This is waymo, not cruise, but it's a comparison between 1 million miles with no humans behind the wheel and the human average. That's about 80 years worth of human driving, and while that's not long enough to provide meaningful data on reducing fatalities, they do show that self driving cars reduce both the frequency and severity of more minor accidents
It turns out that humans monitoring an AI are basically useless at stoping the AI from messing up. We don't have the reaction time to take over and be able to drive the car instantly when the car has been driving itself. That's why there aren't any companies making level three autonomous vehicles, they all skip from level 2 (like Teslas) to level 5ish (like waymo and GM cruise). It either needs to be all human, or all AI, we can't do an in between
The human override is done by an employee in a control center, not by the passengers. They can and do override without using the steering wheel, and that's never going to change. The steering wheels are already obsolete, because passengers aren't even allowed to sit up front or touch the wheel.
This is as level 5 as we're ever going to get.
The cars they're talking about here are fully autonomous, not driver assisted. It's a completely different stack from their commercial vehicles that are driver assisted, and they're not for sale yet.
That dream is a reality right now in at least a couple of cities. At this point its a matter of expansion and scaling, because right now it's geofenced into a couple of small areas
Their track record is literally better than the track record of the average human driver. The data doesn't back up what you're saying at all
Self driving is a lot more than just AI though. Our current AI gets really smart really quickly, but fails way too often to be used in critical roles like driving, which is why most of the code in these self driving cars isn't actually AI. They use it to predict the path of other cars, decipher data, and make high level decisions, but the actual control of the car (steering, brakes, etc) is all traditional programming. Waymo even talked about how they used to have more AI, and removed some in favor of traditional logic
We trust our lives to wires all the time, brakes, acceleration, airplane controls, elevators. This is no different, you just put in enough redundancy to make failures safe
Oof, that's relatable. I can't explain how many dating app conversations go stale like that.
<1hr after I get a text: pretend I didn't see the notification because I can't deal with it right now
1-5 hrs after: read it, and stress constantly about what to say
2 days after: realize I didn't respond, get stressed because it's now long enough to be awkward, and I have to change what I say
1 week after: realize it's really too late now, the conversation sits for eternity
It's a really cool technology, but the main problem is that letting people around the world inspect and verify just isn't needed in most use cases. It does a great job at removing the central source of truth, but rarely does anyone explain what the problem with a central source of truth was. Especially when you're talking about a company setting, startups don't want to build open source software without a source of truth, they want to be the source of truth
They already operate on the road without a driver behind the wheel at all, haven't they already "replace[d] a regular human driven car in its regular setting"?