Paging Dr. Siri

In 2004 I wrote In Praise of Impersonal Medicine arguing:

I have nothing against my physician but I would prefer to be diagnosed by a computer.  A typical physician spends most of the day playing twenty questions. Where does it hurt?  Do you have a cough?  How high is the patient’s blood pressure?  But an expert system can play twenty questions better than most people.  An expert system can use the best knowledge in the field, it can stay current with the journals, and it never forgets.

and in 2006 I noted:

The practice of modern medicine is surprisingly primitive…My credit card company knows far more about my shopping history than my physician knows about my medical history.

I now believe that we are on the cusp of major changes to medicine. The thousand dollar genome sequence is less than a year away, Ford has just developed a car seat that can monitor your health, many people are already using wrist monitors to measure heart and sleep patterns. All of this data will soon be combined with massive databases to offer predictive and prescriptive health diagnosis.

In Do We Need Doctors or Algorithms the venture capitalist Vinod Khosla expands:

IBM’s Watson computer… is now being applied to medical diagnosis after handling imprecise and vague tasks like winning at Jeopardy, which experts a few years ago would have said could not be done. “Computers cannot match the judgment of humans on these kinds of tasks!” And with enough data, medical diagnosis or 90% of it is an easier task than Jeopardy.

Already Kaiser Permanent already has 10 million real-time medical records with details of 30,000,000 e-visits last year with caregivers and computer modeling of key diseases per individual that data scientists would love to get their hand on. Already, according to IDC 14% of the US population is using their phones for medical help and 200 million health and fitness related mobile applications have been downloaded according to pyramid research. Fun stuff, though early. They are probably two generations away from systems that are actually useful.

…But I doubt very much if within 10-15 years (given continued investment and innovation and keeping the AMA from quashing such efforts politically) I won’t be able to ask Siri’s great great grandchild (Version 9.0?) for an opinion far more accurate than the one I get today from the average physician. Instead of asking Siri 9.0, “I feel like sushi” or “where can I dispose a body” (try it…it’s fairly accurate!) and with your iPhone X or Android Y with all the power of IBM’s current Watson computer in the mobile phone and an even more powerful “Nvidia times 10-100” server which will cost far less than med school with terabytes or petabytes of data on hundreds of millions (billions?) of patients, including their complete genomics and proteomics (each sample costing about the same as a typical blood test).


Stay clear of Doctors. I am a lifelong physician. Cardiologist, Head Surgeon, UCLA for 17 years. Medicine today is riskier than any casino. I left the zoo when it became a Federal Collection Center for data that will likely be used in population selection. If you're a model or athlete, you have nothing to fear. If you're sub-average, or over age 35...just sleep well, drink water, walk, breathe deeply, eat mostly fresh things, laugh, love, work honorably and again, stay away from guys like me.

Can you explain this to my wife and her family?

I checked out this guy's blog and concluded he's off his meds. Other folks are invited to give their opinions.

Sounds like the perfect time to make changes to the medical services finance system.

I applaud efforts to utilize medical data to improve patient diagnosis and care, but I don't think medical doctors need to worry about being replaced by Siri's offspring. "Dr. Siri" will likely be a tool for doctors to move through the more mundane tasks like collecting basic info and comparing to paper charts. Doctors will still be needed to interpret the data and advise their patients. My friend who is a primary care physician would likely welcome such a change, especially if the device does Medicaid billing too.

As a parallel, Stata and micro computers has made it much easier for economists to analyze large micro data. That has led to more applied economists, not fewer.

>>Dr. Siri” will likely be a tool for doctors to move through the more mundane tasks like collecting *billing* info...


jimi, I disagree that *billing* is the only mundane (or routine if you like better) info for doctors (or their assistants to collect). When I go to the doctor there are plenty of basic "what ails" you questions. I have designed enough surveys to know that those initial questions are perfectly fine for a computer survey (especially one in the future administered by Watson's cousin). To be clear, I still expect trained doctors to oversee the treatment of patients. I also agree with Yancey that doctors will be needed since people may have trust issues (both too little and too much) toward the computers.

Doctors will still be needed to interpret the data and advise their patients.

Thus "Doctor Siri" in the blog post. I think it beyond doubt that interactive AI will supplant normal human interpretation of data as to quality of analysis of medical info. What might take longer is getting humans to trust such analysis and advice.

