>I also recognise that if Tesla doesn’t get out in front of every single incident, it may set back the replacement of human drivers with safer AI drivers.

Tesla doesn't need to get out in front of every incident. Musk feels the need to because of the outlandish claims he continues to make about how close Tesla is to having autonomous vehicles. This is something Tesla's board should have reined him in on long ago.

>I don’t really trust any famous person for privacy, because they necessarily don’t have even close to as much of it as a normal person.

I agree that famous people have less privacy, and there's a lot of unfairness around that. But even if Musk hadn't chosen to become a public spectacle, that doesn't give him the right to treat others the same way.

In this case he felt the need due to outlandish and incorrect claims about the nature of the accident.

> But even if Musk hadn't chosen to become a public spectacle, that doesn't give him the right to treat others the same way.

Of course.

What I’m saying is, I don’t expect the famous to be able to empathise with the rest of us, not that it’s good that they can’t.

> Tesla doesn't need to get out in front of every incident.

Tesla has been attacked in the past so maybe Elon takes such issues personally.

I'm not convinced that 'autonomous' driving will ever be better than non-distracted/drunk drivers. Suppose this is a task that, short of overhauling all our road infrastructure, ultimately requires GAI? I think we generally have way too much confidence that cars will ever drive better than people, at least on the current road system. I am not saying that they definitely won't, only that I'm not sure we should take the attitude that anything that gets in the way of autonomous driving is sacrificing human lives. We don't know that.

I also worry that mixing autonomous systems with people will result in more accidents, both in the case of drivers nodding off instead of supervising the car, and in the case of other drivers being surprised by the autonomous car's behavior. On top of this, it would not be surprising if autonomous car accidents were more psychologically devastating - they would likely commit different categories of error than human drivers.

Conversely, if the tech industry wanted to save lives on the road, easily the most impactful thing they could do is shut off any phone screen moving faster than five miles an hour. This would inconvenience passengers, but we did after all survive for many years without phones in the car.

> Suppose this is a task that, short of overhauling all our road infrastructure, ultimately requires GAI?

I would be surprised if level 5 self driving requires an AI smarter than a mouse or an eagle — freeform environment, arbitrary weather, hostile others, camouflage(!), long range navigation.

The rest of your points I broadly agree with though.

How many birds have crashed into your windows? At my parents house, this happens a couple of times a year.

One, over my lifetime.

I don’t think this a great comparison. Firstly, sheets of glass are not a common road obstruction, and even if they were cars have the added advantage of more sensor types (doesn’t require much more brain, it’s equivalent to just another ‘colour’); Secondly, I’ve also been hit by a car driven by a human.

The driver had stopped at a give-way (major-minor junction) [0], then failed to look in my direction and pulled out into me as I was turning into that road.

[0] right where this unrelated white van is in this Google street view photo: https://goo.gl/maps/Xe87aruFbbUazXhp7

> at least on the current road system

I came to the conclusion years ago that we're going to need RFID tags on everything (or something similar) to make FSD work safely enough without AGI. We'll probably also need dedicated lanes in some places. Nothing I have read or seen yet has given me any reason to think differently.

Edit: The common counter-argument I see to this is that tags could be altered maliciously. My response is that we've already been building systems which scan for and filter out malicious changes for a long time, and we keep getting better at it. It seems to me like this kind of system would be far easier to design & update than any FSD software I know of.

RFID is no solution at all, how do you ensure a toddler running into the road is properly RFID chipped?

If you need every possible obstacle to be tagged with a locator beacon, you’ve essentially just reinvented 1970s autonomous subway or PRT technologies that require the track to be completely fenced off from surprise animal or humans or debris ended up on the track.

The idea is not that it would be RFID only, but to reduce the number of non-tagged objects & people so false positives happen less often. Obviously you still need a way to deal with random, unexpected obstacles, but it's a much smaller problem space when the surrounding environment is mostly properly tagged.

There's a very real risk that half-assed "AI" driving just leads to worse human drivers that stop paying enough attention to driving normally. Sensors are definitely better than humans at reacting to collisions though.

Hah, I thought you meant worse drivers outside of the Tesla. I guess if I wanted to be an ass, I can cut off most Teslas because they have the auto-brake feature. (Well most modern cars have it).

You can, but if you cause an accident - you now have 8 cameras recording your every move. I bet I know how that goes down in court.

You know who’s making real progress on AI driving? Waze is. I don’t want to be too dismissive of EV progress thanks to Tesla but I feel like there’s no contest about who is pushing forward AI driving beyond the status quo.

And Waze doesn’t need a professional troll to go around garnering media support 24/7.

* Waymo not Waze

* Operates in a tiny geographically fenced area

* Despite success, has not expanded to any new areas or rolled out any further.

* As a consumer, I have no access to buy Waymo, but I can buy at least the current implementation of Tesla Autopilot.

Waymo believes that the the roads need to be fully 3d mapped, with crosswalks, street lights and signs before their car can autopilot safely on it. More mapping will happen if/when the AI autopilot is ready.

Tesla autopilot is not a comparable product though! It operates nowhere at Waymo capacity.

My understanding is that Tesla is extremely behind on these fronts compared to Waymo, and that gap keeps widening every day. I know who I would bet on

But they aren't trying to solve the same problem.

I believe Waymo works better in a pre scanned area, where there are no construction sites or major changes, but in a general context you can atleast attempt to use Autopilot, which you can't with Waymo. Thus I would not say that Waymo is miles better.

About Waymo: I used to think they were a leader in this space, partly based on their published safety statistics, but I am no longer so sure.

I had the great fortune of living in "Old Palo Alto" (think "mega rich tech CEO paradise") for a year while my wife was at Stanford. This is a grid of low-traffic, low-speed, sidewalked residential streets, ~10 minutes on all sides from the nearest highway, with typical mid-Peninsula weather (in other words, essentially perfect driving conditions).

I would would frequently see Waymo vehicles going all through the neighborhood at 6-9AM on Sunday mornings when I was out on long runs. They would often be essentially the only cars I saw on the road.

If there is a better way to pad your urban safety statistics, I do not know of one.

Old Palo Alto on Google Maps, for the curious/unfamiliar: https://goo.gl/maps/cAqifusdRL647YgWA

Do you mean Waymo?

Yes