Hacker News Reader

Comment by Andys | Hacker News Reader

The first time I saw it detect pedestrians and cyclists that the human driver hadn't even noticed yet, I realised it is also going to save lives.

View on HN

Considering the resources spent on Tesla's “self-driving” systems… if you used that money on public transport, how many lives could you (improve and) save by reducing the number of cars on the road? (My guess: you could build a massive amount of public transport on a local scale, but not much on a global scale, so you'd probably only save a several dozen lives and significantly improve a few tens of thousands.)

Factoring in that bad drivers start being really negligent when there cars are “self-driving”, I'm not certain this will save more lives than the opportunity cost. But if it's mostly used by good drivers, and bad drivers get off the road, it could be huge.

Public transportation in the US will never replace the automobile. There is simply too much infrastructure to replace and not enough GDP. It's a fine solution for SF, Chicago, and NYC, but not the rest of the country.

Illinois used to have electric rail connecting every podunk town for a nickel. It was called the "interurban" and was scrapped for the war effort in the 1940s. The reason half the towns (that now require cars) even exist is because the train had to stop for fuel and water. Not enough GDP is a pretty sad state of affairs for the richest nation on earth.

The last routes actually closed in 1956 and 1958 - none closed in the 1940s: https://en.wikipedia.org/wiki/Illinois_Terminal_Railroad

Thanks for the correction, I must be misremembering a second hand story.

I found lots of photos from the 50s trying to find more info

https://www.cera-chicago.org/Blog/3315468

US urban sprawl is costing so much GDP it is not even funny. And i don't understand why they don't invest in septic and would rather install a full sewer system for less than 200 people.

that is quite the hopeful view. The High Speed CA rail is now projected as a 100$ billion project. With nothing to show for it.

We will never get massive public transport, no matter how much money is spent. That is the harsh reality

Tesla has spent about $8.5B on R&D in the past decade. Of course, not all of that was spent on FSD, but let's assume it was.

Let's also assume our hypothetical public transport is built in San Francisco. San Francisco recently spent $1.5B on a 1.7 mile subway line. It will also take a decade from when work begun to its opening next year.

That works out to be a 9.7 mile subway that opens in 2032.

You can't conflate overall R&D development with manufacturing and construction costs, that's a fallacy. I don't know what the exact numbers are but Tesla has definitely invested well over $100b in their supercharger networks, manufacturing facilities, and even cost per car and it will still be far less of a public good affecting far less less people directly and indirectly than a reasonable public transport network.

And the need is there. Tesla's previous CTO, JB Straubel, recently lost his wife to a driver who hit her while she was cycling [1]. Everyone on HN shits on Autopilot (and I'll be the first to admit it has flaws and isn't perfect), but humans are just terrible (tens of thousands of deaths per year, hundreds of thousands to millions of injuries). You don't have to be perfect, just better than humans.

The software will improve faster than humans will, as arguably, we've already hit our limit. Software doesn't get tired and it doesn't drive impaired. It makes mistakes, but so do we, and it's mistakes can be improved upon at scale. This is not blind faith in technology or software, but an acceptance of the limited capabilities of your median human (as well as the disadvantaged outliers).

[1] https://www.nytimes.com/2021/07/01/business/boryana-straubel...

Humans are no where near the limit to being attentive on the road. When you consider how many people are either distracted or tuned out to their surroundings, there is a ton of room for improvement, and not theoretical improvement, there are clearly some drivers that are consistently able to make correct decisions, whether there is risk or not.

This is not to say that software and hardware will not quickly overtake us, but while humans are on the road and in control, improvement will be possible.

I really don't think he referring to some physiological limit of individuals. It is the aggregate of society, which includes the prevalence of distracted driving.

That story is heart-wrenching. So young!

And if me make mistakes we face the consequences, who will be trialed for software errors?

We don't face the consequences, at least in my experience seeing the justice system play out. Drunken driver? Affluenza? Old person turning in front of a motorcycle killing the rider? You're getting off relatively easy for taking a life or maiming someone. I've followed the cases for each example given, the last being a close friend. The retiree who pulled out, in their 80s, still has their license and lives comfortably in their retirement community and my friend is dead.

You are right, the software should absolutely be held to a high standard (otherwise we're socializing losses and privatizing profits, also bad!), but please don't perpetuate a lie that existing drivers are held responsible to the degree they should for their mistakes (or even intentional acts) that cost lives. You'll find examples where someone is made an example of, but that is rare.

So disagree, please (checks and balances are important, as is the dialog we're having), but having seen what humans have to offer in this context, I will be the first, on my own time and resources, in front of regulators to argue for aggressive driver assist system development and implementation. Because the alternative is hot garbage we get by with because there are limited alternatives (ie no public transportation most everywhere and the inability to deploy it cost effectively), and we do a disservice to those who die or suffer because we could do better.

This is maybe the best reason to allow autodrive. We need to let it learn because it will get better. Whatever we pay now in terms of lives lost is made back in the future, in lives saved.

All depends on our faith in the system getting better of course. There's always going to be dissatisfying things surrounding it.

Gotta be up front about the state of the art though, can't let people sell a thing like it's the finished item.

I absolutely agree with your entire comment. The system must be taught, but everyone must participate to the best of their ability. Can't mess it up, there are people we haven't even met yet who are counting on us, collectively, to do better.

Did you watch the video? Clearly the car doesn't care too much for what the vision detects.

Not with vision based autonomous sensors.

Reply on news.ycombinator.com