As for your first paragraph, I don't think you can reasonably treat causation like an everywhere-binary like that. Not for arbitrary choices of X and Y.

Take "hunger" and "war" for instance. There are many causal arrows pointing from war to hunger, and several pointing the other way too. You have to zoom in to specific details before the language of causation starts being useful. "good for your health" and "being kind to others" are similarly aggregate concepts.

Causation is a myth, and a damn good one. We use it to reason about nearly everything. But the tendency to use it to reason about actually everything is a dangerous one.

Most of the time you can't "hold everything else constant". We usually have to settle for frobbing X back and forth many times and letting its continued correlation with Y convince us that there is indeed causal wiring between the two. At the end of most experiments is an inductive leap of faith that is justified by the high frob cout.

But sometimes X doesn't go back and forth. Sometimes it's a one-way deal and you only have one of them. Like maybe X is "finding carbon trapped in the earth's crust and releasing it into the atmosphere", and Y is "a global change in climate that poses an extinction risk to humans." In cases like that, we don't have the luxury of waiting for enough induction to bring about the leap of faith.

The situation with kindness and heath is similar. Ideally we'd take whatever wisdom we can from the correlative data that was presented and run with it--maybe there's a cause in there and it would make us healthier, maybe not. Instead we have this artificially high bar for argumentative strength, and when an argument fails to meet it we throw the baby out with the bathwater.

But I am proposing that we should have some skepticism when importing an explanatory style that has worked well in physics and expecting it to perform equally well everywhere. I think we're likely to make mistakes of this sort:

If a cat's head appears from behind the couch, and then later its tail becomes visible--you wouldn't say that the cat's head caused the cat's tail--they're just different parts of the same phenomenon. We're familiar with cats, so we don't make this exact mistake, but I think that we are quite susceptible to misapplying causation to things that we don't understand well, like happiness.

Years ago I had this idea that many of our ideas about causation are flawed in this way. So I developed a habit of taking a causal claim that seems true, flipping the arrow around, and then testing the mutated hypothesis--just to see if it worked backwards. I was surprised by how often it did.

I don't have the data to convince you that the habit of mistrusting my instincts about causation and using "deep down, it probably goes both ways" as a heuristic has caused me to reap benefits that I would have otherwise ignored, but I suspect it strongly enough that I plan to keep doing it.

So the difference between the alternative perspectives is that if you're skeptical about the universal utility of causal nitpicking, you'll learn to recognize when you're wasting your time trying to prove that the laws of physics demand whatever you've noticed. You'll bail sooner on perspectives that don't work, and you'll be more creative about finding new ones.

On the other hand, if you demand a certain style of causal argument, you're more likely to double down against the opacity of human behavior to other humans. You'll design experiments and write articles that inspire further nitpicking about which is a cause and which is an effect, and the matter will remain in stasis--perhaps indefinitely--where its capacity to positively affect people's heath is limited.

I'm not saying that we shouldn't try to uncover underlying mechanisms when we can. When that's in cards it's pretty great. I'm just saying that we shouldn't always expect results of that sort to be within reach, and that reasoning from correlation alone can get us further than we tend to let it.