How to learn machine/deep learning/coding relevant for medicine before the AI storm hits?

[deleted1009711]

Hi there!

II have 7 months until med school starts. What can I do now/in medical school/after medical school, to learn the basics, considering I majored in a Bio major in college?

I was thinking of taking 7 months of stats and biostats before starting (useful in any case) with a tiny tiny amount of intro to python, and have been looking into Data Science Masters programs after my third year of medical school (specifically those targeting machine learning), but for example is there anything I can do while IN** medical school, like electives?? What is the best way to go about this?

Thanks!

---

Members don't see this ad.

 

[SeitenKishi]

I won't comment on whether AI can replace most functions of doctors, as that is up to debate. But I think I am qualified to comment on the "learning AI" part, as I have a Master's in engineering from a top engineering school (think MIT, Stanford), and I worked as an AI/ML engineer at a major healthcare company before entering med school.

You have to be pretty good at math, especially linear algebra (e.g. matrices, regression, eigenvalues/eigenvectors) and statistics (e.g. bootstrap) to understand machine learning. You also have to learn a programming language (e.g. Python, MATLAB), a coding framework, development environment, etc. If you don't know either of these, you won't be successful at truly learning how machine learning works.

I would start with Khan Academy, Udacity, Google courses to see if they have free courses (usually intro courses are free). Don't spend extra money for a Master's if you can avoid it. Better to avoid debt.

 

[deleted1009711]

I wonder if I should then go ham on math now...jesus, it's absolutely CRAZY that med programs are not incorporating this kind of stuff into their programs yet, or at least making it an option!!!

Ok, so basically: 1.) Be competent in basic mathematics 2.) learn some language, which I guess by default should be Python 3.) Jump into machine learning and applying it to medicine.

 

[SeitenKishi]

If you can get an MS for free, go for it. Just make sure it doesn't interfere with your med school studying (Step 1 takes priority MS1 and MS2). At my school, there is a Precision Medicine club consisting of students and a faculty mentor (covers AI/ML). Maybe your school has a AI club or something.

If it makes you feel better, you can *technically* learn AI/ML without knowing any mathematics, but then you would be blindly using AI/ML techniques on random data, without knowing what you are truly doing. This is dangerous because the conclusions and patterns you derive can be completely false or irrelevant (in other words, since you didn't know the theory and mathematics, you used the wrong AI/ML approaches on different types of data).

 

[deleted1009711]

I might have more questions soon, but I really really appreciate the help man/gal!!!!!!

 

[Thesimplelifeofamyloid]

Hi there!

I'm an upcoming M1. Needless to say, all the reassurances about "empathy" and "creative human thinking" etc. have not convinced me a single bit that most of what doctors will do will be automated by AI in a decade or two, and at best most physicians will be glorified technicians/counselors with huge pay hits (matters more considering loans). I want to be in on AI research in healthcare so I won't get totally wrecked.

TLDR: I have 7 months until med school starts. What can I do now/in medical school/after medical school, to learn the basics, considering I majored in a worthless Bio major in college?

I was thinking of taking 7 months of stats and biostats before starting (useful in any case) with a tiny tiny amount of intro to python, and have been looking into Data Science Masters programs after my third year of medical school (specifically those targeting machine learning), but for example is there anything I can do while IN** medical school, like electives?? What is the best way to go about this?

Thanks!

Used to think the same way as you. Spending a year at the VA with patients that can't even tell you what's wrong with them, speak under 3 words, have 20 comorbidities convinced me otherwise. The more you go up in years, the more you see how complicated medicine is and how important the physical exam is. Nevertheless, I don't think learning any skill is a bad idea. If it's something you enjoy, go for it.

 

[Cornfed101]

Don't burn yourself out over this though. Programming is not for everyone. Speaking from experience, it can be extremely repetitive and boring if you don't love it. For the foreseeable future, doctors will still be the pilots and machine learning will only help.

As an aside, you learn programming and integrate engineering into medicine at Carle SOM in Illinois. You could check that med school out (they do require a lot of the math that @SeitenKishi mentioned).

