Quantum mechanics is not magic. Magic specifies the outcome, but not how a system evolves to reach that outcome. Quantum mechanics has precise equations describing how a system will evolve over time, but is famously bad at describing the outcome.

By the same token, we can see that thermodynamics and conservation laws, while widely accepted, are magic. I have heard legend of a deeper magic known as “Lagragians”, although knowledge of that lost art remains confines to the warlocks’ ivory tower.

https://xkcd.com/2904/

The AI developers understand how AI works, but that does not mean that they understand the thing that the AI is trained to detect.

For instance, the cutting edge in protein folding (at least as of a few years ago) is Google’s AlphaFold. I’m sure the AI researchers behind AlphaFold understand AI and how it works. And I am sure that they have an above average understanding of molecular biology. However, they do not understand protein folding better than the physisits and chemists who have spent their lives studying the field. The core of their understanding is “the answer is somewhere in this dataset. All we need to do is figure out how to through ungoddly amounts of compute at it, and we can make predictions”. Working out how to productivly throw that much compute power at a problem is not easy either, and that is what ML researchers understand and are experts in.

In the same way, the researchers here understand how to go from a large dataset of breast images to cancer predictions, but that does not mean they have any understanding of cancer. And certainly not a better understanding than the researchers who have spent their lives studying it.

An open problem in ML research is how to take the billions of parameters that define an ML model and extract useful information that can provide insights to help human experts understand the system (both in general, and in understanding the reasoning for a specific classification). Progress has been made here as well, but it is still a long way from being solved.

When I was learning Japanese, I came across a sentence along the lines of “lets buy stuff at the <shoppingumouru>”, I could understand most of it fine, but didn’t recognize bracketed word, which was conveniently written in a script that denotes loan words (and I have trancscribed phonetically abovr). I probably spent at least half an hour trying to look up “shoppingumouru” simce I couldn’t find it in my dictionary. Eventually, I turned to Google translate and immediately facepalmed when I saw the answer.

The actual difference between a working new mouse and a failing double click mouse is in the button itself (mechanical parts are almost always the problem).

However, it is not some exotic failure mode. All mechanical switches have a “bounce”, where the contact makes and breaks a few times before settling into the connected position. Switches are typically designed to make the actual contact spring loaded (which is the origin of the click sound you here). As they age, this mechanism degrades, making the bouncing problem worse.

However, this is a well understood problem that any electrical engineer should be familiar with. One solution is to install a filter capacitor. Now it takes longer to switch between the on and off state, so the inherent bounce in the switch is smoothed out to the point where you cannot detect it.

They probably did testing with a new switch, and decided that they didn’t need to include any explicit debounce component, ignoring the fact that the switch would degrade over its lifetime.

The annoying thing is that fixing the double click is stupidly easy. Years ago, I got frustrated with that exact problem (after a string of 3 mice that each lasted only a few months); so I opened one up and soldered on a random capacitor I had lieing around.

Capacitors like that cost literally less than a penny, and are no more complicated to install at production time than any other component already on the circuit board.