I don't get something. When modelled in software instead, I don't see why a simulating function couldn't use pseudorandom functions and inputs only (nothing tr

I don't get something. When modelled in software instead, I don't see why a simulating function couldn't use pseudorandom functions and inputs only (nothing truly random, no source of input), and, therefore, be deterministic. I don't see how it makes its results worse. But if it's deterministic, I'm having a hard time seeing how actually running it can have a moral side effect, or how that effect is worse depending on how many times you run it. Especially since there are further levels of silent optimizations (such as by the compiler, or in the CPU) based on the abstraction that the result is all we really care about. Would the calculation involved have a moral side effect at one level of abstraction, but not another? What if you cache intermediate results? Final results? It is really hard to give pure deterministic calculation itself a moral component, yet it seems a simulation could in the end be just that. Any solution to this problem?