It depends on what the models are about. I’m old enough to remember when the government was experimenting with seeding clouds with iodine crystals to try to precipitate rain.They quit. The reason they quit was that someone ran a weather prediction computer model - one which he knew to be too crude to actually be useful - and printed out the results over a certain duration. He thought better of terminating the computer run as soon as he had, so he took the printed results from partway thru the run, and fed them back into the computer to repeat the end of the run and extend out further.
So the computer put out some result which he expected to match his previous run - not precisely because the printed results he had were not printed out to the full precision of the internal values, but close. Or so he thought. What happened instead was that the first results from the new run were similar to the first run, but very quickly the results diverged utterly. In short order there was no similarity at all. He was shocked at first, but when he analyzed the equations from the new perspective, he was able to recognize that he had a “small difference of large numbers” problem embedded in them.
Take the example of profit and loss. Profit is revenue minus expenses. Unless you have an extremely profitable (or unprofitable) business, revenue is fairly close - within, say, 10% - of expenses. Fine. But suppose you try to estimate next year’s revenue and expenses to predict next year’s profit. And suppose your estimate of each is only good to within 5%. We’re talking about the future . . .
Under those conditions, what is the % tolerance on your estimate of profit? There isn’t any bound on the possible percentage error! Because the difference between revenue and expenses might be very close zero - and if it is, then what ever your estimated profit was is all error. And that error, compared to the actual result of near zero, is off of the mark by an unlimited percentage.
So the researcher wrote up his finding, and predicted that the weather could never be accurately forecast for long periods of time, because the input data would never be good enough, and extensive enough, to support that possibility. He had a hard time, back in the early 1950s, getting his result published, and then in a relatively obscure foreign journal. That “little gem of a paper” is now looked on as the beginning of Chaos theory.
They stopped experimenting with controlling the weather when they realized that they could never predict the weather - and if they couldn’t do that, they could never know the effect of their intervention.
I read about chaos theory in Scientific American many years ago. They showed how chaos generators were everywhere, using the timing of a dripping faucet to generate chaotic patterns for example. The Computer Recreations column also had some fun little projects on chaos theory. Interesting and fun.