 

[Dro133]

Current MS4, self-taught programmer with no formal ML training. My advice is to not expect to become an expert on algorithm development or get too heavy into the maths -- there are people who do PhD's in this stuff, and it's likely out of the scope of what will love useful to you. My advice is to take a free Coursera course on intro ML and get a basic overview of ML theory, the difference between supervised and unsupervised learning, etc. Check out codeacademy.com or the book Python for biologists to get some programming experience under your belt.

If you're still interested after all that, the best way to truly get your hands dirty IMO is to find a faculty member who is using programming for medical research, and see if you can get on a project. In my experience, the best way to learn is to have a project to work on and figure it out as you go.

I don't think AI is going to replace doctors any time soon, but if you have an understanding of the basics, even if you're not the one working under the hood, you're going to be ahead of the game.

Best of luck

 

[Select All That Apply]

Patent the idea for adding a multi-layered "net" to catch patients on Avasys machines so the voyeuristic operator who is probably taking cat naps can actually do something useful while I'm stuck in a rapid with another patient three rooms down.

 

[Cornfed101]

Or get EHRs to communicate with each other... Ya know like computers have been able to do since the 90s.

 

[HKSZYU]

I'm in a similar boat to @SeitenKishi. I got an applied stats MS, and did computational bio research for a bit. I'd echo the sentiment that there's a pretty big difference between knowing how to throw some code into python, and really knowing the math behind ML. If you have 7 months, a Udacity nano-degree could be a good balance for you. I think it's very practical, and provides you just enough math that you have an idea of why things work without getting so theoretical as to be unaccessible to non-mathematicians.

I don't know if a masters in necessary. A data science masters degree gets very math-y, and if that's not where your skills or interests lies, it can be quite dull and painful. But if you have a bio background, can amass good coding skills, and have a basic theoretical foundation, you'd be a good fit for many comp bio or biostat labs, where you could build more skills. I think that's likely the most practical and applicable path forward.

 

[LoGo]

PhD in enegineering, and MD here. The key to your first post is this: you haven’t even started M1, and therefore you literally have no idea what being a doctor is about. Consequently, your assumption that AI will replace all doctor functions is completely wrong and most machine learning researchers will be the first to tell you that AI will not do that - it’s an adjunct tool that doctors can use to improve efficiency and overall patient care.

As such, my recommendation is don’t waste your time with this. You went into medicine to do medicine and indicated you don’t like math. Doing AI research because you’re afraid of losing your job as a doctor makes zero sense.

 

[NotAProgDirector]

You are worried about nothing. Computers are not going to replace physicians. When I was growing up my father was an engineer. His colleagues kept talking about how computers were going to replace them and design everything themselves. What happened? Computers definitely changed the type of work that engineers do, and a single engineer can now do more with computer support. But the projects get bigger and more complicated, and there's plenty of engineering jobs. The same will be true in medicine also. You will not be replaced by a computer, it's just that the nature of your job will change over time as technology progresses.

Even radiologists will be fine, I think. Computer technology can probably read images better than humans, but I expect you'll always need a human to review the machine's findings. And as imaging technology progresses, there will be more and more images to review.

If you want to learn about ML because it interests you, then fine. If you're doing it to "save your job", that's silly. And if machine learning really becomes good enough to replace physicians, then it's replacing most/all of the workforce so the entire structure of society will be different. And last, basic coding skills is unlikely to be helpful in such a future -- in fact computers will be programming themselves, and non of the languages you could learn today will be of any value.

 

[TrpIleAsnGluArg]

OP, AI/ML is highly overrated in my opinion. Its exceptional at image recognition (depending on your definition of "recognition" though) and "solving" simulated environments (OPEN AI, Alpha GO, etc...). At everything else its not that useful. So unless every task in medicine becomes image recognion, or we become able to simulate a whole patient, I think it will be fine.

You are worried about nothing. Computers are not going to replace physicians. When I was growing up my father was an engineer. His colleagues kept talking about how computers were going to replace them and design everything themselves. What happened? Computers definitely changed the type of work that engineers do, and a single engineer can now do more with computer support. But the projects get bigger and more complicated, and there's plenty of engineering jobs. The same will be true in medicine also. You will not be replaced by a computer, it's just that the nature of your job will change over time as technology progresses.

Even radiologists will be fine, I think. Computer technology can probably read images better than humans, but I expect you'll always need a human to review the machine's findings. And as imaging technology progresses, there will be more and more images to review.

