Is AMD going to dedicate entire server farms to make use of this? Direct ML is machine learning and any machine learned upscaling like DLSS is going to require dedicated server farms to run the AI learning network on, open or not.
Also the article you linked shows just how little expertise AMD has in this area, because Microsoft used Nvidia hardware to demonstrate DirectML. Which means Nvidia is ready out the gate DirectML ready while AMD is just "compliant".
BTW what does this have to do with Freesync/Gsync?
I think amd can run raytracing and inference but without dedicated hardware, don't think they will be too efficient, imho.