty similar all around the world. I'm not going to bother discussing this further. If you'd like to join me in surgery or a medical consult one day you're welcome, but currently it seems like you're more keen to tell me about my work than you are to learn from me.
Sorry to sidetrack but being able to see a surgery would really exciting for me. Do you think it is possible to find a surgeon to allow you to watch a surgery? That would be pretty amazing.
There may be rules around it where you are. I'm in Australia. All you'd need is to find a surgeon who's happy for you to watch. Just don't touch anything green/blue.
If your presence there is of no benefit to the procedure then it's a good idea to get consent from the patient too.
I think it will be a step-by-step progress where we create specialized tools first. For example robot that helps with precise suturing. So the surgeon will input all the parameters, and the robot does the suturing. Then we can go to more complex things.
There won't be one robot that can replace a surgeon anytime soon. Surgeons and specialized robots/tools will work side-by-side for some time before that happens.
Aren't these algorithms trained on expert-annotated data? If so how does it end up performing better than radiologists? Or are these radiologists only mediocre at detecting cancer?
I'm a doctor but not a radiologist. I imagine it would go like this:
Computer makes diagnosis on image of lesion that has already been biopsied and has a known correct diagnosis.
I don't know how it would work for human false negatives that were never biopsied. Possibly the AI could be trusted enough to make a biopsy on its recommendation.
It's only a matter of time before all X-Ray, CTScans, etc. use algorithms for secondary if not primary screens.
What makes this different from where IBM Watson failed?
Watson was mostly a marketing jargon built by IBM. They marketed and sold the product before the technology was ready.
Google, Microsoft, Facebook, (and probably soon Amazon) have advanced the AI industry further than IBM at this point. You can predict this by the number of papers released by each company. Stats from 2017: https://medium.com/@chuvpilo/whos-ahead-in-ai-research-insig...
It’s going to be an even better fit than Google Flu trends https://www.wired.com/2015/10/can-learn-epic-failure-google-...
One thing for sure is that all these AIs (or their creators) are pretty good at marketing than the actual doctors :).
What is the Google's long term goal here? Is there a plan to monetise this?
Seriously? Build AI hospital, charge 1/2 the cost across the board, begin replacing doctors/surgeons worldwide w/ google doctor. If google could take home 1/3rd of all GDP spending on medical that would make them the top company in the world.
Are not they searching for new area to expand their business? Their turnover is around $150b, if they go in healthcare they could reach $1 trillion. Which other business path can safely offer similar potential?
Charge $100 for a hospital to send them an image for screening. Google sends back a cancer confidence score.