Auto-pilot risk discussion maybe OT, and much broader than Tesla of course - but I find very interesting.
I share Amped and Hank's concern that it will take time to perfect auto-pilot, and that humans can/will botch it up. Glad that Tesla has "outsourced" most of this to Mobileye, which has deep experience across many marques... so much of the function will come from dedicated specialist.
Ecarfan and Skilly, true that many of the MS (and most lux cars these days) are laden with "drive by wire"... steering, accelerator, brake, ABS, traction control.... but I believe that this is a different class of software with respect to complexity. For example, steering software does one thing: steer front wheels lock to lock - and is controlled by one input device: steering wheel. Pretty simple set of use cases, relatively easy to bulletproof. Auto-pilot software, on the other hand, has to take multiple inputs (photographs, sensor input, etc.), interpret them, relate them, devise a response, and relay it to multiple systems (braking, steering, acceleration, driver alerts). Much, much, much more complex. And, the "drive by wire" functions have a failsafe involved on the other end - the driver. When we get to full auto-pilot, drivers are going to check out, so this incredibly complex software is going to have to be pretty perfect.