You also have the issue of performance. If you have a NN churning predicting every frame what your remote player would do, it would add even more load to that 1

You also have the issue of performance. If you have a NN churning predicting every frame what your remote player would do, it would add even more load to that 16ms loop of calculating and rendering everything else. At best maybe some very basic ML might help, but the "assume no input changed" seems to be best guess from the sounds of empirical testing in the article.