Cool animation, but not a very sound argument. Human drivers also aren't supposed to swerve at highway speed. You slam the brakes. Not only do you slam the brakes, but you slam the brakes and then trust a computer to prevent those brakes from locking up.
Why aren't we posing these dilemmas to our anti-lock braking systems?
Agreed. I understand the argument, but ultimately the reaction time of the computer being better than the human reaction time would limit this scenario anyway. not to mention self driving cars don't have that much time to swerve in this scenario anyway.
Programmers aren't programming decisions in Convolutional Neural Networks, which is how a lot of companies are tackling self driving right now. A CNN takes inputs (what it sees from the camera), passes it to a black box (the CNN), and then generates an output (gas amount, brake amount, steering angle).
One of the problems with CNN's is that there is no way to actually know how it will react to an unknown scenario. There is also no clear way to debug the hidden layers in a neural net to figure out exactly why it chose to do what it did. In many ways it's very analogous to how a human works. The only difference is that it's not eating, playing on it's phone, changing the radio, or doing anything else while it's supposed to be "driving".
Very interesting, we were discussing this the other day in the office, the top comment on this video is a good solution.
Its a fantastic solution, but I think this thought experiment was focusing on if you are the only self driving car