If you want to learn about ML because it interests you, then fine. If you're doing it to "save your job", that's silly. And if machine learning really becomes good enough to replace physicians, then it's replacing most/all of the workforce so the entire structure of society will be different. And last, basic coding skills is unlikely to be helpful in such a future -- in fact computers will be programming themselves, and non of the languages you could learn today will be of any value.

Agree with everything here except the last sentence. Computers will never program themselves unless we have a true AI, a situation which your comment addresses. Learning a specific language is the least important part of programming as well - it is all about the concepts and knowing how to design programs and how computers generally work. Putting the finished design into code is the easy part.

 

[7331poas]

You are worried about nothing. Computers are not going to replace physicians. When I was growing up my father was an engineer. His colleagues kept talking about how computers were going to replace them and design everything themselves. What happened? Computers definitely changed the type of work that engineers do, and a single engineer can now do more with computer support. But the projects get bigger and more complicated, and there's plenty of engineering jobs. The same will be true in medicine also. You will not be replaced by a computer, it's just that the nature of your job will change over time as technology progresses.

Even radiologists will be fine, I think. Computer technology can probably read images better than humans, but I expect you'll always need a human to review the machine's findings. And as imaging technology progresses, there will be more and more images to review.

If you want to learn about ML because it interests you, then fine. If you're doing it to "save your job", that's silly. And if machine learning really becomes good enough to replace physicians, then it's replacing most/all of the workforce so the entire structure of society will be different. And last, basic coding skills is unlikely to be helpful in such a future -- in fact computers will be programming themselves, and non of the languages you could learn today will be of any value.

This is really not true at all. There is no question that computers have reduced the number of jobs, not increased them.

Your point about the radiologist misses the point. If previously a radiologist could read x images in a day, an AI assisted radiology device could improve their efficiency to reading 1.5x per day. That is now less jobs needed for a single center's radiology demand.

You are correct on one point however, learning AI will not help a medical student. The amount you could possibly learn while in medical school is not helpful to a tech corporation. Your best bet would be to join these companies as a consultant in improving their algorithm. I know a radiology graduate who does something similar at Google and gets paid very well.

PhD in enegineering, and MD here. The key to your first post is this: you haven’t even started M1, and therefore you literally have no idea what being a doctor is about. Consequently, your assumption that AI will replace all doctor functions is completely wrong and most machine learning researchers will be the first to tell you that AI will not do that - it’s an adjunct tool that doctors can use to improve efficiency and overall patient care.

As such, my recommendation is don’t waste your time with this. You went into medicine to do medicine and indicated you don’t like math. Doing AI research because you’re afraid of losing your job as a doctor makes zero sense.

And if a program is invented that improves the efficiency of doctors then the amount of jobs available for physicians will...

Used to think the same way as you. Spending a year at the VA with patients that can't even tell you what's wrong with them, speak under 3 words, have 20 comorbidities convinced me otherwise. The more you go up in years, the more you see how complicated medicine is and how important the physical exam is. Nevertheless, I don't think learning any skill is a bad idea. If it's something you enjoy, go for it.

What makes you think that a human will be better than a computer at this task by 2060? Or perhaps a better question, what makes you think a physician will outperform a nurse practitioner + AI combination at obtaining and interpreting an H&P.

 

[deleted480308]

This is really not true at all. There is no question that computers have reduced the number of jobs, not increased them.

Your point about the radiologist misses the point. If previously a radiologist could read x images in a day, an AI assisted radiology device could improve their efficiency to reading 1.5x per day. That is now less jobs needed for a single center's radiology demand.

You are correct on one point however, learning AI will not help a medical student. The amount you could possibly learn while in medical school is not helpful to a tech corporation. Your best bet would be to join these companies as a consultant in improving their algorithm. I know a radiology graduate who does something similar at Google and gets paid very well.



And if a program is invented that improves the efficiency of doctors then the amount of jobs available for physicians will...



What makes you think that a human will be better than a computer at this task by 2060? Or perhaps a better question, what makes you think a physician will outperform a nurse practitioner + AI combination at obtaining and interpreting an H&P.

