This is both helpful and scary.
It is helpful because it gets all of the info we can gather into a single compatible framework that allows for better development of insights, plus it helps us learn where we need more or better data.
But it is scary because it will not be completely correct, and will probably be used to make decisions that are problematic because people tend to over-estimate the accuracy of such models and underestimate the ranges for potential errors in its results.
Think about how long it has taken to develop weather models that can predict even a few days in advance with useful accuracy. That required a lot of trials and tuning for decades to get to where we are today, which is still far from perfect for planning even a week ahead.
Now consider how many additional variable parameters are involved in long term climate modeling, and realize that we can't make a prediction and wait to see how it went wrong so as to make the model better. The opportunities for tuning are much more limited, so the accuracy will also be much less certain.
One test will be to see if the model can "backcast" the climate over the last few million years to produce the ice ages at the timing and intensities that the geological record indicates. But, that is still not detailed enough to deal with the rapid changes we see today.
Modelers have a saying: "All models are wrong - some models are useful." We need to remember that the models can be very wrong when first developed. I am concerned that the idea that the models must be "smarter" than their developers because they used "artificial intelligence" will be used as a political argument to the public when some climate scientists start having misgivings about some part(s) of the results. Modelers typically already have an overconfidence in their model's accuracy.
Facebook
Twitter
Instagram