Sorry to take things off-topic but

>Needless to say we will not be uploading consciousness anytime soon

I'd say that uploading consciousness is an entirely orthogonal problem to AI. Uploading a consciousness is all about scanning the brain and simulating it accurately. It doesn't involve any AI since the intelligence already exists. Conversely if you succeeded with uploading it wouldn't necessarily help with AI, since it would still be hard to see how the brain was working.

So I don't think progress in one field can tell you anything about the other.

I feel like the comment was poking fun at the "fields", which are so hyped that they might as well be called hunting for the philosopher's stone.

Where do you upload it to? What hardware? How much RAM? What's the CPU's clockspeed? Bus speed? Those are the right questions to ask when researching a new electronic gadget's capabilities, but I'm sure those are the wrong questions to ask if presented with hardware that could hold consciousness. Humanity understands so very little about consciousness, that, were we able to simulate it under the umbrella of AI, it would still advance the pre-natal field of uploaded digital consciousness.

I can think of specific exceptions, especially using a very strict definition of the field of AI research, that says that the fields are separate, but until we've succeeded at both, it's impossible to say what was actually necessary in order to get there.

>After that he had a range of little annoying problems, many of which were likely software bugs, but then he (actually his GF) experienced something in autopilot mode known on the Tesla forums as "shadow braking" -- the car braking hard on the highway because it seems to see the shadow of an overpass and think it's a wall. This is extremely dangerous. It was on the I-405 (SoCal) going 70-80mph, and such a thing could easily cause a rear end collision.

Why is he not leaving the beta of Autopilot then? The car would still be better than the competition in most domain (acceleration, navigation, range...) wouldn't it?

Does your friend believe the car is worthless if it isn't on Autopilot? All benchmarks of driver assist systems show that Tesla is far better than comparable cars, so it's pretty dumb to ask for a buy-back for a car because of such a small and extra feature. He could have asked for a buy-back of the Autopilot option but then, your comment wouldn't add to the conversation (that would be a non story, actually, considering all the Tesla owners who love it).

What does he think of the rest of the car?

It's been an overall bad experience. Here's a rough chronology:

(1) Serious and clearly visible body panel alignment issues. They got fixed, but it took a while and Tesla didn't have a loaner car. There was also a scratch on the car when it was delivered.

(2) Door handles not extending. This problem occurred multiple times. The first "fix" was to reboot the computer.

(3) A whole host of software bugs including constant apparently false (according to Tesla) warnings about suspension problems, inability to pair with Bluetooth devices, UI glitches, inability to lock the car (turned out this was a 'feature' but one with no documentation and no UI feedback), etc.

(4) Dangerous autopilot failure modes.

The car's drive train is indeed excellent. It accelerates great, has great range, and performs and drives well. The rest of the car however is really problematic.

The big thing causing him to want to return the car is that Tesla service has apparently been less than helpful and evasive. It's been back and forth, back and forth, back and forth... The dangerous failure modes in the autopilot are the brick that broke the camel's back.

He said he's been many times more patient than Tesla than he would have been with any other car company because he like me believes in what Tesla is trying to do.

Given that Tesla owners are the most satisfied (cf. https://eu.usatoday.com/story/money/cars/2017/12/21/tesla-po...), that implies that the other owners are even more happy with their Tesla that I would have expected.

All surveys about panel gaps points to Tesla being above average (expect in the first few cars out of the line, which are generally owned by employees).

If your friends' Tesla has serious software bug, I'm curious to know his experience with other brands! Unless he can't stand a small bug in an optional, beta feature.

As for autopilot, he knows he can turn the autosteer off and still get the best ADAS available on the market?

>inability to lock the car (turned out this was a 'feature' but one with no documentation and no UI feedback), etc.

I don't own a Tesla and only test drive one and somehow I was aware of that. Looks like PEBCAK to me.

Most of this you’re arguing against someone’s subjective opinions, which makes no sense. They are unhappy with the car and service and are taking advantage of a program to return the car. This is how it’s supposed to work. The customer has less bad will against the company and gets out of a car they don’t like. And other potential customers can see that there is less risk for them is they don’t like their new Tesla.

But this part...

Unless he can't stand a small bug in an optional, beta feature.

That’s not okay. Shadow braking is not a small bug. It’s a dangerous thing that can happen. Calling something “beta” doesn’t work when you’re driving down the road at 70 MPH. That’s just plain dangerous. Putting something like that out into the public as a “beta” feature puts not only the customer but everyone else on the road in danger.

I have no problem with a company testing car features on the road with test drivers who are well trained (with well marked cars). But “beta” doesn’t belong on the highway in the hands of a customer.

>If your friends' Tesla has serious software bug, I'm curious to know his experience with other brands!

Most brands have actual knobs and handles instead of putting everything in a digital interface in the entertainment console.