You are bad at economics. Tech advances aren’t reducing the number of jobs in total, they are reducing the numbers of very specific jobs for specific tasks thereby creating new jobs in other tasks necessary to support that new tech and freeing labor to find other new jobs to improve life or meet other needs

It isn’t a net negative

 

[deleted480308]

OP, study hard for step 1 and ignore this idea

 

[Dro133]

You are bad at economics. Tech advances aren’t reducing the number of jobs in total, they are reducing the numbers of very specific jobs for specific tasks thereby creating new jobs in other tasks necessary to support that new tech and freeing labor to find other new jobs to improve life or meet other needs

It isn’t a net negative

That's more nuance to it than that. According to a study by the Bureau of Labor Statistics, "Not surprisingly, large decreases are found within blue-collar jobs that have routine manual operations, such as assembly workers, transportation workers, and machinists. The negative impacts slightly diminish as a worker’s education level increases, and no impact exists for workers with graduate degrees." (Source: The impact of technology on labor markets : Monthly Labor Review: U.S. Bureau of Labor Statistics)

So, it will probably hurt some portion of the population, and whether it has a net positive or negative impact on society vs. just a positive impact for the highly-educated sector of the population remains to be seen. Regardless, physicians will probably be safe for the foreseeable future.

OP, I'm just an MS4, but my advice is to only pursue AI/ML/programming if it's something that's interesting to you and you are excited by the prospect of being involved in health tech development/research some day. If you hate it and you're pursuing it simply out of fear for your job, you are probably just going to be wasting your time and making yourself miserable; if that's the case, then focus on becoming the best clinician you can be.

 

[deleted480308]

That's more nuance to it than that. According to a study by the Bureau of Labor Statistics, "Not surprisingly, large decreases are found within blue-collar jobs that have routine manual operations, such as assembly workers, transportation workers, and machinists. The negative impacts slightly diminish as a worker’s education level increases, and no impact exists for workers with graduate degrees." (Source: The impact of technology on labor markets : Monthly Labor Review: U.S. Bureau of Labor Statistics)

So, it will probably hurt some portion of the population, and whether it has a net positive or negative impact on society vs. just a positive impact for the highly-educated sector of the population remains to be seen. Regardless, physicians will probably be safe for the foreseeable future.

OP, I'm just an MS4, but my advice is to only pursue AI/ML/programming if it's something that's interesting to you and you are excited by the prospect of being involved in health tech development/research some day. If you hate it and you're pursuing it simply out of fear for your job, you are probably just going to be wasting your time and making yourself miserable; if that's the case, then focus on becoming the best clinician you can be.

Of course people who refuse to learn marketable skills go unemployed. Advances change which skills are marketable, they don’t decrease need for workers/labor/skills

 

[Dro133]

Advances change which skills are marketable, they don’t decrease need for workers/labor/skills

We don't know if this is necessarily true. According to the study I quoted earlier, "Little evidence exists of the equilibrium impacts that this spread may cause. This study is viewed by the authors as a first step in evaluating how robots influence labor market equilibriums. The authors provide an empirical methodology to address the lack of research in this area. Their research concludes by stating that if the spread of robots continues, there could be sizable future declines in the employment–population ratio."

Thus, the future effects of AI on the workforce seem very much unknown. To say it will definitely increase/decrease/not change the demand for labor seems premature, in my opinion.

It's very possible that the number of employed people required to build/deploy/run/maintain a robot for a given task is less than the amount of people required to perform that task manually (as in an assembly line, for example); in this case, one would expect a net decrease in the demand for jobs, and the people filling those highly-skilled jobs would likely be the highly-educated elite. This could lead to an even larger wealth gap than currently exists, which, in my opinion, would be a net negative for society. You don't have to look much further than what happened to the Rust Belt after automation and outsourcing of manufacturing led to the loss of low-skilled labor to see what negative effects this can have on a population.

 

[deleted480308]

We don't know if this is necessarily true. According to the study I quoted earlier, "Little evidence exists of the equilibrium impacts that this spread may cause. This study is viewed by the authors as a first step in evaluating how robots influence labor market equilibriums. The authors provide an empirical methodology to address the lack of research in this area. Their research concludes by stating that if the spread of robots continues, there could be sizable future declines in the employment–population ratio."

Thus, the future effects of AI on the workforce seem very much unknown. To say it will definitely increase/decrease/not change the demand for labor seems premature, in my opinion.