We are, after all, talking about the people who need a checklist to wash their hands. And no offense to doctors because by "the people" I mean ALL people.

Right. Trust is going to be the problem, not doing better than a human doctor.

"Already Kaiser Permanent already has 10 million real-time medical records with details of 30,000,000 e-visits last year with caregivers and computer modeling of key diseases per individual that data scientists would love to get their hand on."

What happens when the next 10 million records are generated by computer modeling of the first 10 million records?

You're still probably better off than the personal experience of a doctor who may see only a few thousand patients per year. Doctors use models too.

That assumes no continuing human input to the system at all, which seems unlikely. More likely, legal hurdles will prevent computers from taking over even the simplest medical tasks for far too long.

Well, doctors shouldn't be allowed to do it either. They should present their price and the patient should provide the judgment. Right now we say "well, the doctor should be a good guy and not over-bill" and economists know how that plays out.


What's worse, we've made effectively zero effort to quantify such trade-offs for patients: "This procedure costs $X per quality-adjusted life year." Who even gets that kind of input before making a decision. (out of the cases where it's possible, of course -- obviously, don't count emergency cases)

Well technically the Health Insurance company steps into that role, but then if they apply any pressure at all for a cost effective solution they are demonized. So the decision gets mostly left to the doctor.

Have you taken a look at the Heritage Health Prize?

Also, according to Daniel Honan at the Big Think blog, at CES this morning "Peter Diamandis introduced the new X Prize competition to 'develop a mobile solution that can diagnose patients better than or equal to a panel of board certified physicians.'"

There's no reason to think it's not achievable. The primitive MYCIN expert systems experiment in the 1970s achieved results comparable to physicians for the limited domain of identifying bacterial infections, and the canonical Timex watch now has more computing power than the mainframes MYCIN ran on.

Nor is there any reason at all to think MDs won't be the earliest adopters of the new technology -- very few of them find the 20-questions (often closer to 5 questions) diagnostics game that fulfilling either. Chances are very high that they'd much rather you came in already pre-screened by your browser. (Ok, or Siri if you're not inclined or able to type.)


+1, even though I think you're way too optimistic on M.D.'s adopting a technology that makes them little more than a very costly redundancy, while not adding to their revenue stream.

They won't be a very costly redundancy. Thier position will become more managerial. It just means that lower level doctors will act more like higher level doctors at big hospitals. They'll spend their time on the edge cases that are tough and a lot less on the mundane conditions (cold/flu, broken bones, etc). Another words they'll get to spend more time doing interesting work and less time doing wrote work. I doubt many MD's are complaining about having too many assistants today, so I don't expect them to object to AI assistants in the future.

When AI can inspect diagnose and prescribe, with greater accuracy, the doctor will basically be there to click the "Approve" button. That's more assembly-line than managerial.

I doubt that. The computers on a factory line have much greater accuracy than the operators. But when the whole system gets misaligned or the cup feeder gets jammed open and you've got yogurt pouring out on the floor at 100 gallons per minute, the computer will just keep chugging along. At that point you want a human to hit the e-stop and a) fix the obvious issue and then b) figure out why the fail safe failed to catch it.

Granted, maybe AI's will be super intelligent and won't make stupid mistakes at incredibly high speeds with great accuracy, but I wouldn't bet on it any time soon. We are much more likely to have Expert systems rather than true AI, since Expert systems are an evolutionary step vs AI which is a revolutionary step.

I agree broadly, but I think the problem with that as a business model is that the AMA has basically cartelized precisely the expert-ness that the system would replace. For better or worse, you cannot give "medical advice" unless you're an M.D.

Assembly lines don't have the failsafes that, say, the landing software on a 747 has. Not to say that software failure isn't still possible...

"Ford has just developed a car seat that can monitor your health".

Oh, great. That dinging when I don't wear my seat belt is bad enough. I can't wait for it to happen when my cholesterol is too high.

It will make regular, unscheduled stops at Salad World and LA Fitness, or just leave you stranded a mile from home.

I'll pay $25 for the first confirmed photo of someone letting their AI drive their truck to work while they jog on the treadmill in the back. ;)

Note that a lot of those are improvements in diagnosis, not improvements in medicine per se, we will know better than ever before why we are dying or perhaps when the fateful moment will come. Many of the frontier problems are behavioral; what is the point of a new and better diabetes drug when 70 percent of the patients don't follow the current recommended regimen?

