It'll be a big clusterfuck
Your statements nicely illustrate the problem with 'AI patents': So your novel, patentable method would be to use Tensorflow to recognize cat pictures? (Feel free to substitute the cats with whatever you consider useful.) Or do you really, in your heart, want to patent the idea of using AI for recognizing said cat pictures? Or do you want to only patent the use of your particular training data in combination with Tensorflow? Or maybe the use of any equivalent AI system that could give somewhat similar input-to-output results?
Whatever you try to do, either you get a monopoly for an idea as opposed to a method -- and that would break the very innovation the patent system is supposed to foster -- or you try to patent data or software.
AI is so interesting to so many, because it's a black box that promises to solve all problems of this world. Because it's a black box, it's really hard to prove that it cannot, potentially, solve a particular problem. You could spend decades of your life trying to understand and solve a hard problem. Or you could just use this magic box.
I predict that AI will go out of fashion, gradually, over the next 10 years. All those great AI projects will be extremely successful. (You wouldn't tell your boss otherwise, would you? After all, it's research and it's OK to move the goalposts a bit, innit.) But after a few rounds of funding, the weeding-out will begin.