> no system in place to enforce up-to-date knowledge

I don't think that's correct. Many states require a certain amount of continuing medical education to maintain a medical or nursing license. It's often not a TON of training and you have some flexibility in what topics you choose to study, but it's something.

Doesn't that apply to any field, though?

Doctors get one thing wrong and then people want to self-diagnose with the internet. It's like a programmer originally thinking a bug is in one function when it's really in another. And then the customer reads a Java for dummies book and thinks they can find the answer.

It does apply to most fields but not all. Notably, pilots are really good at what they do and improve constantly.

It's hard to compare doctors with other fields because a fuckup in most fields will cost money, but a fuckup in medicine will cost permanent damage and lives. If doctors can't live up to a higher standard (and they can't) then we need to minimise the human element in diagnosis and treatment as quickly and aggressively as possible.

Anecdotal but I've personally been saved from a potentially permanently-crippling medical procedure only because my med student friend visited me in the ward and questioned the procedure as it didn't make sense given my condition. If he wasn't around, I would be spending the rest of my life with a mild disability. That's not acceptable, but it happens all the time.

> Notably, pilots are really good at what they do and improve constantly.

This is a fantastic example! I often think about the distribution of talent/competence in fields and that is a very interesting anomaly.

Does anybody know why this is the case? I understand that it's a (fairly) repetitive task, and technology certainly assists, but I think I recently read that there are .2 "incidents" per million commercial flights.

How has an entire industry created a standard that seems to be well above any other?

Because the aviation industry has powerful regulators and an absolute culture of safety first. Every mishap is dissected until a root cause is found, and steps are taken to ensure that root cause never happens again. After a major incident, it's almost impossible for the same incident to occur the same way again.

Nothing like this exists in the medical industry. In fact surgeons and doctors are notoriously anti-process, anti-statistics, and basically don't want anyone scrutinising how they work. I'm not sure how they've been able to get away with it for so long.

Funny thing is, no one programmer can know all languages and systems, and with a short consult into an aspect of a system they're unfamiliar with, would be unlikely to give more than general advice.

Sounds an awful lot like the situation every GP is in... no one human can know everything about the body, so there are GPs that legitimately can't diagnose you.

On the other hand, the person with a particular bug to troubleshoot in themselves has all the time in the world to become their own specialist and do the knowledge synthesis that the generalists don't have the scope for.

I've read way more about my relatively common condition than anyone but the specialist I'm seeing, and I've come up with some things that he was not aware of in the process. Once down a certain condition's rabbit hole, the patient/specialist dialog really should be a give and take.

Late reply, but it often is if you find the right doctor. The problem is the right doctor is often very expensive. All doctors should do this however there are so many quacks that read webmd and think they should be part of the given and take. It reminds me of the thread about words you can use that let people know you know what you are doing. Usually once I mention "peer reviewed" it seems to help.

What makes you think ML will help with diagnosing unusual cases? If anything I would expect ML to make that problem worse, not better.

ML is already helping diagnose unusual cases.

The reason it's unquestionably better than humans is because it's not possible for humans to be aware of every recorded case of every recorded disease and the various ways it presents itself. But as long as data capture capability exists, an ML model can be updated to incorporate all that knowledge on a regular basis.

The second element of this is that the vast majority of doctors just follow a primitive flowchart to diagnose conditions. It's not like Dr House where they're experts on human anatomy and drug interactions etc. GPs are better compared to phone customer service reps running through a script. ML is better than that.