This is just silly. Of course you have to understand a process to emulate it, but there would be no purpose to emulations if they did not produce predictions. The whole point of emulation is to learn previously unknown things in an environment that is safer, cheaper, faster, or more ethical than the "natural" environment.
The error here is assuming that understanding a process automatically confers understanding of all possible outcomes of the process. That is a key misunderstanding. Even designed things do not spring fully formed from the forehead of Zeus. Even designed things evolve.
There's nothing to dispute WRT the above. It is my thinking exactly.
But the point seems to be that you can't produce reasonable predictions from an emulated process if you can't emulate the process because you don't understand it in the first place. You do have to understand something about it, or you wouldn't be able to simulate it. You might say, "Well, whut the hey, let's just go ahead and model it anyway." But how good would the resulting information be? What would it really tell you? What value would the findings have as a "prediction" of anything?
But IMHO there's an even deeper issue involved here. Biological complexity may not be reducible to digital emulation at all. You can perhaps isolate pieces of that complexity and study it; but isolating a particular piece -- abstracting it from its place in the whole -- may give you a false picture because the piece needs the whole to express its function in the total system -- which I think is probably the thing one is trying to get to by means of digital simulation: The piece's function in the total system.
This is simply an attempt to render into language an intuition about complexity. It probably isn't sufficiently "scientific" from your point of view. I will, however, be studying up to see whether I may be able to put the idea in a more rigorous form, in due course.