If two different x can produce the same y information is lost on the hash transformation. Finding _an x_ for your y is not the same as finding _the x_. That said, Laplace's Demon is, well, a demon. Presumably the devil can get the demon to unspool time to see what the input was and the math doesn't matter, collisions or not.
> Finding _an x_ for your y is not the same as finding _the x_.
Not in all cases, but in the case of NFTs and other cryptocurrencies, it is. If a second private key fulfills all constraints of the original one (like size and all computed results so far), it is functionally equivalent to the original one.
In cryptography either you only need 'an x', in cases that the original x does not matter, or if it does, then you can just as easily find all x of size less than large enough N and find 'the x', it is still solely restricted by time.
Laplace's demon, information loss, Landauer's principle etc. are at most tangentially related to the problem discussed
You understand that isn't the same mathematically as reversing the hash? If you farm the set of all possible x and try them to see what works you do not have an inverse function in your hands.