It's a matter of defining "full self-driving."

If you're accustomed to horses and see someone selling cars, it'd be reasonable to say, "Those are ridiculous. There are numerous places only my horses can reach. There will virtually never be a time cars fully replace horses." And you'd be right. It's still true--there are places horses can reach that cars cannot. But because of the value of cars, society has reoriented itself such that reaching those places doesn't matter to the vest majority of us.

I think self-driving cars will be the same. We don't need them to operate in every single road / circumstance ever. There may always be roads that self-driving cars can't drive on. But it'll become irrelevant because the vast majority of places people want to go will become navigable by self-driving cars, whether through the cars getting better or the roads becoming more easily navigable.

I agree and think a lot of smart people still don't really comprehend how non-linear the task is. In other words, they believe if a self-driving car handles itself properly in 95% of situations, the final 5% is right around the corner, when in fact the final 5% (often the difference between life and death) may, in all likelihood, take longer to get right than the previous 95%. It's just that much harder to be perfect than to be pretty good.

Furthermore, even 99.99999% reliability is not good enough, and that's already incredibly hard to achieve. Five 9s of reliability still means 5 minutes of unreliable behaviour per year, which is unacceptable in the context of driving.

> Furthermore, even 99.99999% reliability is not good enough, and that's already incredibly hard to achieve. Five 9s of reliability still means 5 minutes of unreliable behaviour per year, which is unacceptable in the context of driving.

40,000 traffic accident deaths per year in the US.

< 400,000,000 drivers.

So more than 1 fatality per year per 10,000 drivers. There's a 99.99% change that you won't kill someone while driving in a given year.

That's just fatalities.

> In 2017, there were an estimated 2,775,608 police-reported motor vehicle crashes in which at least one passenger vehicle (i.e., a passenger car or a light truck) was towed from the crash scene in the United States, which resulted in an estimated 1,686,240 known passen-ger vehicle occupant injuries.[0]

So 0.6% per driver per year.

So far today I've seen one driver going the wrong way down a one-way road, and one run a red light. (Plenty entering the intersection on yellow and leaving on red, but at least one where it was blatant). I was on the road for maybe thirty minutes.

Do you think the average driver spends less than five minutes per year driving unsafely?

[0] https://crashstats.nhtsa.dot.gov/Api/Public/ViewPublication/...

Ok, point taken, I take my previous comment back. I guess we just have to wait until the risk posed by autonomous driving systems becomes socially acceptable. Right now it just seems like a scary unpredictable risk.

Just being pedantic, but those are seven 9s, not five. For five 9s, the downtime is indeed 5 minutes per year, for seven 9s it's about 3 seconds.

Still an interesting question: is 3 seconds of erroneous/unreliable behaviour per year acceptable from an autonomous vehicle?

I would be interested to meet a human driver who didn't have 3 seconds of erroneous/unreliable behaviour per year.

Depends what's happening. If those 3 seconds consist of randomly veering into something on a highway, that would be... Not good.

Even 5 minutes per year is totally acceptable. 5 minutes is 300,000 miliseconds. Split over the course of 365 days at 86400 seconds per day, we are looking at roughly 1 milisecond of erratic behaviour every 105 seconds. Not enough to crash a car

Perhaps that'd be OK, if the 5 minutes per year is evenly spread across the entire year (including time the vehicle isn't even being used?!)

OTOH, if it means that the vehicle is liable to have a 25-second "blackout" in the middle of your morning commute roughly once a month, it seems very different.

25 second blackouts are too much. However a 1 second blackout every 5 minutes might not be a problem in most cases: you can keep going the direction you are going and other cars that are not blacked out will avoid yours while the computer recovers. Of course this depends on the blackout - if the car is still steering itself but not looking for problems (a much easier problem) it works, if the car isn't steering itself it could run into a wall.

My practical experience with system failures is that they tend to not be evenly spread out in any kind of predictable pattern. Instead they tend to be acute and concentrated at a few points.

I agree that my assumption that failures will be evenly spread out is too optimistic. But on the other hand, if failures are concentrated on a few points, they should be that much easier to detect and correct.

You literally just proved theieindjri's point. (Assuming you're a smart person) these edge cases, even if they are clustered, even if they happen repeatedly, are NOT easy to fix. Partly because there are so many of them, partly because they only arise with a specific combination of events/circumstances that are difficult to replicate, partly because navigating them safely relies on facets of the human brain that AI doesn't perform, partly because the consequences of failure are so severe. "That much easier to detect and correct" vastly underestimates the difficulty of providing safe, automated driving.

My explanation is that some writers in the technical press got taken in by the idea (possibly encouraged by exclusive stories and other favours by the industry) and wrote a series of articles about self-driving cars, how they will change driving for ever, save lives, etc. This started happening a few years ago, around the start of Google's self-driving project.

These writers, being tech writers and not AI experts, didn't understand the subject very well and basicaly just repeated the industry's claims without really understanding the state of the art.

Then people reading those articles who also were not AI experts and didn't have much of a technical background anwyay (except being interested in cars) got taken in, thiking that if the press says that self-driving cars are just around the corner, then they must be just around the corner.

