How do Driverless Vehicles Make Decisions?

Graeme Pattison sent in a link from Autonomous Vehicle International. This short video gives a little insight in AV decision making.  

https://www.autonomousvehicleinternational.com/videos/how-do-driverless-vehicles-make-decisions.html

Here’s an extract of the commentary from the video:

There's no program that if you see a small girl to the left and an elderly man to the right, [decides to] kill that elderly man because of utilitarian ethics. 

Instead what we're seeing is deep learning neural network. 

Deep learning is the one that is the most powerful and talked about right now. It's what's used by Google and Facebook for a lot of their ads that kind of thing. But what makes it somewhat unique from other types of classifiers is it's not transparent. So we can't actually go in and see why a vehicle did what it did. Let's take that Uber case in Phoenix recently. So Uber did a full investigation of that death and they were able to learn a lot but it could not actually go in and see why did the car make the exact decision it made when. 

So Ford's autonomous vehicle and Google's autonomous vehicle could be in the same exact situation and do something slightly different. 

One issue that has arisen in discussions about AV decision making is whether each vehicle can continue to collect data and use machine learning to improve its performance.

The short answer is no. The system capability is built into the software version when it is loaded and it would be too risky to add data from a single vehicle without background analysis.

Alan Finlay replied 
Note also that the drunk driver might still have some responsibility to intervene if the AV has ‘issues’. 
See also the article in SMH Drive supplement where Nissan does not believe it will be possible for zero human involvement.