You're right, that is shockingly low! 43,000 people die in car crashes in the US every year, and only 17 in Teslas since 2019. That's only 1 in 7500.

These statistics are meaningless.

How many Tesla cars with full self driving are there compared to regular cars?

A Tesla with FSD will only be using FSD a small fraction of the time.

If we had statistics on the number of hours FSD was active compared to the number of hours all other car driving of all cars we might be able to compared these numbers.

Hour wise, I think normal driving is way above 7500:1.

Why are people willing to excuse things because they happen in a car? If an AI power tool were going haywire once in a while and chopping the arms off of bystanders near by, would we find that acceptable because people using non-AI powered tools also chop off their body parts sometimes? 'We can't know if it would have happened if a person were in control' is not an argument that would fly for anything else just as dangerous.

> would we find that acceptable because people using non-AI powered tools also chop off their body parts sometimes?

Yes, if the AI power tools chop off less limbs than the human controlled power tools, I would find that acceptable.

For that to be meaningful, one would have to consider what proportion of miles driven are in teslas, and where.

Just because you crash that doesn't mean you aren't able to navigate without human intervention. Humans are also capable of crashing their car.

He means under their supervisor control. There are "normal" people who do their best to cause a crash.

Without any comparison to humans those numbers are completely meaningless

> Without any comparison to humans those numbers are completely meaningless

I disagree.

What I want to know, is is it safe? Not is it safer than the average human driver, but is it safe in an absolute sense.

When a human driver crashes we can almost always pin the cause down to human error. Errors caused by some human failing, being less than they could be. The promise of machines is that they are consistent. They are never "less than they could be", but are consistently at their best.

Comparing humans and machines is comparing dogs and roses - it is interesting a rose smells better, interesting a dog is more loving (or fierce), but not a valid comparison

Self driving cars stand or fall on their own capabilities.

Sure it's different whether you were hit by a malfunctioning machine that was confused by sunlight or whether you were hit by a driver who wasn't paying attention because they texted. In one instance you have the person genuinely apologize to you while in the autonomous car case, maybe there wasn't a passenger at all.

But on the other hand, if you can identify the most accident prone group of humans, and require them to use AI cars, and with this you could significantly reduce the number of road kills/accidents, wouldn't that be an improvement?

1. It's the adaptive cruise control, not the FSD.

2. What would a good absolute number for that period be? Our threshold can't be zero or we'd have to give up every technology from showers/baths (~60x more deadly) to fresh produce (E. coli, salmonella, listeria).

> 2. What would a good absolute number for that period be? Our threshold can't be zero or we'd have to give up every technology from showers/baths (~60x more deadly) to fresh produce (E. coli, salmonella, listeria).

Zero accidents caused by the failure of the software. (Clearly a good system may have some obscure bug that makes it fail - but failure of a self driving system should cause uproar and consternation - unlike the reaction of Tesla to the deaths caused by their systems, it seems)

If I slip over in the shower it is not because the shower head went rouge and strangled me. If a shower was designed that in the period of one year strangled eleven people (a fair comparison to Tesla's record) imagine that?

Point being your comparisons (food poisoning or household mishaps) are not relevant

- It's not clear why you no longer care that a technology is killing a lot of people if there's a human in the loop that you can blame. If a driver runs over your sister/son/friend, does it matter if you can blame someone?

- About 43,000 American motorists died in 2021. If we have a way to prevent many of their deaths with software, would you not want to unless you could prevent every death?

- Why is a software tool failure different from any other tool failure? Brakes can fail, wheels can fall off) (happened to me once), etc. Listeria in produce is a failure in the production chain. Showers can be made safer. You use your car/shower/spinach and some day, for reasons entirely beyond your control, it might kill you.

- Why isn't the driver to blame in a Tesla? They're supposed to be watching and responsible.

- There's no clear distinction between what's a software failure and what isn't. Collisions have multiple contributing factors. Perfect software will still have collisions. The numbers reported aren't just collisions where the Tesla was at fault. One person was killed by a self-driving vehicle when a person jumped a concrete barrier and ran across the highway at night. The human driver couldn't react in time, and the software didn't see them. Is that a software failure? At what point do we accept that a collision is no longer the car's fault?

Autopilot is just fancy cruise control. How many crashes have there been with cruise control turned on? It's just a completely meaningless thing to even look at. How many crashes would there have been during those miles without autopilot?

Tesla doesn’t sell Autopilot as cruise control though.

Yes they do? Autopilot is standard in all Teslas, which is different from the Tesla FSD (Beta) that Mercedes is being compared to here.

Because even if they did, would still be worst than many others.

"Ford’s BlueCruise Ousts GM’s Super Cruise as CR’s Top-Rated Active Driving Assistance System - Tesla Autopilot falls to midpack as other high-tech systems improve" (Jan 2023) - https://www.consumerreports.org/cars/car-safety/active-drivi...

"...Of the 12 ADA systems we just finished testing, Ford BlueCruise came out on top, followed by Cadillac Super Cruise and Mercedes-Benz Driver Assistance. Tesla, once an innovator in ADA with its Autopilot system, fell from its second-place showing in 2020 to seventh this time around—about the middle of the pack. That’s because Tesla hasn’t changed Autopilot’s basic functionality much since it first came out, instead just adding more features to it, says Fisher...

...“After all this time, Autopilot still doesn’t allow collaborative steering and doesn’t have an effective driver monitoring system. While other automakers have evolved their ACC and LCA systems, Tesla has simply fallen behind.”..."

That's the number of crashes where Autopilot was enabled, not the number where it was at fault. So its a meaningless number.