Re: I'm all for safety but
If every other car was autonomous - and no pedestrians or cyclists were allowed on the roads. We've probably got the tech to do that.
I really don't think we do. Let's consider some failure modes.
The "Top Gear Kevin Failure": If cars continue to be privately owned, some percentage will be owned by "a bloke named Kevin" who does his own maintenance with no idea what he's doing. In general, if maintenance isn't regulated and enforced (and certainly no US states are currently doing a decent job of this), you'll have a variety of nasty failures happening on the road, and other vehicles will have to account for them. I've seen cars lose parts (bumpers, for example) on the highway. I've seen a wheel come off a Jeep when the axle sheared. A friend was riding in a rear-wheel-drive car which had its driveshaft decouple from the torque converter, fall down, and hit a crack in the paving and burrow under it, bringing the vehicle to a dramatic stop as the rear axle gleefully disassembled. You probably don't want to be in an AV following behind that, because I'm betting its tires don't provide enough force of friction to decelerate to a stop in time.
The "Transponder MTBF Failure": People keep saying that AVs with transponders will let us do away with external traffic flow control (lights, stop signs, etc) and allow close-following convoys to reduce drag. Sure, up until someone's transponder fails.
The "Private Roadway Failure": I live on a private road. So do lots of people around here. Plenty regularly drive on Forest Service roads or powerline access roads. It's no small task for AVs to deal with these; they're often not in the mapping information, are often poorly maintained, typically aren't signed, and are generally gravel (or worse). So AVs will need a way to let humans take control, or we'll have to solve the last-mile (sometimes last-quite-a-few-miles) problem some other way; and if humans rarely drive, they likely will be even worse at it than they are now.
The "It's Not an AV Failure": Sure, forbid pedestrians, bicycles, etc. Animals get on roads. People get on roads even when they're not supposed to; how many times have you seen someone walking on the side of a limited-access highway, or a railway, or the like? Other roadway obstacles include things like rockslides and flooding, so the AVs will have to be able to continuously monitor for non-AV obstacles. And they'll also have to be able to distinguish non-obstacles. Trying to drive an AV though eastern Colorado (a trip I make several times a year) would be miserable if it kept slowing or stopping for tumbleweeds — and Colorado isn't even a prominent tumbleweed area, compared to some others. So AVs tracking one another might reduce the scale of the environmental-awareness problem, but it doesn't change it qualitatively.
The "Software Sucks Failure": Software sucks. Sorry, trufans, but that includes Apple software; you really don't want your AV stumbling into the next goto fail. Nontrivial software is really complicated, and complicated systems are difficult to get right and have revenge effects. And security failures in systems that lives depend on are kind of problematic, too.
Sure, it's easy to argue that if the public roads were all AVs rather than all human-controlled vehicles that the — admittedly terrible — accident rate would be lower, perhaps much lower. But there are still a host of failure cases that require both overlapping safety systems on the one hand, and the option for human takeover on the other.