Can we have a less flamey discussion of self driving cars?

[hefferbub]

I know that righteous indignation is the new national pastime, but I would find it a lot more interesting to hear a more nuanced discussion from this community on the topic of self-driving.

Some questions that come to mind:

- Would it contribute positively to our world if competent self driving existed?

- What are the various approaches to get from here to there, and what are their relative merits and risks?

- Assuming some approach showed promise, how should we expose it to real world conditions to allow for validation and ongoing improvement?

- Do the technically-savvy folks on this list think that a heuristic approach (i.e. a bunch of rules written by humans) could ever succeed at effective self driving in diverse conditions? Or is the machine learning + massive data approach more likely to solve it? Or do you consider it fundamentally unsolveable?

- What interesting moral and legal questions arise if effective self-driving becomes available?

- What might be the economic and environmental ramifications of a world with ubiquitous “robotaxis”?

Anyone care to pick one and take a swing at it?

[Cupertino95014]

Sometimes when people say "let's have a civil discussion" they're really saying "first, everyone agree with me -- then we'll discuss how to achieve my objectives."

I'm not saying that's what your post is. However, asking people to assume "competent self driving" could be likened to saying "suppose we had perpetual motion -- how would that change things?"

We have science fiction for that.

[slickdork]

AI/ML just isn't at a point yet where it can encounter something it's never seen and make a decision about it in a predictable way. Which is problematic with driving for two reasons.

1) crazy things happen on the road all the time, none of which are 100% predictable, since humans have free will. Sure, the majority of driving is very predicable, but it's that tiny fraction of clown cars that's a problem.

2) humans are bad in scenarios like the above. Imagine the car has been driving by itself for 200 miles. The human will probably start reading a book or looking a their phone. And then, dang, something happens that requires human intervention. But the human wasn't paying attention. so, there's a crash!

If we can solve the problem of AI being able to think for itself, rather than drawing conclusions based on past data sets, then 100% fully autonomous driving will exist. As will a completely society changing AI!

[h2odragon]

> none of which are 100% predictable, since humans have free will.

Not just humans. The judgement call of "hit the critter with my car or try to avoid a crash while swerving to avoid it" is hard for humans too; how will we program our machines to make that choice for us?

My wife ran into a falling tree branch once, wasted the windshield but no other real damage, fortunately. It was traumatic for her and the circumstance wouldn't have been avoidable, human or computer driving.

[ItsMonkk]

The typical car is on the road 1% of the time. The typical driver-less car will be on the road whenever possible, so will see 100x usage. Just 1% of cars being driver-less will therefore double cars on the road. That will go to things that are good - taking the kids to soccer practice, go to things that are neutral - sleeping in the car as it drives to the beach, and it will go to things that are wasteful, like shipping a single package to a single customer and then going back to the store. But traffic congestion scales exponentially, and cars don't scale at all. So in reality, the entirety of the city will be a traffic jam and you won't be able to get anywhere. Just as we've seen buying GPU's during the crypto shortage, the bots will win, your time can't compete with automation.

So in reality in a world that has driver-less cars, the first thing we will see is a congestion tax. And the first thing we would like to see as a result of that congestion tax is better public transportation. Driver-less buses will do well, especially if they have dedicated lanes. Trains will do better. If you think that you can avoid public spending on public transportation because driver-less cars will save you, you are mistaken. You should be funding that such that if driver-less cars become a reality, but even if they don't, you will be ready.

[MattPalmer1086]

Surely those 100% utilised driverless cars would reduce the other cars on the road?

[ItsMonkk]

I mentioned this in my post - soccer moms who used to drop off their kid and go home, then pick up their kid and go home will save an entire trip.

My argument is nothing more than a fairly obvious case of Jevon's paradox. If something becomes easier or cheaper, it gets used more. Our road networks can not handle being used more.

And there are other side effects of driver-less cars as well. If driver-less cars become a thing, people will slowly retire their cars and go driver-less. As that happens, the required parking lots in malls and offices and shopping areas will drop. It will start dropping the more driver-less cars are used. This will mean that these areas can now build on these parking lots. Malls might see high rise office buildings in their lots, shopping centers might see some nice 5-over-1's. This will allow the higher density, and higher density means the average person will travel less distance, so that could be a huge win. But they can only do that if the road network can handle the congestion, and they won't be able to handle it if you don't plan ahead with public transportation.

But we'll also see the other side, where people who used to not like driving, so they only lived 1 hour away from the office, now live 2 hours away from the office because they can now read in their car.

So once again, if cities are prepared their land owners could reap enormous profits up-zoning their parking lots into highly dense highly walk-able neighborhoods, and if they don't they will lose the ability for these investments to come. It's a first-mover advantage.

