Speaking from my own areas of experience, I've been struck by how much some celebrated methods and results in machine learning, broadly speaking, basically amount to tinkering without any hard generalizable proofs. In at least one case I'm aware of, there was a statistical proof later that made it clear the original researchers were close, but if they had a different dataset they would have come up with a different result. In fact, as I write this, I think I recall that different researchers did come up with different results, and it was in large part due to tinkering on different datasets.
So yes, even in computer science you run into similar things. Maybe not the same, but similar processes leading to similar problems.