Tesla’s “Self-Driving” System Never Should Have Been Allowed on the Road | Hacker News Reader

Tesla’s “Self-Driving” System Never Should Have Been Allowed on the Road

prospect.org

43 pointsyoelo16 comments

I’ve noticed when Tesla stock goes on a run, a wave of such articles and posts begin appearing. This particular posters history only posts about Tesla in a negative light.

Im so fed up with this click garnering nonsense. I've had two teslas, and driven probably 200k miles on them since 2018, the majority of freeway miles on autopilot. How many times has autopilot created a dangerous situation? Zero. Zero times. Do I play with my phone and fall asleep while using it? No of course not. But it removes the cognitive load of making micro adjustments to the course of the vehicle. Has it done odd stuff? Yeah for a while it would brake for bridge shadows. Did this cause a ten car pile-up? No of course not, it was rectified by a tap on the accelerator. For me the issue is clear. Bad drivers are bad drivers, given any tool, they will remain idiots.

This is not about Autopilot, it’s about the “Full Self-Driving” system.

1. As mentioned, this article is about Full Self Driving Beta Pro Black Edition, but it's important to note that it is commonly confused with Autopilot, and vice versa. Even the most die-hard Tesla fans routinely confuse the two (example: you saw "Self-Driving System" and somehow that registered as "Autopilot").

2. It's only click garnering nonsense if you don't bother reading the article.

3. I respect your thoughts and experiences with Autopilot, but your experience will never invalidate other people's experiences, nor cold hard data.

4. If bad drivers will always be bad drivers given any tool (which I'm inclined to agree), doesn't that invalidate Elon's claim that FSD will actually allow you to be an absent driver, one that doesn't look out the windows, plays games and drinks alcohol?

> 1. As mentioned, this article is about Full Self Driving Beta Pro Black Edition

The article does mention both. Notably it uses autopilot+FSD crash count (but then only divides by the miles driven using FSD).

Taking their calculations at face value, they estimate the fatality rate of Tesla's system as 11.3 deaths per 100 million miles.

That is more than a standard car, but it isn't the most dangerous thing we allow on the road. For example, the fatality rate of riding a motorcycle is 31.64 deaths per 100 million miles.

source: https://www.californiaaccidentattorneysblog.com/california-m...

The per-mile risk of a self-driving Tesla is roughly the same (within the range of estimates this source gives) as the per-mile risk of riding a bicycle.

source: https://bicycleuniverse.com/bicycle-safety-almanac/

It doesn't really make sense to ban this new technology which is basically in the middle of the pack in terms of risk. Personally, I don't have a Tesla, and I would be reluctant to use their self-driving technology. But I don't ride motorcycles either. I don't think we should ban either of those things; consenting adults should be allowed to take the one-in-a-million risk that is driving ten miles in a self-driving Tesla, just like they are allowed to ride motorcycles.

Motorcyclists and bicyclists take their own risks. Tesla drivers risk killing others as well as themselves.

> The per-mile risk of a self-driving Tesla is roughly the same (within the range of estimates this source gives) as the per-mile risk of riding a bicycle.

That may be true, but the reader is reminded that a typical car trip is probably five to ten times as far as a typical bike trip (is that a reasonable guess?), so the risk per trip is five to ten times as high. The safety figures often quoted for plane travel share similar characteristics.

Not to mention the disparities in cycling infrastructure. Riding in typical North American suburbia, on stroads with maybe some paint to protect you from fast driving SUVs is completely different from riding in a physically separated cycle lane in a speed controled city.

Also... if you're in a car that's more dangerous than usual, you have a higher risk of injuring yourself and a higher risk of injuring/killing others. If you take a bike, you're probably not going to slaughter a pedestrian.

There are three major problems with this argument.

First, Tesla's system is primarily engaged on easy roads, and will automatically disengage in tricky situations. This means their safety figures are heavily skewed by the implicit exclusion of many accident-prone locations.

Second, comparing per-mile numbers across wildly different modes of transportation is impossible. Cars often drive literally an order of magnitude faster than a bicycle in a low-risk environment: it's called a "highway". Meanwhile, bicycles spend all their time in high-risk environments.

