As far as I'm aware, computers are better at diagnosing a very small and very specific subset of diagnoses. If I'm wrong about this and there's a computer out there who can diagnose patients with better or equal accuracy than the average physician, please link me to the study proving so immediately, because I desperately want this in my life, and so do ninety percent of my colleagues. (There's a stubborn subset who wouldn't trust a computer if Asclepius himself descended from the heaven to sing its praises. The rest of us have ten times more work than time and would gladly welcome the help.)

As it is, we can't even seem to find a program that can analyse an ECG better than an overtired and exhausted resident at 4 am, which is really quite a low bar to cross.

Yeah, I'd like it, too. But the only one of which I'm aware is IBM's Watson, and that's still wildly impractical.

Computers are better at diagnosing when the problem is very specific or low/counter-probability.

So, let's take heart attack diagnoses in emergency rooms--a fixed procedure laid down by following statistics is WAY better than human judgement.

Generalized abdominal pain--computers aren't so good. There are a lot of possibilities and most of them are about equal. The doctor is more likely to tease out the problem over time through conversational randomness.

Some obscure genetic disease--a computer is going to be FEROCIOUSLY better than a human. A computer can keep all of those .00001% options in mind as tests and diagnoses rule out the far more common problems until the low-probability things pop up into statistical relevance.

You think software for those computers that are better at diagnosing was created by engineers who don't know the difference between a stable and unstable sorting algorithm?

There is a massive delta between knowing that, and being able to code examples of each from memory on a whiteboard during an interview.

Also, it may surprise you, but there are many highly experienced and extremely capable programmers that didn't study computer science in college. Filtering those people out because they can't answer programming trivia questions is, at best, hugely short-sighted.

If you're writing diagnosis software, your approach to sorting should be "use sort()". Sorting is not your core competency, diagnosis is.

Actually, yes. It's very likely that the core algorithms for these systems were written by "crossover" types (e.g. math PhDs) who are perfectly capable of getting up to speed with the esoteric aspects of these systems, without having to go back and master the standard undergraduate CS curriculum.

It's called being flexible and able to learn, and it's the single most important thing you're supposed to get out of your formal education.

I have discovered over and over that most genuinely good things are created by one smart person. After that person creates it, then the masses can run with it.

Bittorrent is a great example. All the pieces had existed for a while, but Bram Cohen put them all together into one place. After that, lots of people created other BitTorrent clients. However, without that first, smart person, it wouldn't have appeared.

Only in certain, highly procedural use cases. But in the large, no.