Ask HN: Is it time for AI to start replacing doctors?

In The Checklist Manifesto, Atul Gawande wrote that he

...came across two professors who studied the science of complexity, which highlighted three types of problems in the world: the simple, the complicated and the complex. They are defined as:

Simple Problems – “are ones like baking a cake from a mix. There is a recipe. Sometimes there are a few basic techniques to learn. But once these are mastered, following the recipe brings a high likelihood of success.”

Complicated Problems – “are the ones like sending a rocket to the moon. They can sometimes be broken down into a series of simple problems. But there is no straightforward recipe. Success frequently requires multiple people, often multiple teams and specialized expertise. Unanticipated difficulties are frequent. Timing and coordination become serious concerns.”

Complex Problems – “are ones like raising a child. Once you learn how to send a rocket to the moon, you can repeat the process with other rockets and perfect it. One rocket is like another rocket. But not so with raising a child. Every child is unique. Although raising one child may provide experience, it does not guarantee success with the next child. Expertise is valuable but most certainly not sufficient. Indeed, the next child may require an entirely different approach from the previous one. And this brings up another feature of complex problems: their outcomes remain highly uncertain. Yet we all know that it is possible to raise a child well.”

For the longest time medicine has been part of the complex problem set. Doctors (very rightly) are highly trained with a minimum of 4 years of training followed by years of residency. The profession is seen as highly complex where they need to combine their book learning, with hands-on training and decades of experience to diagnose and treat a wide variety of health conditions. However in Low and Middle Income Countries, where doctors are in short supply, can some level of healthcare be provide by applying Artificial Intelligence to the SOAP (Subjective, Objective, Assessment, Plan) Note along with a touch of the Bayesian Priors?

I think most people can relate to the horrible experience of calling in to a bank or insurance company and being forced through their chatbot voice recognition system instead of getting to talk to a real person.

If we can't even do something that simple (granted I understand the goal with chatbots is to prevent people from reaching a human rather than actually helping them) how could we possibly believe that an "AI" doctor is a good idea? Find a simple to moderate complexity consumer interaction that AI does well, and maybe we can begin to think about how it could help on healthcare. For now, it will only lead to the hellscape we see in other areas of customer service

Let’s say starting tomorrow we decided to pay all doctors in the US $0 dollars a year. How much would the US actually save an annual healthcare expenditures? Approximately 8%

Doctors salaries are not a major driver of healthcare expenditures - they’re just an easy target because doctors tend to be the most public facing part of healthcare. How many health care administrators do you meet when you go to the hospital?

Also, even if we were to use AI to replace a doctor, the for-profit hospitals would pocket the difference as profits.

For-profit medical services are inhumane, immoral, cruel, barbaric at their core.

Every time I describe the US medical industry and medical insurance industry in Europe, people react with disbelief, disgust and horror.

The amount of mass propaganda needed to keep this monstrosity in place explains a lot.

> Many doctors, from my observations, function as pen pushers in what seems an administrative bureaucracy, relying as often as not, on remembered-information rather than improvisation or critical thinking.

This contempt towards doctors never fails to impress me... I don't see it aimed at other professions either.

I am unsure, but maybe because we go in expecting more attention and all we get is an automated and canned response from the doctor, and this makes us think that an app could do it as well.

>contemptuous towards doctors

If it seems so in this circumstance, you've misinterpreted. I am contemptuous of certain aspects, but not blindly of all doctors. Not at all. You can take my word, or ask me questions to clarify.

There's a big problem with healthcare and doctors are only part of it (pharmaceutical corps, nu?).

Our society isn't kind to those who fall behind. We need a way to help people without indenturing them. We need a way to help others by evaluating them more earnestly and less capriciously. We need a lot, and the current system isn't keeping up. Clearing the path to better health and less suffering may require changes that are threatening to some. Or perhaps we could go full euthanasia and market it as compassion.

