3 ms·
The problem with code validation in the context of a self driving care is that the space of inputs to the software is essentially the entire world. The code is
by zaphar 11y ago
The problem with code validation in the context of a self driving care is that the space of inputs to the software is essentially the entire world. The code is necessarily complex and the invariants nearly impossible to adequately specify. A general statement like "Does not collide with deer in road" is in fact insufficient.
This isn't like an airplane autopilot or your typical other sorts of highly vetted software. It's many orders of magnitude more complex.
- TeMPOraL 11y agoAnd that's why I think questions like "whether the car will be programmed to kill you or someone else" are missing the point. The car won't have an explicit decision point like that! Evaluating pros and cons is what humans do because it's hard for us to do any better. A car can be continuously observing, updating and navigating the solution space, and if it ends up killing someone, it's either because of not having sufficient data, or not being able to stay within its envelope on the solution space (i.e. a situation when it's physically impossible for it to avoid a collision). I don't see why a car would ever have an explicit if(something()) { killDriver(); } else { killTarget(); }. "Killing" won't even be a concept in code.