Musk has made this argument many times. Basically don't let perfect be the enemy of better.
But if the vendor is legally liable in each case then this software is untenable.
If you're actually trying to increase safety, the right things to build are computers to supervise the human driver, and not systems that require a human to supervise the computer driver.
If the human drifts out of the lane, nudge the car back in and beep, but don't prevent the human from intentionally going over the lines.
If the human appears to be driving into an obstacle, beep and apply the brakes, but allow the human to override.
If the human isn't paying attention, beep, then turn on the hazard lights and slow down / try to find a safe place to pull over.
Tesla also doesn't say that it's safe to let the car drive. There are ample warnings and messages and feedback that say that you have to keep your eyes on the road and hands on the wheel at all times.
There are likely more non-AP Tesla/non-Tesla auto accidents and fatalities than there have been AP related ones. I feel safer riding in a Tesla with AP engaged and the driver having his eyes on the road and hands on the wheel.
Boeing promised a brand new and improved jet that was cheaper because it packed in a bunch of improvements without having to go through certification on a new airframe or pilot retraining. Engineering tried their best but either through negligence or "just following orders" and bureaucracy they failed catastrophically, as any engineer will tell you would happen with a non-redundant (in a 3+ vote configuration) sensor that could override pilot control inputs based on faulty readings
Tesla had to be "encouraged" to say that "at all times" was in fact an amount less than "every 15 minutes", which it was just a couple of years ago, touch the steering wheel four times an hour.
A few percent better than "humans" is quite bad relative to a semi-responsible human driver.
Yes it should be (imho) at least an order of magnitude better than humans before we consider it OK to not pay attention.
Not a few percentage points better as you say.
However, to overcome media drama, it probably will need to be not just one, but more like three or four orders of magnitude safer than humans, because of the “oh look, a Tesla had an accident” effect.
If you assume vans have about the same crash protection as cars and some van drivers are idiots then I'd guess you could get 10x better than the average human just through driving skill/technique.
Per passenger mile. So if busses have 30 people on on average, the accident rate could be the same, just typically only one passenger dies.
That also makes sense with the vehicle characteristics. It's much harder to die by being ejected from the vehicle or by intrusions into the passenger compartment in a bus. Also, heavier, longer bus = more gentle deceleration.
Your statement might still be true (and I suspect it broadly is), but it just isn't backed up by the cited data.
Or maybe I'm overthinking it and you're just talking about how hard it is to get the statistics (number of accidents) down overall?
Since drivers under the influence are less likely to die in an accident because most drugs either relax the muscles or delay reflexes enough to keep the body relaxed during an accident, preventing a lot of potential injury, it's an important nuance to consider. If the people most likely to drive drunk are the least likely to adopt self driving cars, there is an absolute floor to how much safer they can be.
The easiest way to deal with the issue is to be safer than, say, 99% of drivers. I don't think Tesla can make that claim at present.
If it's better at avoiding harm, across the board, in all situations where humans currently operate vehicles, then yes, that would be an improvement. But Tesla is nowhere near being able to do that.
Tesla seems to be trying to claim that it's sufficient for them to be a few percent better at, for example, keeping the car in its lane, or maintaining speed, or spotting obstacles. That's not an improvement. Driving is not a contest to see who can identify the most obstacles or maintain speed or lane alignment the best. People want to get where they're going safely.
I don't think this can be true for a 90 or 95% solution. I think too many people would think the system can drive for them, because they're not Tesla enthusiasts and they're not reading all the caveats about what the aids can and can't do, and they'll gradually start to rely on autopilot until the day it lets them down, possibly with a very negative outcome.