Most people don’t understand that there are two very different kinds of computer models.
(While I used to be a bicycle mechanic and a bicycle journalist, I’m now a mechanical engineer specializing in finite element analysis (FEA). FEA is basically using a computer to model physical stresses and strains. I tell my parents that I make sure the wings don’t fall off the airplanes, and that’s pretty accurate).
Social scientists often say things like “we’ve created a computer model of the phenomenon and our model predicts X.” At it’s weakest, this means “we did a little curve-fitting!” More often, social science computer models have more rigor than that, but they’re still largely probabilistic. Those models might be right, but in most cases all we know is that they more or less fit the data we have so far. They also make falsifiable predictions; evaluating those predictions is where the science happens.
The models I make are overwhelmingly deterministic—that is, if I get all of the inputs right (within a certain precision), I will get the right answer (within a certain precision). Put another way, the result I get is always the right answer for the inputs I provided. If my model accurately predicts that your beam will bend by theta degrees under force A, it’s also going to accurately predict the bending displacement when the force is A/2 and when it’s A/pi.
Some people dismiss deterministic simulations without realizing this, saying that “it’s just a simulation.” That applies more to probabilistic/stochastic models more than to deterministic ones. Sometimes, “It’s just a simulation” is tantamount to “I think Hooke’s law is about as valid as alchemy”—and that’s an extraordinary claim.
Of course, “I don’t buy the results of your deterministic simulation” can be a totally valid thing to say—or even the only valid thing to say—but to make this a credible objection, you need to either point out an invalid assumption in the model or you need to invalidate the model (prove that one or more assumptions are invalid) by experiment.
Statistician George Box famously observed that “All models are wrong; some models are useful.” That’s worth remembering for everyone including those who, like myself, build deterministic models. But in my field, it would be more accurate to say “all models are incomplete; some models are complete enough to be useful.”
Now, there are probabilistic aspects to my modeling and deterministic aspects to statistical modeling; the lines definitely get (Gaussian) blurred. Most people in these fields are well aware of the distinctions, naturally. But they often get ignored in the popular press and thus in the popular understanding of these models.