I am also not a technologist. Far from it. I do use tools though. I also observe things, some through difficult experience. There are places I absolutely do not want AI. So as I wrote; why not use it in at least one way that would have tangible benefits to the majority?

There is a lot of administrative bureaucracy that forces doctors to push pens.

Recently a doctor prescribed me some medicine, and I went to the pharmacy to pick it up. Apparently the insurance balked at paying for it (the pharmacist told me this was common), so the pharmacy sent some additional paperwork to the doctor to fill out.

I know multiple doctors (in the US) that are so fed up with "the system" that they refuse to deal with it, operate their own self-pay practices, and refuse to deal with insurance companies in any capacity.

Even the kindest, most talented, and most dedicated doctors are representatives of a system that seems pretty well rotten to me.

If you're referring to HN specifically, I fully agree. I've thought about starting an "ask HN" about it, but I figured I would get downvoted to oblivion.

The issue is that a for-profit medical industry squeezes doctors out of their position of healers and pushes them against their Hippocratic oath.

They become the face of a malignant industry aiming to maximize profit by extracting the maximum market price under the threat of sickness, suffering, death.

Check expert systems, there were already experimental good medical experts systems in the 1980'. None of them ever hit the market, because medicine is an incredibly difficult market to penetrate due to regulations, but also the structure of power: When doctors can vet what they buy or use, nothing that they perceive to weaken their power can succeed.

One area you can try is alternative medicine, such as naturopathic medicine, which is a social group much less powerful.

U.S. jurisdictions that permit access to prescription drugs to naturopathic medicine: Arizona, California, District of Columbia, Hawaii, Kansas, Maine, Montana, New Hampshire, Oregon, Utah, Vermont, and Washington.

This post is extremely stupid, reeks of ignorant arrogance. And one like it seems to pop up every month.

Why don't we ever folks like this ever ask, "when will AI replace SWE?"

Afterall SWEs deal with deterministic systems that follow well-understood laws of physics, while doctors deal with humans with all their capricious complexities. By any logical thought, software engineering is orders of magnitude more replaceable by AI than doctors, yet the people writingb these posts don't ever consider themselves threatened to be replaced by AI. Arrogant ignorance.

>Why don't we ever folks like this ever ask, "when will AI replace SWE?"

People ask this all the time.

Ok, I'm extremely stupid. I confess.

Now please offer an intelligent, insightful, un-enraged response. Maybe with examples. And if you're truly brilliant, include something that considers the patient, in some way, preferably constructive, figuratively.

Your "observations" are ignorant, condescending bullshit and thus the entire premise of your question is wrong. Go spend a few years working with physicians in an actual healthcare delivery organization and you'll understand how it really works.

AI can be somewhat helpful in medical research and clinical decision support. But compared to simpler deterministic algorithms it is a marginal improvement at best. In particular AI sucks at gathering actionable clinical data from patients.

Harsh. But unimpressive. There are things that don't require years to understand.

And if after those years, one still does not understand, maybe they don't care to. Do you?

Regarding condescension, I humbly suspect some ignorance on your part too, not for the tone of your reply, but for what I've seen some doctors do and not do, say and not say, and... your unwillingness to consider. Your statement followed by "thus" is very much beneath your intellectual acuity. I assume I hit a sore spot, but my intent here was sincerely, if you'll entertain the possibility, to read through a hoped-for extensive discussion. You're actually among the type I was hoping to receive (slightly less confrontational) insight from.

No offense taken and insight well taken, though I could use a bit more.

I'm not feeling well at this point tonight and will probably end my replies here.

> Most people, regardless of their politics, understand that healthcare, notably in the US, is too often unaffordable and sometimes destructive through financial means.

The US is an outlier. There are countries with a similar and a much higher GDP/per capita that pay almost 50% less.

https://www.finder.com/healthcare-costs-by-country

I’m a programmer, and my dads a doctor, and for ten years I’d always tell him that something like 20 questions could do his job.

