Does using a probabilistic model for a real-world event make it harder to identify its causes? I recently read this odd critique of statistics (the author calls it a critique of probability theory, but I think he doesn't understand the difference probability theory and statistics).
http://mitchgordon.me/math/2021/04/02/probability.html
In this question, I'm concerned with this specific critique (I already asked another question about the remaining ones):

by hiding the causal factors of an event behind the abstraction of a “probability distribution” we deprive ourselves of the ability to identify when those causal factors change and our assumptions no longer hold (i.e. the distribution shifts). [..] For example, I may experiment with a coin and decide that it is fair when tossing it onto a wooden surface, only to discover later that the coin is magnetized and slightly biased towards heads on metallic surfaces

Isn't this backwards? Actually, statistics allowed us to understand a great deal about a physical system, without having to know the precise laws which determine the behavior of the system, similar to what happens in statistical mechanics. What am I missing here?
 A: Mitch Gordon confesses that he doesn't exactly understand the role of probability theory in science and you should believe him. IMHO he gives adequate evidence of his confusion in the link.
If you think you may have a biased coin, a probability model for a fair coin
is a hindrance to checking the coin for fairness only if you're too lazy to
toss it repeatedly in an effort to find out. In fact, significant departure
of the results from tossing the coin from the probability model for a fair coin
would be how you would evaluate
evidence that the coin might be biased.
For example, 489 Heads in 1000 tosses
would not provide significant evidence of unfairness because there is about
a 50:50 chance of getting a number of heads as or more different from the expected 500.
2*pbinom(489, 1000, .5)
[1] 0.50666

However, 459 heads in 1000 would provide statistically significant evidence near the 1% level either that the coin is not fair$-$or that the method of tossing is is flawed.
2*pbinom(459, 1000, .5)
[1] 0.01038813

You may find this link
of interest. It mentions some thoughtful approaches to testing for randomness.