[michaelt]

> - Do the technically-savvy folks on this list think that a heuristic approach (i.e. a bunch of rules written by humans) could ever succeed at effective self driving in diverse conditions? Or is the machine learning + massive data approach more likely to solve it?

The only solution that sounds plausible to me is a hybrid approach.

Much like when building a car you can make the engine as complex as you like - but you keep things simple with the brakes. Aircraft autopilots sometimes have three separate processors, programmed by three separate teams.

In a self-driving car, I'd expect a heavily audited core of code to take care of not hitting clearly visible stationary objects like concrete barriers and fire trucks even if other parts of the system relied heavily on ML.

> Or do you consider it fundamentally unsolveable?

I think self-driving vehicles will trade off some types of accidents for others.

For example, you see a running child who disappears behind a truck, then a ball rolls out into the road? A skilled, attentive driver would anticipate that the kid might run into the road. But that needs fairly detailed scene understanding. Accidents like that might increase.

On the other hand, a lot of accidents are due to lack of attention - and a machine can provide consistent vigilance. So accidents arising due to drowsiness and drivers fiddling with the radio could be expected to decrease.

[hefferbub]

I too wonder about the way good drivers anticipate circumstances and prepare themselves to respond. That might indeed be hard to replicate.

But then I realized that detecting these kinds of patterns (kid chasing ball, car next to you that is stuck behind a slow driver, bumper to bumper with a merge approaching, etc.) are precisely what machine learning is good at.

Arguably it could get better at it than any human because it draws not just on the situations one person happens to have encountered in their lifetime. Rather the model that gets uploaded to each car is built from the experiences of millions of cars on billions of miles of driving taking place (eventually) over decades.

Perhaps at some point the quantity of relevant data more than compensates for the greater reasoning and fluidity of a human brain.

It’s hard to know until someone’s tech actually gets there…

[candiddevmike]

Until car companies can be sued into the ground for their self driving car killing someone, there won't be proper incentives in place to ensure the tech is safe and not a move fast and break things race between manufacturers that puts everyone at risk.

[generalizations]

> sued into the ground

The consequences should be severe, but should not prevent the company from finding ways to learn from their mistakes. There's a middle ground here: if there's no consequences, then we get careless implementations without improvement. If the consequence is bankruptcy, then there's no version of the tech that would ever be released by a respectable company. So we need something in the middle, where the consequences are severe enough to motivate good tech, but not severe enough to block the tech altogether.

[candiddevmike]

The tech is inevitable IMO, so I disagree that bankruptcy shouldn't be on the table given the potential for harm (an errant update could result in nationwide or worldwide catastrophies). You want to be the first to deliver self driving? Better make sure it works.

[generalizations]

> The tech is inevitable IMO

That's where we disagree. I think it would be entirely possible to prevent development permanently (or almost...50-100 years or more) if the regulations are too harsh.

Even with some fatalities, I think self driving could reduce total deaths. Perfect is the enemy of the good.

[michaelt]

Well, there are already companies that accept nigh-unlimited liability for the consequences of car accidents - insurance companies. And drivers are used to paying enough in insurance premiums to cover the costs of the payouts.

Accepting liability for car accidents isn't necessarily a death sentence for a car company, any more that it is for an insurance company. Just instead of paying a $x00/year insurance fee you'll be paying an $x00/year self-driving subscription fee.

[h2odragon]

We had all these discussion years ago. The flames are from the "consensus / conclusions" from those debates; people have Received Wisdom and must push it to others or defend it. Going back to first principles and asking "should we be doing this?" is gonna draw ire from all sides: you're implicitly questioning the conclusion people have already adopted and invested emotion into.

Here's one for "Legal questions": At what age can you send your child off unaccompanied in a robotaxi? And how far? If the child doesn't arrive is it parental neglect or should the taxi owners be held liable?

[satisfice]

This is a moral issue. It’s only possible to have a non-flamey moral discussion if people trust each other and share some sort of moral operating system in common.

I am offended, and I think we all should be, at the way self-driving bros characterize this issue as “pro-life” (because self-driving cars that work would save lives) while ignoring that they are inherently irresponsible (because they force human drivers to interact with and adapt to the socially incompetent avatars of those tech bros in a public space).

Driving is a social act. AI cannot be made to be socially competent or to have a social status. Without that, any problem caused by self-driving cars will create rage among the humans it interacts with. They are rolling arrogance-mobiles.

Self-driving cars could work only if human drivers are outlawed and pedestrians banned and all roads standardized. And if you do all that, there is still a problem: you’ve just spent a shit ton of money to solve a tiny problem.

Solve global warming instead.

[fxtentacle]

The way I see it, there's very little negativity about Waymo.

All of the flame discussions are about Tesla FSD because there it's really more a matter of religion if you consider the name accurate or not.