Why did software folks in particular got taken in? Because most people who write software don't write AI software and even many who use it don't understand it. Computer people are used to thinking of computers like magical dream machines, so the common reaction to overinflated claims about AI in general is wonder before surprise and cheering before skepticism.

Doesn't seem obvious to me. FSD using affordable sensors seems like an unsolved problem, and I haven't heard a convincing proof that it is or isn't solvable.

Did high accuracy OCR without AGI seem possible before it was developed?

Yes, because it was clear from the beginning that it would be just a pattern matching exercise. No reasoning necessary, no actual, deep understanding of the written text.

It was always known that pattern matching would be what ML algorithms would be great at - since the 60s or do. Reasoning about and "understanding" things - well, not so much. Some people hoped that this behavior would at some point just "emerge". But they basically resorted to this because they did not see any promising path to these goals.

Where is the high accuracy OCR? All the OCR software I've seen so far is less accurate than me, especially with handwriting, damaged material, odd viewing angles, etc.

It takes a while to internalize what's difficult about the problem. Partly I think it's that humans take for granted their ability to see, which requires no conscious effort but expends 70% of the brain's computational resources at any time. It's not clear AGI is required... computer vision is not known to definitely be AI-complete. It's a very difficult problem but could potentially be solved in the next several years.

A combination of:

- People with a vested interest in FSD being RSN playing a game of topper with each other, investors, and the market

- Extremely rapid progress in supervised learning, especially vision, being extrapolated into thinking along the lines of "How much longer can getting to FSD really take if we've made this much progress in just a few years?"

- A bunch of young urbanites with dreams of never having to buy a car or even learn to drive bought into all the hybe because they so wanted to believe. (And no small part of the backlash is probably that many of the same group feel basically kicked in the teeth at this point as reality sets in.)

If I had to have an opinion, this would be it.

I don't see why self-driving vehicles can't be dependant on additional self-driving-vehicle-related-signage or sensors embedded in / around / near roads, etc etc.

Humans are fairly dependant on additional signage, maps, and navigation tools (Google / Apple Maps).

I would say safety dictates that the car would still have to be safe when such external signals are either missing, wrong or flat out hostile. At that point I don't really see the value they provide.

Make it self driving on specifically marked highways and switch to manual on normal roads. That would free up hours for many people and might even improve traffic congestion.

And where is this going to come from? Our public infrastructure is crumbling after decades of insufficient spending and maintenance as it is. It will take a significant social shift to reverse this trend and stop robbing the future to support short-term political agendas of neoliberal austerity and privatization that merely enrich the rich further.

If we can accomplish such a shift, there are already much more worthy ends to put the money and political will towards: increased public transportation, rebuilding of streets to safely support non-vehicle traffic, public ev infrastructure, etc. Spending massive amounts to subsidize the few rich who could afford autonomous cars would just be further regressive policy.

It takes more than a tech-savvy person to understand the difference between a difficult problem and a Hard Problem. It might be a thought experiment to pose during a technical interview, though you might encounter highly optimistic people that are otherwise very talented engineers.

If it requires AGI, it is difficult to see how it would happen even in that event - ethics would demand "human" rights be considered, which would hopefully quickly rule out sentient vehicular slaves.

> How is it not obvious that FSD most likely requires artificial general intelligence

Because it doesn't. The issue is that all the self-driving car companies ignore how people actually drive.

People memorize their local driving area--it's why teenagers are so incredibly dangerous and why if you put a new stop sign into an area you cause chaos for several months. A self-driving car needs to do the same--the advantage is that a self-driving car can download that "memorization" for any new area it goes into.

Between memorized zones, people mostly "keep station"--they get onto a freeway and keep a relative position with respect to the lanes and the other cars on the road until they get back off.

When something goes very wrong, people panic and random chance takes over the fate of the vast majority of people.

None of this represents general or artificial intelligence. However, solving these tasks requires a lot of very grubby, non-scalable work over a lot of grubby data.

Except that when people travel away from their local area and even rent a car they're not familiar with, they don't go randomly crashing into things. They probably have to pay more conscious attention than on a daily commute. But they don't go around crashing into things.

> Except that when people travel away from their local area and even rent a car they're not familiar with, they don't go randomly crashing into things.

Collision Damage Waivers and the fact that they are relatively expensive suggests the opposite.

Accidents near an airport are more than 50% of auto accidents I have had totally in my life. All of the airport accidents I was literally not moving when someone drove into me.

We vastly overestimate the competence of the average driver.

> The issue is that all the self-driving car companies ignore how people actually drive. People memorize their local driving area (...) A self-driving car needs to do the same

Self-driving car companies have been memorizing ("mapping") areas for years.

> How is it not obvious that FSD most likely requires artificial general intelligence

Waymo car fully self driving three days ago:

https://www.youtube.com/watch?v=t_1YLHW65M8

Wow, I couldn't even properly identify the turn signals in 4 of the 6 turns. It was never as obvious as it should have been. Dirty windows, yes, but nevertheless. In an age of LED-lighting there's no excuse for not having excellent and properly colored lighting.

Now this felt like a student driver, overly busy with correctness, insensitive to the actual experience.

I think these cars should be marked, like student drivers, to signal their inevitable difference in behavior.

Frame shift drives? I would need one of those very badly.