Exactly. Even as sophisticated as they are, computer models can only be as good as our understanding of the phenomenon being modeled. It's the climate models that are showing catastrophic positive feedbacks for temperature with an increase in "greenhouse gases". But those catastrophic positive feedback scenarios present in the computer model may or may not exist in the real world, because of interactions that are not accounted for in the model because they're presently not known or not well characterized.
The comparison to modeling the macroeconomy is well taken. Both are complex nonlinear systems in which the validity of computer models are highly dependent upon detailed knowledge of the initial conditions, and in which the information content of the phenomenon being modeled cannot reasonably be captured except in a highly simplified fashion. The same information problem that plagues macroeconomic models (you can't really gather all the information necessary to know the initial state, because there's far too much information required, and even if you could gather it, by the time you've gathered the information, the system's state has changed) to some degree applies to modeling climate, particularly where there are direct interactions between human actions and the system.
Paleoclimate data appears to show we're on the downward side of the peak of the current interglacial, with the amplitude of short term warm periods actually decreasing over the last few thousand years. And the current computer models didn't predict the "pause" in increased global mean surface temperature observed since about the turn of the millennium. The models simply aren't good enough to restructure the basis of the entire global energy economy on.