He showed me pictures. The panel issues were really bad. It wasn't something you had to squint to see. It was visible from a distance. There was also a clearly visible scratch on the delivered vehicle.

Maybe he got a "lemon," but in that case Tesla should have been much more ready to fix or exchange the vehicle instead of dragging their feet and arguing over minutia.

I also question what's happening at Tesla when they even ship a vehicle with such obvious defects to a customer.

My Subaru never shadow brakes. Its driver assistance technology is certainly more rudimentary than Tesla's (only has lane keep assist and adaptive cruise control, and the adaptive cruise control doesn't see speedups far ahead and doesn't catch people switching into my lane until they're already halfway switched), but it is very reliable at doing what I expect to, and not having any dangerous shadow-braking like that.

> My Subaru never shadow brakes.

Do your Subaru is electric? Does it range compare to the Model 3? How does it compare to its acceleration?

Don't ever buy a product for a feature in beta. For the sake of anyone, never ever EVER do that.

My Subaru is not electric. Its range is significantly higher than the Model 3. It has much lower acceleration, but it's plenty to meet my needs (and indeed, I did not get the version with higher engine power because I do not need high acceleration). It also cost me half as much as a Model 3 would.

On the other hand, none of the features on my Subaru are in beta. Most importantly, it has fully functioning adaptive cruise control, which is a crucial feature for making highway drives more pleasant (particularly when driving in traffic), and in this day and age, I wouldn't buy a commuter vehicle which lacks a fully functioning adaptive cruise control system.

Indeed, though I think tesla is part of a strategic solar plan, so losing one pillar could see the roof fall.

"The shadow braking thing really reveals how primitive "AI" actually is-- this is something a bumblebee would not do, and its brain consumes milliwatts of power."

I'm not impressed with self-driving AI so far, but I think you should be fair if you're comparing it to bugs. Insects are notoriously not good at avoiding collisions on the highways.

I'm comparing path finding and other navigational and body maneuvering abilities within the context of their world and constraints. Insects are very good at navigating their space and other very complex behaviors that are at least as difficult as driving a car.

I was comparing the two to illustrate how far behind biology we still are when it comes to AI. Our AIs consume hundreds or thousands of watts and are demonstrably inferior to the brains of insects that consume milliwatts. We are many orders of magnitude behind what biological brains can do both in absolute terms and in terms of efficiency. The difference in efficiency is so large I think it argues strongly for unknown unknowns -- things living brains are doing that we simply don't understand at all.

"Insects are very good at navigating their space and other very complex behaviors that are at least as difficult as driving a car."

And yet, they are commonly seen splattered over the front of cars. Which is failing at exactly the same problem driving AI needs to solve.

You seem to be dismissing this as not part of "their world", but if we're going to compare them at all, I think it is. Conversely, avoiding being eaten is not something driving AI needs to solve.

> After that he had a range of little annoying problems, many of which were likely software bugs, but then he (actually his GF) experienced something in autopilot mode known on the Tesla forums as "shadow braking" -- the car braking hard on the highway because it seems to see the shadow of an overpass and think it's a wall. This is extremely dangerous. It was on the I-405 (SoCal) going 70-80mph, and such a thing could easily cause a rear end collision.

This is a known issue. It isn't a software bug. It is autopilot learning that the overpass isn't a tractor trailer or other large object the camera can't discriminate crossing the path of the vehicle. [1] It was introduced after this fatality [2].

With that said, if that causes concern for the driver, the vehicle isn't for them.

[1] https://www.tesla.com/blog/upgrading-autopilot-seeing-world-...

[2] https://electrek.co/2016/07/01/understanding-fatal-tesla-acc...

Holy shit!!!?!

Potential life threatening maneuvers on the road are a feature of the self-learning Autopilot system???? Because without it the car will just ram itself into a wall???

>With that said, if that causes concern for the driver, the vehicle isn't for them.

What if it causes concern for the driver behind them, what then? My family will get a settlement after a Tesla kills me?

How the fuck are these cars legal anywhere???

> How the fuck are these cars legal anywhere?

Because product defects don't (in general) make the car illegal, they just incur strict liability for the manufacturer (and the chain of commerce, but in Tesla’s case that's moot except as to defects that originate upstream of them, since they sell direct to consumer) when they cause damages.

The driver behind a vehicle is primarily responsible for any rear end collision (US-specific, YMMV elsewhere). Should probably not tailgate a Tesla (or tailgate in general).

> How the fuck are these cars legal anywhere???

They're legal because Autopilot is driver assistance. The driver is still 100% responsible, just as they would be for cruise control.

> The driver behind a vehicle is primarily responsible for any rear end collision (US-specific, YMMV elsewhere).

Not if you slam on the brakes without cause on the highway.