Third, it completely ignores the cause. A "bicycle fatality" is almost always actually a car killing the cyclist. Bicycles don't do the killing. Meanwhile, an "FSD fatality" is usually someone who actively got killed by a car with FSD engaged.

Once your are comparing apples to apples, Tesla suddenly looks a lot worse. And even if it were as safe as they claim it is, they should not automatically be allowed on the road. A lot of accidents are caused by poor or distracted driving. That's already illegal, and anyone caught doing so would be prosecuted for it. If Tesla acted exactly the same on the road, it should be prosecuted too. Right now they are shifting all the responsibility onto the driver who is supposed to "supervise", but considering how their product is marketed I do not believe they should get away with that loophole.

"The American Prospect is devoted to promoting informed discussion on public policy from a progressive perspective.

Founded by Robert Kuttner, Paul Starr, and Robert Reich"

Here is Robert Reich claiming that corporate profiteering is the "structural driver" of inflation:

https://twitter.com/RBReich/status/1595880687344689152

"If inflation comes up at the dinner table tonight, don’t forget to draw attention to the deeper structural driver of inflation: the concentration of the American economy in the hands of a few corporate giants with the power to raise prices, even as they rake in record profits."

And not the five fold increase in the money supply between 2019 and 2021:

https://fred.stlouisfed.org/series/M1SL

This, by the way, is exactly what Latin American politicians claim:

https://twitter.com/nanexcool/status/1463679262381199365?s=2...

Money Supply conspiricists have predicted 8000 of the last 1 high inflation eras.

In the meantime: https://www.reuters.com/business/energy/exxon-smashes-wester...

The idea that supply and demand determine prices is not a conspiracy theory.

Expanding the money supply five fold in two years will create inflation.

Discussion of the multifaceted issues causing inflation across the globe (even in countries with a less severe increase of their money supply across the same timeline) aside, does this in any way negate their position on Tesla operating their not nearly sufficiently tested "Full Self-driving Beta" on public roads using untrained members of the public?

I feel that the article, if anything, could have gone even more in-depth, as one key criticism I see rarely mentioned is Teslas constant changes to the FSD stack, the implications of which a lot of people "using" the "beta" simply will not be able to fully understand and has on multiple recorded occassions caused behavior that were different from what the untrained drivers had come to expect using previous versions.

This is a unsupported analysis. Despite the California DMV mandatory reporting requirements for autonomous driving systems (ADS) with a safety driver [1], FSD under open beta testing with a safety driver has been deliberately underclassified by Tesla to not qualify as a autonomous driving system to suppress disclosure of their raw usage data [2]. As the only entity with the raw information is the manufacturer which has a inherent conflict of interest, no robust untainted safety analysis is possible.

The National Highway Transportation Safety Administration (NHTSA) does have a standing general order (SGO) [3] requiring the disclosure of all ADAS incidents [4]. However, Tesla redacts which system and version was in use in all reports citing proprietary information unlike other manufacturers which only redact in a small fraction of their reports. In addition, a cursory examination (filter to Report Type 10-Day Update) shows 749 Tesla incidents with 711 having unknown injury severity. This is distinct from "No Injuries Reported" where they know that no injury occurred; they only know for certain that nobody died in less than 5% of the cases. This is again in contrast to all other manufacturers combined which only have a total of 203 incidents of which only 63 have unknown severity. That is 5% known for Tesla versus 70% known for all other manufacturers, a 14x difference. The deliberate censoring of the incident information by Tesla makes it even more impossible to do a meaningful analysis.

So, not only is the denominator suppressed, even the numerator is censored. It is straight up impossible for any third party to do any analysis. As the default assumption with safety-critical systems is that they are unsafe, and it is literally impossible for any untainted safety analysis to be done, it is affirmatively unsafe until at a minimum they stop suppressing the incident and usage information that they are legally required to disclose.

[1] https://www.dmv.ca.gov/portal/vehicle-industry-services/auto...

[2] https://thelastdriverlicenseholder.com/2023/02/17/2022-disen...

[3] https://www.nhtsa.gov/laws-regulations/standing-general-orde...

[4] https://static.nhtsa.gov/odi/ffdd/sgo-2021-01/SGO-2021-01_In...

Reply on news.ycombinator.com