Hacker News Reader

Comment by pwncake | Hacker News Reader

Every incident is obviously tragic, but statistically with the number of Teslas on the road is self-driving roughly on par with the number of expected deadly accidents with manually driven cars? This is still a nascent technology with a lot of promise and it is naive to expect that there will never be a fatality or issue.

View on HN

Statistically, Teslas are much, much more dangerous than other new cars in the same price bracket.[1] To be fair, I suspect the driver demographics skew younger, maler and more urban than most luxury cars and that has at least as much impact as any actual driving characteristics, but the statistics really don't bear out the people killed by glitches or systematic flaws in Tesla's software pale into comparison compared with those saved by it line of defence.

I find it truly remarkable how on the one hand HN can rage for days over Boeing introducing software with flaws that killed people and on the other hand when it's Tesla's software flaws killing people in a similarly repeatable manner, the general consensus veers towards deaths from poor software implementation not mattering nearly as much as the potential of the technology...

[1]https://medium.com/@MidwesternHedgi/teslas-driver-fatality-r...

>I find it truly remarkable how on the one hand HN can rage for days over Boeing introducing software with flaws that killed people and on the other hand when it's Tesla's software flaws killing people in a similarly repeatable manner, the general consensus veers towards deaths from poor software implementation not mattering nearly as much as the potential of the technology

And you can replace Boeing with banks, car companies, oil companies or any other legacy company. Fraud, worker exploitation, shoddy product...these things are acceptable to some here, as long as the company has come from Silicon Valley.

The Medium article referenced points out the difficulties in getting accurate up to date information out of the NHTSA (National Highway Traffic Safety Administration) database. The survey described in the article include data about the Model 3.

Also, they're shorting Tesla. They say that their position on Tesla is due to their research (as opposed to the outcome of their research being driven by their attitude), and that's fair. Nonetheless, I'd be as skeptical of their claim that Teslas are significantly more dangerous than average as I am of Musk's claim that full autonomy will be available and safe before 2020.

Tesla is a virtuous startup. Boeing is an evil defense contractor. What do you expect.

They may or may not be evil, but Tesla is a defense contractor, too. https://dod.defense.gov/News/Contracts/Contract-View/Article...:

”Tesla Industries Inc, New Castle, Delaware, has been awarded a maximum $17,056,347 firm-fixed-price contract for battery power supplies.”

In the Boeing, the pilots were aware and trying to mitigate the problems, but the system prevented that and ended up crashing the plane.

If the Tesla kept accelerating and steered straight into the truck, after the driver disengaged autopilot because he saw the truck, this would be a fair comparison. We don't know whether that happened, but it seems highly unlikely.

Sure, the two are not exactly the same, as the MAXes appear to have crashed because some pilots weren't trained/knowledgeable enough to understand how to address a software malfunction, whereas the Teslas appear to have crashed because some drivers weren't trained/skilful enough to react quickly enough to a software malfunction.

I'm not convinced that distinction makes a tendency to accelerate into static objects a comparatively acceptable teething problem for the autopilot software though, even less so for people who support Tesla's stated aim of full autonomy asap...

> Statistically, Teslas are much, much more dangerous than other new cars in the same price bracket.[1] To be fair [...]

To be fair, how many of the other new cars in the same price bracket and category class (midsize sedan, large sedan) can be ordered with factory 700+hp and sub-3.3s 0-100km/h times?

I haven't seen numbers but I'd suspect that the vast majority of Tesla fatalities are due to the fact that the cars are stupidly much faster than anything new Tesla drivers will have driven before.

One of the 4 fatalities investigated did happen after the tesla was trying to outrun a police car... Not exactly Tesla's fault!

the model s with the performance characteristics you mention starts around $90k (or $110 with ludicrous mode option). at this level you can start getting into 911s, BMW m4/m5, or the highest trim American muscle cars. these cars have comparable acceleration figures to the performance model s and the sports cars will thrash it in any situation that doesn't involve 0-60 in a straight line.

edit: I reread your post and realized you are talking about midsize sedans and larger. Mercedes amg e63 is a better counterexample. most of the cars I mentioned don't have 700hp either, but you don't need it when you start with a 500-1000 lb weight advantage.

Part of the discrepancy is also that I'm from the Australian car market and our prices for fast cars are insane compared with the U.S. so my baseline is skewed. :/

Also a P3D is much cheaper for similar price (https://forums.tesla.com/forum/forums/p3d-base-price), and the 3.3s 0-100 mentioned was for a P3D+ not a S100D.

oh okay, I shouldn't assume everyone can buy cars at the US price. still if you move down to BMW m3 or amg c63, price and performance are comparable to p3d+ (though the Tesla does have a 0.3 - 0.7 lead in straight line acceleration).

I think your point still has some merit though. it could be that fast and quiet EVs don't provide as much sensory feedback to say "hey guy, you're actually going really fast". it could also be that people who are not usually interested in performance ICE vehicles might go for a performance EV for the novelty.

tbf I tend to agree and nearly put "sportier" as another reason why Tesla might have more driver errors than, say, a 5 series BMW (through more aggressive drivers choosing Tesla, or it being more tempting to manually drive too fast) but wanted to avoid a completely tangential debate on the extent to which Tesla was responsible for that (unlike young men on busy exurban freeway commutes having more accidents than the average driver regardless of car type, which definitely can't be blamed on anyone at Tesla). But the key point is that not only is gleaning statistically significant and properly controlled info from accident data hard, but also a lot of aggregate stats don't paint the rosy picture Tesla's PR figures do

If you read this article, definitely read the top-rated comments too.

Are statistics available on airbag deployments?

It seems the numbers of deaths are low enough to make stats difficult, but 'car crashed badly enough to make the airbag deploy' might have more data, and might also help separate out the effects of autopilot vs crumple zone design.

Statistics don't include situations where Tesla went wrong and the driver manually intervened to avoid an accident. One is just human drivers and the other is a team effort between human drivers and AI, if the results are the same then that's quite condemning of the AI.

At the moment the relevant choice is between human and human+AI, not human vs only AI. And I believe human+AI outperforms in what rough statistics we do have.

We'll have 'full self-driving' in a year, so option 3 is on the table very soon, according to Tesla's marketing.

I assume that this is mostly pr in terms of how autonomous the ai will be once again, but many people may buy it and then get blamed for it when accidents happen.

Of course. But there obviously is a real issue with overhanging obstacles, since there have been multiple of these crashes now.

Do we really accept a known flaw just because it's still better than the alternative?

Honestly? Yes. We take the best option available at current time, and simultaneously work on improvements.

So... human drivers.

.. yes? I mean, it depends on what you mean by "accept", but yes we choose the option that has the least risk of killing people, even if it has "issues", because those issues are already encapsulated in the relative risk estimates.

Airline flying is safer than driving even though planes have known issues too.

Really struggling to find a logical point in here. Maybe some nuance I'm missing?

Interesting that you would bring up airplanes. Because when the Max 8 is suspected to have a systemic problem which may repeatably cause crashes, the entire worldwide fleet of them gets grounded. No one makes the argument that a Max 8 with a faulty AOA sensor is still safer than driving your car so whatever.

> Do we really accept a known flaw just because it's still better than the alternative?

Would you really choose an overall-worse alternative because the better option has a single known flaw?

If you're that worried about Autopilot, just don't buy or use it. The rest of the car is still best in class.

Edit: Be less inflammatory.

Adding Lidar is an alternative - but Tesla/Musk keep ragging on it as their cars keep slamming into objects

Reply on news.ycombinator.com