Here is the general state statute you're up against (as the person rear ending someone on the interstate):

“Any person driving a vehicle on a highway shall drive the same at a careful and prudent speed not greater than nor less than is reasonable and proper, having due regard to the traffic, surface and width of the highway and any other conditions then existing, and no person shall drive any vehicle upon a highway at a speed greater than will permit the driver to bring it to a stop within the assured clear distance ahead.”

Maybe if you have a dash cam. Maybe if you can prove there was nothing in front of the person in front of you. Lot of maybes.

You've described the statute in some state for spacing between cars or something, but in some states "brake checking" is mentioned explicitly in the criminal code, and in others it can be prosecuted as vehicular assault. It's harder to prove without a dashcam, but if there are witnesses it is certainly possible.

That's a bit of a sideshow. Tesla needs to worry about the civil liability involved in making a car that stops suddenly in the middle of the road. For example, "The driver is still 100% responsible, just as they would be for cruise control" is absolutely not going to impress anyone as a defense if there is enough property damage or personal injury involving such a feature to justify a class action.

> For example, "The driver is still 100% responsible, just as they would be for cruise control" is absolutely not going to impress anyone as a defense if there is enough property damage or personal injury involving such a feature to justify a class action.

Class action vs. direct action is immaterial; Tesla can't disclaim it's liability for manufacturing g defects even if it tried, and stating that the driver remains responsible for following traffic laws isn't even superficially a disclaimer of product defect liability anyway.

Product defect liability is strict liability for damages produced, and there's a fairly high bar for excluding liability due to intervening causes which would not generally apply to the simple negligence of a driver (either of a Tesla or the other involved vehicle) which contributed to a collision where the defective operation of an assist system was also a contributing cause.

If you’re not at a safe following distance behind the car in front of you, regardless of why they’re stopping, your driving skills are deficient. Following distances don’t change because it’s a Tesla; you shouldn’t follow closely as a tire blowout or a piece of debris falling into the path of travel of the vehicle in front of you is just as possible. Im sure there are states with more strict statues; they aren’t the majority of states.

Anyone can sue anyone in the US. Take it up with their general counsel.

Being in a car is life-threatening.

Which is why there are so many laws and regulations surrounding them, the most basic of which is a technical inspection every few years.

If your brakes don't work your car is illegal on the road (at least here in Europe). But apparently if its a software bug in the brakes then its okay to drive it around...

The number of question marks and exclamation points in your original post does not fit the relative severity of the issue you are describing. If you used that many question marks to describe a car with an imperfect auto-braking system, I can't imagine how many you'd use to describe a car with a human braking system. In the U.S., those humans fail to brake, with consequences, 2.5 million times a year. https://www.scmitchell.com/blog/2017/may/rear-end-collisions...

Even when self-driving cars become popular, people will still be injured and die in them. And it won't be just a subset of the people who would have died without such cars. It will almost certainly be a different, randomly selected group of people. If it's your time to go, does it matter to you whether you died because a texting teenager slammed into your stopped car? Or because an edge case triggered in a car's software? The former case is just stupid. The latter case is something society can learn from and prevent from ever happening again.

If you are in favor of more people dying in "normal" accidents, then by all means, please continue expressing your outrage.

The cost of progress in this area is computers making mistakes instead of people making them.

The US NHTSA and state regulators have allowed Tesla to continue to offer Autopilot, therefore I believe your argument to be invalid (versus that of domain experts), although you are entitled to your opinion.

There is no way that Autopilot is worse than human drivers: https://i.imgur.com/au8A1o3.gifv

Oh well then I guess high speed braking maneuvers on a highway without user input are just fine then.

Hard braking at 80mph isn't as dangerous as not seeing a tractor trailer, but it could definitely cause a serious accident.

I'm starting to understand why all the other self-drive companies use active sensors like lidar. You need that to compensate for the fact that the AI actually driving the car is less intelligent than a honeybee.

A self-drive "AI" is dumb. It's just a glorified expert system with a lot of pattern recognition and other stuff bolted on. But the performance of dumb things can sometimes be improved by giving them more information. Self-driving cars need super-human senses.

Lidar is doubly special because its an active sensing system. If the car hits an undecidable edge case it can probe it to see if it's physical or a lighting artifact. It needs to be able to do this because it has no higher order conceptual model of the world to reference. It doesn't "know" what an overpass or the sun "are."

Edit: This principle is also why "big data" has helped revive "AI." We need big data because our AI isn't very I. Biological intelligences do incredible things with very small amounts of data because they have more I. What's so impressive about biology is what it can do with small data.

> I'm starting to understand why all the other self-drive companies use active sensors like lidar.

