As an engineer, I can say that models are incredibly useful.
You absolutely can demonstrate causal relationships. I've done enough Ab Initio modelling in my lifetime to say that as a certainty.
Complex physics modelling doesn't rely on holding other variables constant. The whole point is to capture the complex interactions.
Engineers rely on models every day. The transistors in the computer you are using right now were designed using complex physics modelling which took into account physical processes you've never heard of.
If I was only interested in modelling the causal relationship between the two variables of human activity and atmospheric evolution while holding constant all other determining variables, I wouldn't need a super computer.
I could write the model out in one line and solve it with a calculator. It would even be correct for the situation where all other determining variables were constant.
But modern climate modelling is more complicated than that and needs supercomputers precisely because it doesn't hold constant all other determining variables.
Is it right to be skeptical? Within reason. When different modelling techniques show similar answers, jaded old engineers like myself take notice even before looking at physical reality.
And when the models start to match reality in the way that last year's climate did, then it's definitely time to pay attention.
