I don't disagree that humans use quite a bit more than vision only while operating a vehicle. There's some intuition and experience that contributes to the decision making of a human driver, but let's not pretend that we as humans drive with perfect information and understanding of what other drivers around us are doing, or that that information can't be synthesized from other data.
Just this morning, the car reacted beautifully to a driver who quickly changed lanes in front of me. I saw what was happening, and before I could react, the car was already taking appropriate action.
The current FSD version no longer relies on 300K+ lines of C++ code and is based on AI that has "learned to drive" by observing countless hours of safe human drivers. A lot of the intuition that we gained through years of individual driving experience has now been duplicated in the training of this AI model. Not only that, but the AI has orders of magnitude more driving "experience" to draw from than the average human. It also has constant 360 degree visual input to feed this model. A lot of things that we intuit based on experience might be able be replaced by actual visual data and better-than-human response times.
I'm hedging a bit, because vision-only with AI could end up being a dead end, but my initial experience is very positive.