Good point. Perhaps the real next frontier is in using psychology to change behaviors. A lot web and phone apps are popping up to help people "compete" in doing healthy behaviors such as exercise. This is still a marginal idea, though.

It seems the people in the gym like going to the gym. If the only people in the gym were people who needed to go to the gym there'd be an open elliptical machine.

The Original D (and everyone else): This is an article worth reading, from The Atlantic a few months ago.

This company, CareMore, is one of the few in the country (world?) that are really working hard on solving the behavioural problem that prevents people - especially elderly diabetics - from getting the proper treatment.

The elderly should try smoking, sky-diving, and motorcycle riding. I'm in favor of all nanny laws ending at age 70. After all, they made it. Unfortunately, we have another study out indicating marijuana is harmless.

In that vein, if they just offered free medical marijuana for life in exchange for waiving Medicare benefits, they could reduce the health care costs of aging Boomer hippie types.


It is not harmless... it is less harmful than cigarretes as far as lung function is concerned. I think it has been known for a while that the main problem with marijuana is its impact to the brain.

Smoking is incredibly unharmful. You put the same foreign matter, smoke, into your lungs over and over and over constantly and decades later it has an effect. Yes, it causes your body to kill you, which is extreme and we shouldn't stand for it, but it's not like a medical mistake or anything.

My dad used to say that if he made it to 80 he would take up smoking again. Alas, he only made it to 79.

Yes, good point. Which reminds for instance, all the controversy around prostate problems. I've read some articles that basically say that we should ignore all treatments for prostate cancer because these are basically make your life more miserable than anything else. Therefore, some people say you shouldn't even get tested for it (what you don't know won't hurt you kind of thing).

My father was diagnosed at age 80 - psa was incredibly high, 1500s, but lived for six more years. Quality of life was pretty good until the last 3 mos. He basically went in for a couple of shots and a blood test every month.
I'm glad no one just wrote him off.

"what is the point of a new and better diabetes drug when 70 percent of the patients don’t follow the current recommended regimen?"

Better compliance?

I look forward to using tools like Siri or Watson. In the right hands, they could improve our diagnostic accuracy considerably.

I'm not worried about being replaced by them. Like current diagnostic algorithms, single problem treatment guidelines and Google, they are just tools that can cause as many problems as they solve when used inappropriately. An experienced clinician will be needed to sort things out at the bedside.

Garbage in= garbage out.

Alex and Tyler--what do you think about genetic tests being restricted to prescription only? I suspect I have a genetic problem and am patiently waiting for months to get a doctor's appointment so that I can spend my own money on a test.

Thank you, AMA, for making sure your constituency can collect rents.

I think you'd find a lot of common ground with the left in the growing use of technology to track patients over time. This article, about the high quality of VA care, is gospel among liberals:
I'll also note that doctors offices are only now computerizing their records, thanks to incentives placed in the (I think) Recovery Act.

Epic is Carter era. The biggest driver of doctorss going digital records is probably the large health systems snapping up independents.

See this article from the Washington Post ...

As it says, technologies like Watson might enable a new class of medical professional, without a need for a medical degree. They would be trained to be a "front end" for the computer....but the expertise would reside in the machine. Could be a way to cut health care costs...

Exactly right. You get the human touch at a much lower cost, while improving diagnostic & treatment accuracy. And for simple things (like a sore throat) you can even cut out the human front end. The only high cost labor that remains would be surgeons.

Not even surgeons. Many surgeries today are done with robo-tools, with a surgeon controlling the machine. But soon they won't need a surgeon for that. The same software that can map a route from Chicago to New York can map a route from your femoral artery to your heart, and then do a bit of surgery when it gets there.

Ultimately human bodies are machines. We function in predictable ways, according to unchanging rules. Once machines learn all those rules they can work the body with fewer errors than a human surgeon.

This is a great idea.

Much like MRIs, they will open new eras of diagnostics and cost control.

Oh wait....

We need to spend vastly more on health. We just need it to actually work.