Every roboticist since the DARPA Grand Challenge knew this, as every team that finished had a Velodyne LIDAR. It was THE piece of tech that made that race possible. So for everyone who learned that lesson, to then see Telsas out on public roads with a feature called "autopilot" that did not have any LIDAR sensor attached to it... well... Let's just say we've been saying "I told you so" since then.

I think it's a little early for people who don't "do" to call it that LIDAR is mandatory, and that optics and radar aren't sufficient. Autopilot is still in development.

LIDAR is not necessarily categorically mandatory, but shadow braking and other autopilot failure modes demonstrate that it or something with its capabilities are probably necessary right now to enable today's "AI" to safely drive a car. My point about active sensing being able to decide edge cases explains why this is likely the case.

IMHO doing without precise active sensing would require a breakthrough in AI that hasn't happened yet.

On the plus side the fact that Tesla's autopilot is able to work as well as it does without lidar might indicte that with a lidar system it might be the best autopilot on the market. Who knows.

Seems weird that the Tesla can't infer that there is no wall there from GPS data.

What if a wall is there tomorrow?

They do over time. The vehicle “learns” from no impact at the overpass and uploads that to the mothership for fleet-wide distribution.

Obviously LIDAR is not mandatory. We have a billion proof of concept controllers that only use binocular image data to navigate just fine. So one day ostensibly that could be enough. But we need a computer as good as the human brain to make it work, and that's not happening anytime soon. So until then, dropping LIDAR from the sensor suite is really going to limit your performance.

It's clearly a software defect. I don't know how you could possibly claim otherwise. The reasons behind the defect are irrelevant from a customer standpoint.

A human driver who did that during a driver's license test would be failed.

"If you don't want your Tesla slamming on the brakes on a highway at 80mph, don't buy a Tesla."

You must own a lot of Tesla shares to make such a ridiculous statement.

> You must own a lot of Tesla shares

I do, but I also entrust my wife and kids in a Model S. I specifically bought the Model S (a used 75D) for them due to its safety record. That's worth far more to me than any Tesla shares. With that said, my wife is aware under what conditions it is safe to engage Autopilot, and is also aware that she is still the responsible party if an accident happens when it's engaged. She doesn't use her phone, she doesn't not pay attention, she is babysitting the vehicle while at speed and still has final authority.

> to make such a ridiculous statement.

You think the statement is ridiculous. 40,000 people a year die in the US driving due to human decisions. Several friends I attended high school with were killed in car accidents. Loved ones have been killed in car accidents. I don't think my statement is ridiculous in that context. These sorts of efforts are necessary to move the needle on transportation safety. Perfectly happy with my TSLA shares going to zero in the process, as long as progress is made.

The ridiculous thing I was commenting on was your suggestion that if sudden high-speed braking maneuvers on a highway were "cause for concern" then you shouldn't buy a Tesla. MV death statistics are completely irrelevant to that point.

Your issue isn't with my suggestion, it's with regulations that specifically allow Autopilot to continue to be sold and engaged in Tesla vehicles.

"The vehicle could do X while under your control. If you are not comfortable with X occurring while under your control, do not buy the vehicle (or buy the vehicle, but do not purchase Autopilot). The vehicle feature that could cause this is still permitted by regulators to be functional. You are solely responsible for the actions of the vehicle while operating it."

As someone who may be behind a auto-brake slamming Tesla one day, where is my opt-out from this dangerous beta?

> where is my opt-out

Don’t tailgate.

That's not a panacea. Realistically, you can easily be rear-ended because you slowed or stopped for a traffic jam caused by something like that Tesla. It is a fantasy that people are in control of every significant danger around them.

Your stopping distance should always allow for the slamming on the brakes of the car in front. If not, you are tailgating by definition.

The scenario is that I am at a standstill because the car in front of me stopped, and am then rear-ended by someone who didn't notice traffic was stopped. Legally, the car behind me is probably responsible, but the car in front of me contributed.

A randomly stopping car is dangerous even to people who aren't tailgating. A human would be charged with dangerous driving for that - which is a felony in my jurisdiction.

Both behaviours aren't great, but the at-fault driver would be at the back. Even though the stopping frequently randomly is likely more dangerous.

Fault isn't necessarily exclusive, and drivers of the front vehicle in a rear end collision may be at (legal) fault. [0]

More importantly for discussion of an automated system shadow braking, however, fault of drivers based on negligence (or more serious mental states, like recklessness) is a separate concept from fault of manufacturers for product defects (which is based on strict liability, not negligence, and therefore bars to liability based on negligence of the injured party due to contributory or modified comparative negligence theories would not generally apply; liability might be barred on a theory of superseding intervening cause, but that would require the intervening cause to be unforseeable—too close freeway following is not so rare as to be an unforseeable intervening cause and thus would not usually limit recovery on a strict liability product defect claim.)

[0] see, e.g., https://www.nolo.com/legal-encyclopedia/is-fault-automatic-r...