That's also the fundamental problem with all existing forms of autonomous vehicle AI. Google was the furthest along with Level 4 and 5 autonomy but even they ha

That's also the fundamental problem with all existing forms of autonomous vehicle AI. Google was the furthest along with Level 4 and 5 autonomy but even they have backed off that development track as they couldn't muster the resources (a massive fleet collecting data) necessary to make the next leap in progress. Anything less than Level 4 autonomy will require constant driver attention. While there is technically Level 3 autonomy, the premise of a system which can operate with the operator intervening within some "limited span of time" in an emergency is fundamentally flawed. The situations in which an AI will have to kick the decision making back to the human driver are inherently time sensitive. Any sort of time-sensitive response from the driver would have to occur in such a short window that one could never safely remove their attention from the driving process for any meaningful amount of time. Even a time window as short as 5 seconds is too long to react to a situation at highway speeds (the car would have traveled 440 feet if it's at 60mph). My point is that this is an existential problem in the field of autonomous vehicles to a point that even the most ambitious players are admitting defeat. Most, if not all, the commonly discussed benefits of an autonomous car (all the things you can do with a comfy seat and an internet connection in the time one spent driving, also no worrying about DUIs) are unachievable with Level 0-3 technologies. They might make driving safer if they're implemented as a fail-safe for bad drivers but if they're allowed to take full command for any period of time, they are fundamentally unsafe and undesirable in terms of what the public is demanding from the promise of a truly driverless car.