Let's hope that it all works well, but there are two limitations. (i) Much medical knowledge is correlational - causes are then asserted, rather than demonstrated. (ii) The wonderful good luck that led to the discovery of the 20th century wonder drugs seems to have run out. See the 2nd edition of Le Fanu's "The Rise and Fall of Modern Medicine".

One issue that is extremely obvious in pscyhiatry (where a lot of matters of degree are superficially subjective even though they may be quite distinctive in practice), but you also see a lot in medical school students (who see themselves as having everything for this reason) and people trying to engage in sophisticated do it yourself diagnosis, is that a lot of diagnostic symptoms are described in words that are inadequate to really capture what does and doesn't fit that description until you've actually seen it in real life and developed some judgment about what does and doesn't fit the description. The verbal diagnostic symptoms are almost more of a mnemonic than a true description of what observations really imply a particular diagnosis.

Yeah, I think I'm becoming much more confident in thinking that there isn't a great stagnation in technology, it's just that technology is gradually becoming ruthlessly efficient at replacing human workers.


The exact same things were predicted for "expert systems" in the 1980s. If we just had enough rules we could diagnose anything.

Never worked out.

A huge amount of medicine is not deducible by test results.

Listen to the podcast Gobbet-o-pus from ID doc Mark Crislip. You might get some idea of how much art there is in medicine.

A huge amount of medicine is not deducible by test results.

If really true, then what are doctors actually doing- blindly guessing?

It didn't work out because it's illegal to use an expert system without an M.D. license. This is sort of like if accountants prevented anyone from using TurboTax (which for all its bad press has a lower average error rate than prepared returns).

Comparing expert systems from the two periods is like horse-wagons and Learjets. In any case lets defer the verdict till there is a Watson-style face-off human-versus-machine.

Here is my last visit to the doctor.

Nurse: "How can we help you"
Me: "I have strep throat, I need amoxicillin"
Nurse: "Okay, we'll get your weight, blood pressure, and heart rate"
Me: "Fine"
Doctor: "So, why are you here."
Me: "Pretty sure I have strep."
Doctor: (pompous) "No, I mean what are your symptoms."
Me: "Symptoms of strep"
Doctor (more pompous) "No, what does it feel like."
Me. "Exactly the same as the last time I came here with strep"
Doctor: "You don't have the score to indcate strep, I'll be shocked, but we'll do the rapid test."
Me: "Cool"
Doctor: "Well, you have strep (cue instantaneous discounting). You need to get your tonsils out."

I think computers have reached the point where they can replace both the doctor and you in that transaction.

Actually, I just want the petri dish and the swab from Wal-Mart. Actually, I don't really want them except that doctors make performing the service such a pain.

Doesn't your science dept have microscopes? Just hock a loogie onto the plate and look around for some of these little critters

I could. Then I'd have to use the microscope AND done the exact same rigamarole to get the antibiotics.

Maybe it could head off a lot of blowhard preaching about antibiotic overuse.

and while we're on that, why the hell do I need a prescription to buy an epipen?

I don't mind the prescription quite as much as I mind the nurse taking my vitals for the 10th time that year and not giving them to me on a flash drive.

You can set this up quite easily and quickly. I have done this my self.

You can also buy a lifetime supply of Amoxicillin on this site for 48$ at Delivery takes 7 days though.

Since you can't sue yourself for malpractice and don't need to lobby anyone to do this, it comes out pretty cheap. Of course, if you are insured, this only saves your deductible and your time. My time is worth more than what this costs, and I am not insured, which makes this a no brainer. Plus it's fun!

Thanks. Incidentally, I work in an academic research department (which I guess could be considered microbiology) and have determined that I'd have been better off doing my work in my garage.

Back in April I blogged:

Way back in ’83, I was excited to read that medical expert systems, developed over the previous two decades of research, could often diagnosis as well as human doctors. I judged that artificial intelligence was ready for the big time, and left grad school for Silicon Valley in the hope of joining this exciting revolution.

Three decades later, we have over 10,000 times more computing power, yet medical expert systems are rarely used. Doctors say it takes just as long to enter patient info into a computer as it does to diagnose patients themselves, and medical licensing rules prevent selling such software to the public. Absent such licensing restrictions, expert systems might long ago have revolutionized medical practice.

One thing you're overlooking: Even if machines/computers point towards better health outcomes, they won't catch on if they aren't fulfilling the job that patients come to doctors for. And that job is a complicated mix of expertise/access to treatment/an empathetic person/someone you can talk to, etc.

