AI is an overused term. At its core, it’s just advanced analytics and machine learning algorithms. We’re just able to do more now because of the advancement of technology that’s allowed us to, for example, feed streaming data into neural networks and the like. I think it’s all pretty interesting from a math and stats perspective, but it’s still garbage in/garbage out, in the same way that bad data will lead to a messed up best fit line on a basic 2D graph. The biggest issue I see is the black box nature of the more advanced architectures. You need a PhD to understand why some of them work, though there is software out now that is focused on interpreting how algorithms come to the conclusions they do. I haven’t used it, but I know it exists. The other issue is that, at least in the weather forecasting AI world, I’m pretty sure the predictions are made based on how well the ML components can identify patterns instead of relying solely on primitive equations. I’m not sure how you study that from a physics perspective, per se.