Models are dubious when you already know the answer
As Dr. Diffenbaugh said, we have to use models instead of running actual experiments on the climate. And models can be quite useful and accurate, as witness the models we use to predict the weather every day.
However, unlike the weather, which essentially gives us an actual new experiment every day to test the weather models, we've only got one climate history. Most of the climate data is spotty at best. We are reduced to using things like tree rings to estimate temperature when what we really need are actual temperature measurement over the surface of the earth, deep in the oceans, and up in the atmosphere. And these models contain all kinds of parameters. If you don't like the answer, just twiddle the dials until you get the answer you like. Since you only have one climate history to duplicate in your model, that's not hard. As scientific proof, it's of limited value.
People forget that the Ptolemaic (earth-centered) model of the solar system was actually quite accurate in predicting the movement of the planets as seen from the earth. Over the centuries, its "wheels-within-wheels" model had been tweaked to the point that it worked. Long before we could send satellites into space, the Copernican model was eventually accepted only because it was conceptually simpler and was nicely explained later by Newtonian physics. The point is that just having a model that predicts observations is not the same as having an explanation.
