I wouldn't get your hopes up on that. Various forms of what could be considered as "AI" have been used in this space for decades - machine learning, advanced algorithms, deterministic vs probabilistic modeling, etc. Quantum computing would help here, but the lack of sufficient hyper-local sensor data across historical time will be a limiting factor no matter how good your trained models are or how much computational horsepower you have to throw at it.The one area I'm looking forward to AI in is weather forecasting. I understand why it's impossible to accurately model more than about 48 hours out, but hopefully AI helps with that some. That way, instead of saying we're going to get a foot of snow and us being pissed off that we got an inch, we would have hopefully been closer to the real amount and never gotten our hopes up. Probably a long ways off though considering all the dynamic variables.
AI output can only as be good as the information is that goes into it, both initially and on an ongoing basis.