/biz/ - Business & Finance

>>22639493
Everything in the muh ethics, muh safe ai, and muh adversarial example themes are complete bullshittery. It applies identically not only to any kind of software in general, but even to humans. Moreover, the methodology is always complete and utter garbage, and the conclusions never actually hold (they're mostly made up from grossly incomplete analysis).
The real danger with self-driving is actual, real hacking. In the case of interconnected cars, it would be being able to broadcast fake car locations or fuck with a gps used by the self-driving car to force it into an arbitrary location, or somehow tricking the sensors through similar actual hacking (e.g. ota pirate firmware update) to make the car think it has to veer immediately, or disabling brakes or manual controls in general, etc.