After about 5 years he did this graduation speech and in it he referenced that: “yes, AI is getting great, your phone is a supercomputer, but the truth is that a computer will never be able to hold the hand of a dying patient and tell them it will be all be okay.”

I think AI already has supplemented doctors. Think about all those prepandemic articles about how early/bad the flu season was based on number of Google queries on "flu symptoms".

And be honest: you've googled some set of symptoms to see what you've got, probably before deciding to see a human doctor. No, you don't have 5-alpha reductase deficiency.

Does this contribute to patient's cost reduction in any way?

Patient cost is a function of how much money is poured into the system divided by the number of potential patients, aka insurance subscribers.

For the first time, I recently had a teledoc consultation and received a prescription as a result. Aside from the fun of getting the camera and mic to function in Linux, it was amazingly easy and efficient [1].

I answered a series of automated questions that would be reviewed by the physician, all far more extensive than anything in my live interaction with the actual doctor. This indicates discretion as the greater force of determination.

The cost was comparatively reasonable, but a machine could have procured the same results at pennies on the dollar.

Anyway, if AI does not eventually pull the rug from physician's feet, telemedicine will certainly shift it a bit. Hopefully enough to get people thinking.

1. My efforts with the IRS interface were futile. Nothing worked, which is odd considering that the browser was the only critical bit. Mozilla. But on spooky Linux

You say that doctors are pen pushers in administrative bureaucracies and come to the conclusion that doctors need fixing and not the bureaucracies?

That seems like a very odd and backwards conclusion which reeks of treating a symptom instead of the root cause. I think you may have missed the forest for the trees.

Doctors are people. Their labor is labor. Unfortunately for those not in the health industry, this labor is expensive, regularly not optimal, and might be surprisingly, easily, bolstered or replaced by AI, or even a nurse, or in rare situations, self help. Regardless, it's unobtanium for many.

I'm going to directly declare it now -- this is my last comment/reply here.

Thanks everyone, or someeveryone

How about replacing the administrators with AI instead, that seems like a far better solution to me

You can even replace the administrators in all the health care middlemen companies too! More opportunity for cutting the fat if you go for administration

The SOAP note (an acronym for subjective, objective, assessment, and plan)is a methodology for documenting a patient's condition on first examination by a medical professional. It serves as an excellent cognitive framework for medical professional to assess their patients.

The SOAP Note originated from the early work done by Dr. Lawrence Weed (an American physician, researcher, educator, entrepreneur and author, who is best known for creating the problem-oriented medical record as well as one of the first electronic health records).

SOAP notes are commonly found in Electronic Medical Records as a means to succinctly communicate patient information in a standard format and are particularly useful between hand-offs between medical professionals.

Modern Electronic Health Record (EHR) Systems offer a great deal more in terms of adding additional diagnostic information as well as a longitudinal patient record but the basics SOAP note could prove immensely valuable as we think about how to design a system that can empower Community Health Workers (CHWs).

SOAP Note in Practice (highly simplified for an illustrative example)

Situation: A medical professional encounters an individual who seeks or is identified for medical care

Subjective data collection

Shaking chills

Muscle ache

Fatigue

Objective data collection

Fever of 102 F (suing a Bluetooth connected thermometer or a cell phone camera based temperature detector)

Malaria Rapid Diagnostic Test (RDT) administered - shows positive

Assessment

The Subjective and Objective information is collated in the medical professionals mind and and she assesses that the patient shows indications of Malaria

Plan

Based on the Subjective and Objective and informed by the Assessment, the medical professional develops a Plan that is specific to the situational context - in this case it may involve sending the patient for a confirmatory Malaria confirmatory test or in some cases starting them directly on anti-malarial treatment.

The SOAP note allows a complicated problem to be broken down into components to then become a simple problem. However, in Global Health today, this complicated problem is treated as a complex problem because it is felt that only a doctor with decades of experience and a medical degree would be able to accurately diagnose and treat Malaria.

