I’ve seen some things suggesting they’re testing it, I’m not seeing that they’ve deployed anything.

Honestly, I’d expect AI to make the same mistake as their current systems, though, for a similar reason

You’re talking about a few different things, here.

• AI could potentially use that info, yes. Teslas don’t use AI to drive
• Self driving cars do have the potential to be significantly safer, this is true. Telsa’s cars, through a series of missteps, are not anywhere near that potential, and in fact are hazards on the road. This is bad for developing true self-driving cars!
• The only obstacle to progress being discussed is covered by a picture of a road…

Human drivers can pull in environmental and other cues though, like “why is there daylight on the road inside a building?”