Neither would meet the definition of modern science, so what you've done is to raise a strawman and then knock it down.
As far as models go ... they do a very good job of replicating events when fed past data, and as I mentioned above, John Christy - a real scientist rather than a second-rate software hack not worthy of tying my shoestrings much less kiss my feet (since you do argue from authority, and from authority only, I am compelled to point out that in this the software engineering realm, at least, I have a pedigree that makes your achievements look trivial) - has professed himself satisfied with the current state of modelling.
The ironic thing about your line of reasoning - "models can't be accurate" - is *exactly* why climatologists urge caution.
We can't predict what is going to happen as surface temperatures climb, therefore we ought to consider slowing down the pace of this particular experiment.
And, of course, you are further implying that the whole global warming gig is an artifact of inaccurate models.
Which is false. There's tons of observed data that document recent warming, and models must fit that data to be credible.