Low Signal to Noise Ratio Info and The "AI" Future

I have a Garmin computer for my bicycle. It tracks a bunch of statistics, like heart rate (with a heart rate monitor) and power (with a power meter) on each ride. Garmin came up with some algorithms to try to calculate "fitness" and recommend training routines, etc... but the output of those algorithms seems like garbage. The default setup of the garmin fitness stuff is to be "invasive" for lack of a better word with the automated tracking. It's pretty tedious.

For example, I can tell how tired I am when I wake up in the morning and then if I head out to ride during the day, I can tell how fried my legs are when I ride down the driveway. If I did a really difficult ride the previous day and I'm not fully recovered, I can tell instantly. The garmin, though, makes some kind of calculation about fresh/tired and presents a number like "+4" which means super fresh, and "-4" which means really tired. Generally the number it spits out doesn't match my feeling at all. I just ignore it.

Anyway, I think people react to the garmin advice and information in two different ways. Some people "believe" what the algorithms spew out. Some people just ignore it, or take the time to configure their gizmo so it is silent. I left it on for a while to see if any of the info is useful, and some of it is, but mostly it's trash.

Most of the info provided by "the science" is exactly like the garmin's info. Data and calculations are involved, but the conclusions are garbage. The vaccine efficacy calculations for COVID were utter nonsense, for example. The initial studies were like 8th grader designed experiments. All the food intake and health outcome studies are very bogus too.

If you look at this class of data analysis problem from the perspective of information theory and a concept like signal-to-noise ratio, it becomes really clear why the conclusions about whether coffee is "good" or "bad" or if vaccines work or not or how much training to do are utter bullshit.

A common problem with these studies is the "signal" can be characterized, but the noise level can't be. Without knowing both things the "information" obtained by analysis of some data is worthless. There's hardly anything that falls into the "knowable", that is determining if a signal is 0 or 1, category in the real world.