It's very possible that the number of employed people required to build/deploy/run/maintain a robot for a given task is less than the amount of people required to perform that task manually (as in an assembly line, for example); in this case, one would expect a net decrease in the demand for jobs, and the people filling those highly-skilled jobs would likely be the highly-educated elite. This could lead to an even larger wealth gap than currently exists, which, in my opinion, would be a net negative for society. You don't have to look much further than what happened to the Rust Belt after automation and outsourcing of manufacturing led to the loss of low-skilled labor to see what negative effects this can have on a population.

yeah, so learn a new skill or be unemployed....I already said that

 

[libertyyne]

You are bad at economics. Tech advances aren’t reducing the number of jobs in total, they are reducing the numbers of very specific jobs for specific tasks thereby creating new jobs in other tasks necessary to support that new tech and freeing labor to find other new jobs to improve life or meet other needs

It isn’t a net negative

yes but you dont want to be left being the farm laborer when the tractor arrives.

yeah, so learn a new skill or be unemployed....I already said that

Thats literally what OP is trying to be proactive about.

 

[Select All That Apply]

Can you imagine what type of AI would result if it was built on years of learning from SDN commentary. God, what a time to be alive.

 

[deleted480308]

yes but you dont want to be left being the farm laborer when the tractor arrives.


Thats literally what OP is trying to be proactive about.

what OP thinks they are doing and what they are doing are different

They think they are becoming a doctor poised for upper end coding AI. What they will do (if they attempt this) is become a mediocre med student if they even pass that is now out of half the residency options in the country who will regret trading their medical future for a level of coding skill dwarfed by even a mediocre undergrad coding professional

 

[Select All That Apply]

Can you imagine being a type of programmer who just copies and pastes things from Stack Overflow.

 

[deleted1011020]

In 7 months you wont learn anything. I've been coding for about 4 years now and Im no where near the level to even be able to make efficient ML models on my own.


To address the other portion of your question:

Good places to learn include leetcode, coursera, edX, and you can search around on github as well for resources.


If you want to try out what ML models can do without having to deep dive into mathematics (and have a Mac) check out the CreateML application bundled with Xcode.

 

[deleted1011020]

Don't burn yourself out over this though. Programming is not for everyone. Speaking from experience, it can be extremely repetitive and boring if you don't love it. For the foreseeable future, doctors will still be the pilots and machine learning will only help.

As an aside, you learn programming and integrate engineering into medicine at Carle SOM in Illinois. You could check that med school out (they do require a lot of the math that @SeitenKishi mentioned).

Could you elaborate more on this? What integration aside from the fact that its a good CS program is available?

 

[deleted1011020]

Or get EHRs to communicate with each other... Ya know like computers have been able to do since the 90s.

They already have this and is in the works. The issue as always is that EHRs are dragging their feet and regulators dont push for it. Check out HL7/FHIR if you wanna read up on it.

 

[nimbus]

Doctors are powerful enough in this country that they will figure out a way to remain employed and continue to get paid even in the very unlikely scenario that AI completely automates healthcare.

Like lawyers?

 

[libertyyne]

Can you imagine being a type of programmer who just copies and pastes things from Stack Overflow.

You have basically described every programmer.

 

[Cornfed101]

They already have this and is in the works. The issue as always is that EHRs are dragging their feet and regulators dont push for it. Check out HL7/FHIR if you wanna read up on it.

Yes, I understand it is more of a regulation/pride/stubbornness issue than anything else. I used to work for Epic and it was hard enough to get different hospitals using Epic to communicate. Ya know "bureaucracy"

 

[Cornfed101]

Could you elaborate more on this? What integration aside from the fact that its a good CS program is available?

You mean at Carle? They are focused on innovation on much more than a surface level (unlike every other med school I've interviewed at). Everyone learns R so they can use it to improve research. You have access to the $1 billion worth of engineering resources at the University of Illinois. Basically, if you see a problem in medicine they will let you pursue it.

One of the specific examples where engineering is integrated into the curriculum is when learning about medical equipment. The students said that they took apart an ultrasound machine in one of their classes and learned how it worked. Doing things like this helps the students to evaluate what they can do to improve the current technology. Medical technology consistently lags 10-20 years behind technology in the market. Carle is hoping to change that. It seems like a great idea for a med school, and I hope more new med schools follow suit.

