Driving is a complex activity. If you stop and consider how many activities your brain is engaged in during a simple trip to the store, from judging your speed relative to that of other vehicles, staying in your lane, watching the bicycle on right while keeping track of a truck turning from a left turn lane in front, to watching for an exit ramp and other merging vehicles, you realize there is a lot going on in that 'simple' trip.
Drivers who are focused and aware while driving can manage those demands well, even under extreme conditions, as NASCAR or Formula 1 races demonstrate. But when drivers become lazy, bored, drunk or distracted, even obvious situations, such as when brake lights on all the vehicles ahead of your car flash on, can be ignored.
The autonomous car, such as those being worked on by Google, seems like the solution. A never distracted, never drunk, always alert computer that can act instantaneously would solve all of those problems.
And bring a completely new set of issues. Weather, poorly marked road surfaces, accidents, police directing traffic, all are seen as challenging to a rule-governed computer.
Some have suggested the solution is "a human in the loop." Meaning a human driver would supervise the computer driven vehicle and step in when needed. The difficulty with that is it is likely that human driving skills would degrade to the point that when needed in an emergency situation, they would be incapable of reacting properly and a crash would still result.
One car manufacturer is trying to cope with the problem from the other direction, which leaves the primary duty of driving with the human, but provides oversight from the computer to deal with emergency situations where the near-instant response of a computer would be most valuable.
This may be the direction all the carmakers will adopt absent some massive breakthrough with the trustworthiness of autonomous vehicles.
Source: slate.com, "The Big Problem With Self-Driving Cars," Will Oremus, September 8, 2015