[plorkyeran]

Yeah, I don't particularly expect Waymo to achieve their goals, but they appear to be operating ethically and I don't have a problem with how they're trying to develop their self-driving tech. Other companies in the space are not like that.

[greenthrow]

> Would it contribute positively to our world if competent self driving existed?

Define "competent". Existing isn't enough. In an ideal case of bug free self driving cars making up 100% of traffic on the road, yes the world is a better place with less vehicle related deaths.

Short of that utopia, we have a reality where software bugs cause injuries and deaths and the ultra wealthy companies avoid liability.

> What are the various approaches to get from here to there, and what are their relative merits and risks?

I see no realistic path to the utopian ideal. Even if some company can deliver a truly safe self driving car, it will be a long time before they are ubiquitous. And short of that, you have issues where even if the self driving car behaves safely, it may be acting in ways human drivers do not expect, and thus cause accidents anyway.

> Do the technically-savvy folks on this list think that a heuristic approach (i.e. a bunch of rules written by humans) could ever succeed at effective self driving in diverse conditions? Or is the machine learning + massive data approach more likely to solve it? Or do you consider it fundamentally unsolveable?

Manually coded self drivint has too mant corner cases, so no, I don't believe it is possible.

I don't believe it is possible via ML either. People have way too much faith in ML. I don't think truly safe 100% self driving cars is a realistic goal.

> What interesting moral and legal questions arise if effective self-driving becomes available?

How do we hold parties liable for the damage they cause? So far this doesn't seem to be happening.

> What might be the economic and environmental ramifications of a world with ubiquitous “robotaxis”?

It's not realistic so I don't think it's worth thinking about. It's like daydreaming about a world where nobody dies. Yes there are charlatans pushing this idea. No, it's not happening soon.

[Gordonjcp]

Okay, here's my opinion, for what it's worth:

1) I don't see how a self-driving car would make my life better

2) At present they cannot be safer than human drivers because they do not anticipate a developing situation and only react to them as they happen. Having been in a self-driving car it feels like being driven by a very tired or drunk and inexperienced driver.

3) From what I've seen they would improve road safety only in situations where human driving standards are extremely poor, with inadequately trained and experienced drivers - you just need to look at dashcam footage from the US to see this.

[michaelt]

> - What might be the economic and environmental ramifications of a world with ubiquitous “robotaxis”?

Personally when I hear people talking about a future where nobody owns a car because robotaxis have replaced them, it feels dystopian to me.

After all, when the government of X demands all travel records in their country be handed over to their secret police; or that the people on a 'watchlist' not be permitted to travel? The robotaxi company will have to conform to local laws.

[dane-pgp]

> The robotaxi company will have to conform to local laws.

What do you think car manufacturers have to do?

https://unlikekinds.com/article/ecall-eu-cars-emergency-call...

[nmca]

- Would it contribute positively to our world if competent self driving existed?

Yes, making things that people want (e.g. pay for) cheaper and better is good.

- What are the various approaches to get from here to there, and what are their relative merits and risks?

Offline deep learning. Merits: likely to work. Risks: crashes will occasionally happen.

- Assuming some approach showed promise, how should we expose it to real world conditions to allow for validation and ongoing improvement?

Gradual roll-out in initially in cities with simple conditions, then at higher volume, then in more complex conditions.

- Do the technically-savvy folks on this list think that a heuristic approach (i.e. a bunch of rules written by humans) could ever succeed at effective self driving in diverse conditions? Or is the machine learning + massive data approach more likely to solve it? Or do you consider it fundamentally unsolveable?

rules-based is not workable, ML will work.

- What interesting moral and legal questions arise if effective self-driving becomes available?

moral: none, legal: corporate liability for crashes seems pretty complex.

[xwdv]

1. No, any time people are saving by not driving a car while riding in one will quickly be gobbled up for productivity purposes rather than leisure. Think taking conference calls or working on spreadsheets while commuting to your office. Think truck drivers sitting in their truck all day while the truck mostly drives itself.

2. Going beyond AI, the fastest approach is commercial vehicles controlled by drivers operating from remote offices. Think buses, trucks, etc.

3. Test the cars in locations where people who could possibly be affected cannot afford to create lawsuits for any kind of accidents that may occur.

4. Absolutely not. Edge cases are fairly infinite.

5. Number of acceptable road deaths per year will be higher than what people want. However, the deaths will be more randomly distributed rather than being concentrated mostly on bad or unsafe drivers as we have today.

6. Driving will become a less valuable skill and thus there will be cheaper and fewer jobs available for such people who specialize in it.

[jacamera]

I think righteous indignation is all but inevitable when you're sourcing hot takes on the internet. Your questions are interesting but one could write a dissertation on any one of them. Here on Hacker News you might have a higher chance of bumping into someone who has written such a dissertation and can link you to it, but there will still always be a ton of noise in the comments due to the very format of the forum. If you want to have a less flamey discussion of self-driving cars I think you should flesh out your ideas and questions in a well-researched and well-cited article and then post it here rather than trying to start such a broad discussion with such little context.

