Putting on my computer based modeler hat, I will say the climate models are useful tools, with two caveats. (hey, I told you I've some experience here, full disclosure of course I'm going to promote the modeling industry) One, that they are used for their intended purpose. Two, that they are used within reasonable constraints.
Unfortunately, what passes for "climate science" today meets neither of those criteria. The models should be used to further our understanding of climate. They should only be run/iterated out a relatively short time, say single digit level years... However, people with a money/power/political agenda are trying to use them to "prove" pet theories (that coincidentally shower themselves with money/power/influence) by running them out dozens or even hundreds of years into the future.
That just does not work. Well sure, you can do it. Sure, you can hack in artificial constraints on your model to keep it from spinning off into unrealistic regions like the entire planet freezes solid or becomes hotter than Venus... But the fundamental problem is, every step/iteration of the model depends on the preconditions (state) and produces post conditions (updated state). The updated state becomes the input for the next time-step iteration. For better accuracy, you want small time steps and a small "mesh" in order to capture those fine-grained interactions. But then you start running into precision problems with your numerical representations. Worse, slight inaccuracies or uncertainties in your results get magnified as they become the input to the next iteration. For a few hundred or even a few thousand iterations looking a couple of years down the road this is a manageable problem. After several tens of millions of iterations to look dozens or hundreds of years down the road you're writing fiction.
So fundamentally, what the models are being used for and how they are being used is just plain wrong. Wrong as in fantasy bearing little resemblance to reality. But there are three other fundamental areas where the models suffer from numerous individual problems.
One is simply our understanding of the climate. As the author points out, we really don't understand with certainty more than just the gross interactions. Unfortunately, the subtle interactions apparently also matter. They matter more and more the further into the future you iterate. This is where the climate models could and should help our understanding of the climate - if/when used realistically. Test a little, measure a little, correct ... lather, rinse, repeat. The models are useful tools for testing theories on what factors are significant, how significant, and how they interact. They are not yet predictive in any way, not even close.
The second fundamental problem is the models themselves. The number of variables that have meaning are at least in the hundreds, probably in the thousands or even millions. (this goes back to the problem above - we just do not know what is or is not important, or how important things are) The number of variables and possible combinations and interactions are basically mathematically and computationally intractable with existing technology. Even with our limited understanding, you have to make simplifications and assumptions in order to create something that will actually fit on and run to completion on existing hardware. Consider, if you want to make a 1 meter mesh over the entire "playing field" - say from the upper edge of the atmosphere down to the bottom of the ocean... That is roughly 5.7 times 10 to the 19th power cubic meters. That's 57 billion, billion data points. Oh, and at each one you need to keep track of hundreds, if not thousands of variables. Then virtually every one of these has to interact in complex ways with it's 26 neighboring cells for every time step...
The third issue is genuinely intractable and basically unsolvable. Even if people started using the models realistically. Even if we figured out all the thousands of variables and how they interacted. Even if we had the hardware to run the models on... There is still no way to predict the climate future dozens, even hundreds of years out. Random chance will, that is will, not may, shake things up. Earthquakes, volcanic eruptions, meteor strikes, and other unpredictable events will almost certainly occur and will have some effect, slight or significant. We just don't know, can't know, and therefore cannot claim to know what will happen based on some computer model.
When the models do not match the measured data there are corrections made to te measured data. The measured data for just over 100 years has been adjusted where the global warming is equal to the adjustments.