And we make the same anti-ai argument again that's made against self-driving cars and what not. Humans are widely different. The "correct" form for an exercise depends on the ratios of your various limbs, the flexibility of various muscles and joints, the current strength of various muscles, past injuries, and many other factors. Most of these factors are not visible on video, and can't even be captured by a medical/exercise expert in quantitative ways.
Getting to the "correct" form is an iterative process over many sessions, and it is never complete. You often don't even discover what is bad form for you till months into the process when you get injured and realize in retrospect.
Packaging it up into a AI that only looks at video and gives feedback, and people trusting it, is a sure shot way of getting people seriously injured.