[haunter]

>I know that righteous indignation is the new national pastime

This is your very first sentence so probably no.

[aurizon]

When I was a kid, all elevators had an elevator operators. You entered, he closed an outer gate on the hole in the wall and then an inner gate. He then asked you the floor. He then pulled a cable that moved the car up/down, and stopped at your floor - often jockeying a little up/down to level the floor within ~~ an inch. Then came push button elevators with automatic levelling and inner/outer door operation. So you push the button, elevator comes/door opens and you enter and the operator ask you what floor - WTF?. Yes, the operators union in cahoots with elevator installers unions saw the writing on the wall. Buildings were shorter, but few wanted to trudge up/down 10-20 floors. Deliveries carried as most freight elevators worked the old way. These strikes for an un-needed job went on for a while and landlords often gave in and wages went to 5 times the prior wage as you were over a barrel and your pants were down, and in the overall rent of space = small potatoes. Eventually it became so egregious that new buildings who never had operators were able to get pickets limited for places that never had a union to 2 information pickets who could not block access at all. Bear in mind these were safety elevators, with dual safety brakes so the doom and gloom of elevators full of screaming people falling 20+ floors that were featured by the info pickets. The walking public saw it as BS and ignored them and in time they went away. We can easily have fully automatic subway trains. All you need are doors on the platform side to mate with doors on the car side and a controlled set of rails and you could do it. In fact many airport monorails do this now. So, safe robot cars? This may be technically possible now - if you have zero people operated cars who can screw it all up. I have never been on Munk's Boring system in Las Vegas. A video that uses a car and driver is on youtube. https://www.youtube.com/watch?v=djfYafWFWtk&ab_channel=CNET This needs to be changen to 10-12 passenger trollies - in time driverless, Tesla should make these trolley as there is a HUGE market for them in controlled right of way. As we sit random mixed human/robo are prey to erratic humans. An asphalt path on a route like the BART (Bay Area Rapid Transit) is good as long as thuggery was dealt with. Any system will fail if wild humans can screw it up - we see wild humans (AKA homeless) in many cities. This is a complex drug/crime linked 'wild' system that has resisted control. It costs too much in wages to have paid drivers for the masses - they can not afford what the wealthy can.

[Atlas667]

Self driving cars and trucks are inefficient and much more can be achieved with public transportation like trains, trams, buses, etc.

The inherent problem is that these are marketing gimmicks permitted by our current technological development, but it doesn't mean they are useful. All the questions this technology poses are solved in one sweep with effective and comfortable public transportation.

[nathants]

there are two kinda of people who discuss self driving cars:

- members of comma.ai discord

- everyone else

conversation amongst one of these groups is more interesting than the other, and there isn’t much crosstalk.

[Apreche]

If our goal is to save the world, the only answer is fewer cars. Drastically reducing the number of cars and roads and instead using public transportation, bicycles, walking, etc. is the solution to so many of our social problems. Even if we somehow replaced every car with a self-driving electric car, and recharged all those batteries with solar power, it would not save us. If you really want to save the world, stop using cars period.

[gizajob]

I keep thinking the key is to make 100% of cars self-driving, and networked so they can all communicate and negotiate, which takes the unpredictability and lack of attention of human drivers out of the equation, and then pedestrians and cyclists will adapt. If this was the case, then you could probably get rid of traffic lights and speed limits...but it's the kind of solution to accommodate software solutions that could handle it successfully (I don't like the term AI, because I don't think there's any such thing on a philosophical level). The only problem then remains with the small percentage of drivers left that are determined to drive, like sports car users and motorcyclists, who could interrupt the system by taking advantage of the stability afforded by it. Again, this could be resolved by fitting them with beacons, say, that communicate information about where they are and where they're going, but it'll be a hard sell. I'm pretty sure that most people aren't bothered in the slightest about driving themselves to work every day and operating a steering wheel and pedals to do it, if you could offer them a solution where they could be on their phones and don't have to drive themselves.

[cpgxiii]

The problem with V2X (vehicle-to-vehicle and vehicle-to-infrastructure) is that it is impossible to make it secure. Lots of work has been done on V2X protocols and ideas of operations, and none of them solve the fundamental problem that no one has a clue as to how to make such a large scale system secure against hostile actors. At best, every car would have to be continuously solving a Byzantine Generals problem for all V2X inputs, and those "solutions" don't really solve the problem.

Either the car is capable enough of operating by itself that V2X information is non-essential, or it is an extortion and assassination machine waiting to happen. Think ransomware locking away your documents and family pictures is bad already? Try "pay us 10 bitcoins or you die in a car accident on your commute home today".