 

[7331poas]

yeah, so learn a new skill or be unemployed....I already said that

The point is there are less jobs available for people with "new skills" than there was for people with "old skills". There is no law of economics that says that each job eliminated by computers has been silently replaced with multiple new ones.

Its naive to think that in the far future our society will sustain itself on the backs of 300 million "AI programming" jobs.

 

[Thesimplelifeofamyloid]

What makes you think that a human will be better than a computer at this task by 2060? Or perhaps a better question, what makes you think a physician will outperform a nurse practitioner + AI combination at obtaining and interpreting an H&P.

First off, mever said 60 years. People act like the sky is falling within the next 5-10 years over this stuff. Yes, there are many predictions about the future that are right, but there are many, many that are wrong. You just don't hear about them anymore because they were so blatantly wrong. You could make the same argument about engineers, programmers etc, or basically anything. For example, the easier and cheaper it is to do MRIs , the more images we will order. It's a fallacy to make these projections just solely based on theory. The point is there are many, many factor that go into this outside of just pure innovation.

 

[7331poas]

First off, mever said 60 years. People act like the sky is falling within the next 5-10 years over this stuff. Yes, there are many predictions about the future that are right, but there are many, many that are wrong. You just don't hear about them anymore because they were so blatantly wrong. You could make the same argument about engineers, programmers etc, or basically anything. For example, the easier and cheaper it is to do MRIs , the more images we will order. It's a fallacy to make these projections just solely based on theory. The point is there are many, many factor that go into this outside of just pure innovation.

The opposite is true as well. No one expected the Internet to be invented and to rise in popularity in such a short time frame.

 

[deleted1011020]

You mean at Carle? They are focused on innovation on much more than a surface level (unlike every other med school I've interviewed at). Everyone learns R so they can use it to improve research. You have access to the $1 billion worth of engineering resources at the University of Illinois. Basically, if you see a problem in medicine they will let you pursue it.

One of the specific examples where engineering is integrated into the curriculum is when learning about medical equipment. The students said that they took apart an ultrasound machine in one of their classes and learned how it worked. Doing things like this helps the students to evaluate what they can do to improve the current technology. Medical technology consistently lags 10-20 years behind technology in the market. Carle is hoping to change that. It seems like a great idea for a med school, and I hope more new med schools follow suit.

I mentioned this frequently in my secondaries but I unfortunately didn't apply to Carle due to tuition. If I had known though I would have definitely applied regardless of the cost.

I guess this is a bright side if I have to do another cycle as that school has officially shot up to my #1.

 

[Thesimplelifeofamyloid]

The opposite is true as well. No one expected the Internet to be invented and to rise in popularity in such a short time frame.

I mentioned that in my post, no? Let me ask you this; why haven't you dropped out yet if you think all professions dealing with basically anything will be replaced in a few short years

 

[deleted927529]

What if the sky falls tomorrow? You should pursue something like this out of genuine interest or with a specific career goal in mind, not because you're trying to preempt an AI takeover (highly overblown, IMO)

 

[Select All That Apply]

South Korea's government was an early adopter of information technology and started to subsidize millions into building broadband infrastructure when the internet was in its very early sages back in the early to mid 90's. Broadcast.com launched in 1995 off the premise of being an internet radio company and was acquired by Yahoo! in 1999 for billions. Napster initially launched in 1999 as a file sharing music platform due to its prediction that the internet could be used to spread music as downloadable mp3 files. The notion that "no one expected the internet to be invented and rise in popularity in such a short time frame" is a disingenuous idea when you realize that 20 years later the United States is still trying to catch up to South Korea in terms of reaching similar connectivity goals, streaming of sports and other events was a progression that Mark Cuban wanted Broadcast.com to pursue since its inception, and Napster is a pioneer for the concept of "free to access" media consumption that exists today.

 

[7331poas]

I mentioned that in my post, no? Let me ask you this; why haven't you dropped out yet if you think all professions dealing with basically anything will be replaced in a few short years

To be pragmatic, I suspect that there will be general societal collapse from mass-unemployment long before doctors are replaced

 

[IsleyOfTheNorth]

I, for one, welcome our new robot overlords.