Originally Posted by InOmaha

I'm a little curious as to how they approach legal driving vs human safety. I know that sometimes the safest route for me to take is the illegal route. I've had several occasions that required me to speed up in order to clear an unsafe condition and avoid a collision versus slowing down and getting into an inevitable, though slower, accident.


Training the software specifically for legal driving boundary conditions may unnecessarily limit the full AV capability. Though I could see from a legal aspect that following the letter of the law would limit corporate liability and therefore it would be better to risk a collision even if it costs lives in an effort to shield them from blame.


The AV transition period will be difficult because regular cars will not be able to communicate intention with AV cars (braking, turning, car stopped on a blind corner, etc.). This could theoretically result in situations where the AV is responding to a sudden change or an accident in front of them that requires the software do make a judgement decision. Is it better to run into the stopped car in front at 60 mph or is it better to run the cyclist on the side of the road in and effort to avoid the collision.


I'm almost certain someone somewhere has had this discussion and it may already exist in the software as a hierarchy of logic. The feedback loop for humans is much different then the feedback look for this programming; which may be yearly accident statistics or which lawsuit costs the company less.