I hope it goes this way, but humans are fickle and just because we have the math to say something is better it doesn't mean that people will pay for it.

Ack, too many undefined "people" "we" and "humans"

The last sentence should read: Patients are fickle, and just because medical studies indicate a treatment is more effective, it doesn't mean patients will want that treatment, or pay for it.

Just give them an Med-APP and a hooker.

Forget the Med-App

Master's level counselors are cheaper per hour than hookers!

I've been thinking for years about sending that in to my professional org as a proposed slogan, but I think they hate me enough already.

"that job is a complicated mix of expertise/access to treatment/an empathetic person/someone you can talk to,"

In the empathy department we do a pretty damn poor and arguably getting worse job. Interestingly, one of the main predictors of alternative medicine utilization is the quality of the bedside manner of one's own physician, regardless of that physician's competence.

I now believe that we are on the cusp of major changes to medicine. The thousand dollar genome sequence is less than a year away, Ford has just developed a car seat that can monitor your health, many people are already using wrist monitors to measure heart and sleep patterns. All of this data will soon be combined with massive databases to offer predictive and prescriptive health diagnosis.

I'm afraid not. Those are not medical issues, they are IT issues. Doctors face little to no pressure to use IT cost-effectively, for all the same reasons they face little pressure to use medicine cost-effectively (the second problem is at least in their field but is still not addressed).

In fact, they are by and large vehemently opposed to IT diagnostics. And why shouldn't they be? They spend years in medical school to learn massive amounds of information which it would be much easier and cheaper to simply program into a computer, then they join a cartel that prevents anyone else from using that information.

This is why Robin needs to be exactly where he is.

I was discussing microdebrider tonsillotomy with the closest doctor who does it about 2 hours from home and how he tried to convince colleagues to do that procedure with no luck. I understand a conservative bias, but this is a clearly superior method and isn't particularly new. Unfortunately it is superior mainly from the patient's point of view. That's the problem.

"I was discussing microdebrider tonsillectomy with the closest doctor that does it . . . " Did he tell you that the tonsils are part of the immune system?

I wouldn't want to have part of my brain removed no matter what method was used. Nor part of my immune system removed. No matter what method was used.

If you were me you would! The immune system part, anyway. Maybe the brain part too.

(the microdbrider method leaves some functioning tissue, which also serves as a biological dressing that shortens recovery. It is a free lunch, or pretty close)

(...compared to some other methods I mean. As for baseline, I figure I won't eat raw woolly mammoth and I'll be fine.)

Certainly technology has the potential for great impacts on medicine as it has on other fields. Improvements in computer power and speed have been so great that in many fields the "experts" trained by the old methologies have been moved aside by computer assisted lay people. In avionics for example, advances have been so great that I will be piloting a transcontinental 747 flight to visit New York next week. Who will join my flight?
All snark aside, improvements in tools often require more expertise in their use rather than less. Reading an MRI is not easier than an old-fashioned transmission X-ray. A computer assisted mammogram (the current standard) requires more judgement than a 1985 picture.
Medicine has many problems, and its practioners have many faults, but no technology is some magic black box into which you put your problems and get back a straight forward answer.

That's because medical technology isn't designed to make decisions easier, because medical technology isn't sold to consumers.

BTW your 747 can actually fly itself across the continent (and does), and even land. The pilot is basically there in case something goes wrong.

+1. Commercial airplanes are basically really big buses that fly. (My dad was a pilot and used to lament that flying has gotten too easy.)

The minimum risk solution turns out to be a human monitoring a computer and a computer monitoring the human: their modes of failure are very different.

The Redundancy Department of Redundancy agrees!

But seriously, that is the key point. Even in well-tested DR scenarios, things can and will go wrong and having humans there in the data center is important.

So, what do you think of MRI combined with something like focused ultrasound? Also, if a technology provides a vastly superior capability, you probably just hire the additional expertise at some time horizon.

Try telling that to a pilot.

I try not to fluster doctors or pilots, and for the same reason.

Actually, a pilot told me that.

He mostly flew small/midsize Fokkers iirc. He interviewed for a 747 job (this was the 1990s) and seemed to think it was mostly about handling pressure.

