On Thu, 22 Dec 2016 02:42:19 -0800 (PST), wrote:

On Tuesday, December 20, 2016 at 1:18:20 PM UTC-5, Quadibloc wrote:
On Tuesday, December 20, 2016 at 9:57:39 AM UTC-7, Chris L Peterson wrote:

That is not the way these systems work. In reality, the car will
almost certainly be making decisions based on deep learning methods,
meaning that there will be no easily identifiable criteria, and
certainly not any provided by programmers.

Deep learning techniques would indeed be used for the things we don't know how
to "program" - how to *recognize* a human being, how to control a moving
vehicle, and so on and so forth.

But it would still have to be possible to control such a vehicle - to impart to
it a request to go to a given destination. One still needs to tell it *what it
is supposed to do*. Crashing into a wall instead of squashing the jelly-like
objects in front of it... would be done in service of those fundamental goals
which would have to be set for the vehicle explicitly, even though the skills
it uses *to* do these things would be achieved by deep learning rather than
programming in large part.

John Savard

peterson doesn't get that a machine can -only- do what some human has "told" it to do, directly or indirectly.

It demonstrates that you are a fool when you speak about things you
are wholly uneducated about.