Exactly. Self-driving is a far more complex problem. The car has to get it right, right here on the road, little to no margin for error.

The script-writing AI, voice AI, actor-image AI will be used to create a multitude of options, and all but one will be thrown away anyways. Perfection isn't necessary, realtime isn't necessary, and a human correcting things here and there is always still an option. Compute resources are also far less limited, it doesn't have to fit in a "car" form factor and power consumption.