That last item is going to be tricky. For example, should a self-driving car be programmed to avoid a pedestrian at all costs -- even if it hits another vehicle?
Shashua does not envision a vague "do-no-harm" directive. He is proposing a concrete set of algorithms that regulators and insurers would endorse, and that automakers would adopt.
If those algorithms are adopted -- and if a vehicle malfunction did not cause the accident -- the automaker would not be held liable.
Shashua outlined his proposal for an industrywide "do-no-harm" protocol in an academic paper published Tuesday.
He will get a respectful hearing. Shashua's company commands a 70 percent share of the world market for obstacle detection software. After Intel Corp. finalized the $15 billion purchase of Mobileye this year, it put Shashua in charge of its automated vehicle group.
It seems safe to assume this proposal will trigger a lively debate. But Shashua's blog noted that automakers will have something to offer in return for legal limits on their liability.
A self-driving car festooned with cameras, radar and lidar will be much safer than conventional vehicles. For example, such a vehicle can calculate precisely how closely it can follow a vehicle without risking a rear-end collision.
If the entire U.S. vehicle fleet were autonomous, Shashua estimates that it would suffer about 40 fatal traffic accidents annually. By comparison, there were nearly 40,000 fatalities last year on U.S. roads.
"Society can accept human error, but what happens on day one when you fully take the driver out of the driver's seat?" Galves asked. "To create an environment for self-driving vehicles, we need to quickly and definitively assign blame for accidents."