Nowhere in the scientific method are computer models mentioned as an acceptable substitute for empirical observation.
Computer models are fine, if they are validated. Computer models can explain observations and allow useful predictions.
The problem with AGW models is that not only are they - at best - poorly validated, but they contain far too many parameters, they are ad hoc and few of them are robust enough to account for off nominal conditions. Less charitably viewed, they are designed to predict global warming, - either intentionally or not - but not much good for anything else, especially for understanding the real world.