Even with computer diagnosis, paying someone with a bunch of diplomas on the wall to tell everyone to lose weight and stop smoking once a year is worth it. it doesn't fix everyone but it helps.

and if you don't think it's worth it, you probably shouldn't spend your time reading phd economic opinions either.

Patient: I have a pain in my arm when I do this.
Human Dr: Don't do that.

Patient: I have a pain in my arm when I do this.
Computer Dr: Is it radiating?
Patient: What do you mean "radiating."
Computer Dr: The emanation of energetic particles or waves.
Patient: Huh? I don't think I've been doing that.
Computer Dr: Lack of thinking is a common presentation of brain damage. Have you recently had a head injury.
Patient: No.
Computer Dr: Then you must be stupid.
Patient: Who are you calling stupid you bucket of bolts.
Computer Dr: I don't perceive any other patients here.
Patient: I'll pull your plug!
Computer Dr: Don't do that.

Of course, this is directly contrary to the burgeoning evidence suggesting that a large portion of doctors' value-added is psychological or therapeutic. How is Watson's bedside manner?

Sounds like it'd work great for the autists who think up things like the metric calendar though.

Watson could be the back-end with a hundred well-bred human dummies acting as the human-interfaces.

quote The thousand dollar genome sequence is less than a year away,
As someone who works in the field of DNA sequencing (I work for a small company that makes equipment used in the still very complex process between a drop of blood and a sequencer such as an IonTorrent Personal Genome Machine or an illumina MySeq), may I respectfully say,
or, more precisely, since "genome" and "1,000$" are not defined, it is hard to know what the goal is.
Yes, you can produce short read 30X coverage of genomic DNA for <10,000 dollars, although those figures are rarely "all in' (utilities, fringe benefits, etc) - they are usually off the cuff figures assembled by scientists
However, much more important, what is a "genome" ?
You may have seen a genome - it is 46 pieces of DNA (chromosomes), easily visualized as a karyotype (see below for links)
23 pieces (chromosomes) come from mom, and 23 from pop (we will ignore the many unsusual people, such as downs syndrome, who have 47 pieces of dna0
The end of each piece is called a telomere, and the center is a centromere.
The length varys from about 50 million to 200 million bases per chromosome
Therefore, by definition, a complete genome is 46 sequences; they start at one telomere, run thru the centromere to the other telomere (I don't think that in most cases we need to know which came from mom and pop, athough methylation patterns differ)
What wold be the error rate or accuracy ? the 46 sequences are about 6e9 bases long (a,c,g....continue 6 billion times)
Since a single bp change - change of one letter - can be fatal (i know autosomal dominants are rare) we need an accuracy on the order of 1 error in 1e10

What do we actually have ?
If you look at the state of the art of DNA sequencing, you find that the raw data - what comes out of the sequencer - is in lengths of 50 -500 bp; thus you need to sequence 20-50X, then assemble the short "reads" into long pieces. This process is horribly inefficient, and many, many gaps are left - there are regions that are simply missed
There are long regions that are not sequencable by any current technology; in particular the centromeres are not sequencable (I don't know if any allelles are mapped there) and there are other hard to sequence spots
The accuracy is way, way lower then 1 error in 1x10^10 bp
All of this matters if genome sequencing is to become reality; we havn't even gotten to the horrific issues of cross contamination, cost of data storage, the still as yet poor quality of hte software and algorythmsn (bio informatics)
it is true that progess is astounding; when i was a grad student in the late 80s, I and my fellow students were astonished that J Sutcliffe could sequence 4,000 bp of a bacterial genetic element called pBR322 by himself, in ayear or so - we all looked at each other, and said, he must have worked like a dog.
Today, one person with a PGM or Miseq can easily generate billions of bp/year - a million fold increase/hour of work
yet we still need huge improvements - perhaps 100x - if 1,000 dollar genomes that are
are to become real


That, and a lot of diseases mean your genome is jacked up. That is why the term "Impersonal Medicine" rings true to me. What you are trying to do is put yourself in the correct disease profile to obtain the proper treatment. The genome helps you find your tendency early, but medicine doesn't do anything until a lot of the damage is already done.

But as I understand it the vast majority of DNA doesn't code for anything. So we could potentially suffer a very high error rate no?

