They could, but that does not mean they do. There are quite obviously rules and patterns to much of human functioning. Denying so seems like human hubris.
Even if each human displays unique behaviour for a particular trait, knowing that it is so for that particular trait is useful and therefore still amenable to scientific exploration. Even if humans reacted randomly in some situation, the random behaviour would be subject to a probability distribution and knowing it would be useful.
It's hard for me to see where exactly the leap to "it's impossible to study human behaviour scientifically" is necessary, particularly when we have so much evidence to the contrary.
It holds for most of science most of the time that you don't reliably get the same output if you provide the same input (because you don't know all the variables or the entire set of equations). Only when a phenomenon is completely known does this stop being true.
But when is a phenomenon completely known? After all, for a long time we've known classical mechanics to be completely known... Except it wasn't. And during the time we thought it was, you could get into exactly the type of situation you describe above: for the "same" input, you could get a different output, depending on the components of the stress-energy tensor you were not aware were relevant. The effect was subtle there of course, but there are many examples where it's not (e.g. the entirety of biology and medicine).
So I disagree with this description of science.
EDIT: Also, it completely slipped my mind the first time around because it's such a stupidly strong counterargument, but by your definition the entirety of modern physics (quantum mechanics, quantum field theory and beyond) is not science.
The same is true of chemicals, rocks, or lots of other categories of things subject to scientific inquiry. In fact, interesting scientific results tend to come from how the behavior of different subsets of categories like that behave in similar conditions, rather than being made impossible by such differences.