Hi Peter, I think there is a nuance two disentangle – IAMs are confusingly used in two contexts: 1) models that try to optimize for some economically efficient social cost of carbon (and by proxy, climate policies), and 2) those that attempt to simulate plausible futures. Where Pindyck's writing was mostly about the first, most IPCC work regards the second. Still, I absolutely agree with Pindyck's criticisms – they translate well over to the second category. We tried to cover that massive topic in the section about deeply uncertain factors and Robust Decision-Making, but with so few words, it is difficult to fully address those points.
A further tricky aspect is that of the second type of models, the scenarios that are explored can themselves be misleading or they can limit analysis. Lamontagne et al. (2018) show how a full factorial of input scenarios illustrates that many combinations can lead to the same outcomes. When we don't know how the future will actually unfold, the chosen archetypes clout our assessment.
Another aspect is that the inputs themselves are actually outputs of others. Pielke Jr. and Ritchie (2021) discuss that in Distorting the view of our climate future: The misuse and abuse of climate pathways and scenarios.
All of this is to say: yes, I agree that all models are wrong, but some are useful. Our argument is mainly that through various approaches, we have some understanding of plausible temperature outcomes. We should prepare for all of these to be robustly prepared.