I have a few questions for you / your AI? 1) what if the malaria test costs a lot - should i do it on this patient? Or save those resources for a different patient? before this patient ever arrived, we need to decide if should we spend $20 on the malaria test kit…or treat 20 kids with the mumps,measles,rubella vaccine for $1 each? 2) what if the test is negative? now what do i do? could it be a false negative? how do I decide? 3) what if the patient has a headache as well? could this be bacterial meningitis? 3a) if so, should I give them antibiotics now or after a lumbar puncture? 3b) could they have both malaria and bacterial meningitis? should i do a lumbar puncture to confirm or rule out bacterial meningitis? but doing an LP in the presence of cerebral edema / raised intracranial pressure can cause severe brain damage and death - should I do a CT scan of the brain to rule out raised intracranial pressure first? and so on. There are some conditions that are so clearcut and some tests that are so effective, that you could safely diagnose it algorithmically…and a whole bunch that you can’t

The malaria RDT costs about $0.25. To be clear this is a screening test - not a diagnostic test. An actual diagnostic would involve taking a small blood sample, smearing on a slide and looking for malaria plasmodium under a microscope. Not expensive but time consuming and requires a microscope and a trained technician.

There any studies looking for differences in outcomes between patients cared for by physicians and those cared for by other highly intelligent, capable humans with fewer years of specialized training in healthcare.

It might be interesting to examine some of this literature in the context of this thread.

I think it might happen in diagnostic specialties like pathology and radiology. But in situations in which "speaking to patients" is required - many of whom cannot write, much less use a computer, you need nothing short of AGI.

An if statement can replace doctors. But then insurers, hospitals, and drug companies would lose money.

Doctors have very complicated jobs.

They need to deal with patients expectations and understanding of their own diseases

Like self-driving, the 1 mistake AI made would be global headlines, people are still too biased

It should be easy enough to maintain a parallel record of malpractice vs the bogeyman. We can't mold our existence to the standards of headlines -- not if we want any valid reason to exist at all.

This is a profound misunderstanding of both healthcare and the current state of AI.

I doubt you'll ever find an insurance company willing to underwrite that risk

I’m a physician, medical oncologist and researcher with a previous background in software engineering. I happen to hear your points often and always reply the same way: if it is that easy and cost-saving, why don’t you build a tool that does this? The healthcare market is huge, the knowledge basis is more or less the same for all countries and almost all information is available online for free. There is a plethora of startups developing symptom checkers, supportive tools, clinical decision support systems and much more. I’m testing a lot of these systems and helping companies to realize their wishes, still in my opinion those systems are ages from what doctors are able to do. The argument has been made before, that physicians make up only a fraction of the total cost of healthcare. There is plenty of data supporting the fact that physicians even save money for healthcare systems and serve to avoid unnecessary, expensive and possibly harmful diagnostics. So if you are able to develop a tool that is competent enough to talk to you for 2 minutes, perform 20 secs of auscultation and correctly diagnose and treat your common cold while excluding signs of potential worse underlying conditions, you will make millions of dollars. But because the problem of turning all the available medical literature into data that can be acted upon, bridging the gap between a device and a human being with more than cameras and voice recognition whilst being ridiculously cheap and easy to maintain I’m quite confident even my grandchildren can be doctors (if they want to) and almost all of their responsibilities will still be performed by them. I assume you are a healthy human being and luckily you weren’t forced to consult doctors for severe healthcare problems, but I assure you, once you happen to touch problems we assume to be even slightly more complex than the healthy population has to deal with, you will be happy that a human doctor is making the decisions, answers your questions, performs diagnostic tests and treatments on you and talks to your relatives. And if your are able to develop a tool that frees me from even 50% of the administrative bulls… I have to deal with, I will more than happily buy it. TLDR: medical problems are more complex and inaccessible by non-human objects than it is commonly acknowledged by the layman. If you have a credible solution to a fraction of the problems we face and are able to sell it, you will be a millionaire.

I wanted to thank you for taking the time to reply, especially as a physician.

I had written a reply, but abandoned it due to personal matters.

No, no, no!