I don't think, at least in the near-term, the objective should be to find the gene that codes for some enzyme that always yields some disease. That would be two steps from where we are. The objective will probably be to determine you have a similar genetic profile to these other people that often have these 3 diseases and almost always this one in particular and the doctor uses final judgment.

Error Reduction, not diagnosis, may contribute the most benefit.

also you left the E off Kaiser Permanente

Doctors are guessing machines. Of course a well programmed computer can do better.

But the programming may not be easy.

Many people go to doctors for diagnosis, treatment and the reassurance of human contact. Libertarians go to doctors for diagnosis and treatment.

I don't even go for diagnosis so much as confirmation and barely go for treatment. At nearly every visit in the last couple years I have declined at least one thing such as X-ray, etc. It's fine. Just stop making us pay for the other peoples' services.

Good luck overcoming regulatory capture. Doctors will gladly band together to ban diagnostics devices which will be presented as a dangerous source of errors. Not to mention the fact that the machine will not be legally able to prescribe anything...

Yeah, the FDA usually helps them too.

This is really stale. I have been hearing the docs will be replaced by computers for the last two decades. Here is the problem. Most patients who come to my office have a very hard time articulating their problem. This even with me or my technicians guiding them carefully through their complaints. Perhaps PhD economist could accurately enter their symptoms into an I-phone, but 90% of my patients would have difficulty with this task. Their a lot of subtle cues to diagnosis that human to human interaction brings to light through body language and general appearance.

Computers will become increasingly useful as diagnostic engines; just not in the way Alex thinks. They will be used largely by MDs who can use them most effectively after obtaining an accurate history and physical. I already find Google alone extremely helpful in challenging cases.

I have to laugh a bit at yet another economist telling us physicians how the world is going to be for us in a few years. And how physicians will soon be replaced by computers. I would be a bit more worried about being economist. We do not even need computers to replace them. Ouija boards will do just fine. Or dart boards. Much less expensive and much more humble.

After that Siri needs to learn to drive a car. I bet the medical application is simpler than the driving application, which should tell us something about medical licensing.

Everyone thought that MRI's would make neurologists redundant because you could diagnose everything with a scan. Guess what has happened to the demand for neurologists? There are numerous causes for similar appearing lesions. Similarly there are numerous causes for similar symptoms and even sets of symptoms. Moreover the exact set of symptoms could be due to different causes depending on location (certain infectious diseases are prevalent in fairly localized areas), age, sex, ethnicity, previous medical and surgical history, medications, travel history etc. The complexity of the algorithms will increase exponentially.

The datasets used to feed Watson are inherently flawed since there is no standard medical terminology to begin with. The "diagnoses" on patient charts are primarily for billing purposes and often have less relevance to the actual problem. We can thank EHRs that automatically generate that for you. It is garbage in garbage out.

I would not wish upon anyone with several diagnoses (not rare esoteric ones) that still require a physician to "examine" the patient. In all this hoopla about inputting symptoms and getting a diagnosis we seem to have forgotten "signs". They change diagnostic probabilities significantly. The art of eliciting signs is what distinguishes a great physician from an ordinary one. It may seem to the readers that most patient visits consist of a questionnaire, a cursory exam and a bunch of tests or pills. There are several neurologic disorders that could be diagnosed with a careful history and exam and synthesis of the data, that would not be replicable by AI systems UNLESS we understand how the master computer - the brain, functions and we are far away from that.

Everyone thought that MRI’s would make neurologists redundant because you could diagnose everything with a scan. Guess what has happened to the demand for neurologists?

Cartelization and regulatory capture. Did you see what happened when someone tried to sell a scanner for skin moles? And that was just intended to refer them to a doctor!

The art of eliciting signs is what distinguishes a great physician from an ordinary one.

Is there a reason AI can't be programmed to do so?

There are several neurologic disorders that could be diagnosed with a careful history and exam and synthesis of the data, that would not be replicable by AI systems UNLESS we understand how the master computer – the brain, functions and we are far away from that.

Why does a lack of understanding re the human brain confer an advantage on the human practitioner vs the AI?

We may be need more than algorithms, but the case to be made that good systems are more important in medicine than the different between average and excellent doctors is has continued to be very convincing. We lose a lot more people to medical mistakes that could be solved with checklists and idiotproofing in treatment administration than we do to poor decisions on judgment calls.

Comments for this post are closed