Nation's first known self-driving car fatality happened in Williston (FL)

In reply to Boost_Crazy:

Unless you were reminded vividly and frequently it would have approximately zero effect on your behavior. We have to dismiss the low probability risks to be able to live a normal life. There are too many.

Is this where everyone went after the brexit thread?

They threxited to here

Boost_Crazy wrote: In reply to Keith Tanner: But will it see the three year old bouncing the ball on the sidewalk and slow before he darts in into the street, or will it simply react to him after he darts? I have a three year old, and his trajectory can be very hard to calculate.

Possibly. When I see toddlers playing near the road ahead of me, I slow the berk down - and sometimes even stop. Getting to my destination 15 seconds sooner is not that important. Children are unpredictable and often don't pay attention to their surroundings. Residential areas have 25 mph speed limits for a reason.

I agree the accident is very unfortunate. But just as drivers make mistakes, so will computers. However, one advantage of computers is when one "learns" a new situation, that information can be passed along to other computers in the system (cue: Borg...).

The accident reminds me of a civil suit my mother was on a jury for back in the 70's. Tractor trailer making a left turn across a divided hwy. Motorcycle rider traveling the on-coming lanes. Maybe he didn't see the truck. Maybe he thought he could go under the trailer (he couldn't). We'll never know. Survivor's family sued the trucking company and lost when the lawyers couldn't prove they or the truck driver did anything wrong.

As far as future automated cars with systems that communicate with each other, I could imagine non-automated cars being required to have a transponder mounted to identify the car and it's basic capabilities. And if it comes to a point when I won't be allowed to drive my classic cars on major limited access highways, I'm not sure I'd consider that a bad thing. While my Spitfire, GT6 and Mini can do hwy speeds, they are not in their element when doing so.

I have an issue with the use of "extremely rare circumstances" in the main story. Don't know about anyone else, but I often have tractor trailers pulling out across my lane of travel. They are big and slow and very easy to see. It's not like they just jump out in front of you like a dog or a rabbit. And yeah, if your car can't detect a vehicle over 60 feet long and up to 80,000 pounds, I'd say it needs a big upgrade.

I drive 60,000- 70,000 miles per year.

If I could do 80% of that in an autonomous car, I could be a better father.

The time I spend driving would be spent doing my online office work and general surfing. The time I currently spend doing my online office work would become family time.

Plus, I'd lease the car out (Uber style) during times I was not using it (nights, weekends, and chunks of day time), so it would create a much needed revenue stream for me.

I love it when people think they can out think a computer!

There was a time when chess Grand Masters could win against a computer. But the machines consider billions of variables in every move, and can no longer be beat. Garry Kasparov was beaten, and he was the best in the world.

Some of the variables they consider are "fuzzy logic"- moves and decisions made by opponents that are not logical, or based on their personalities, culture, or age, not the progression of moves.

Similarly, a driverless car will keep getting better. A kid bouncing a ball will not be an unpredictable event- it will be as normal as counting. There is no way a human can ever make the number of decisions a computer can, and it is just silly thinking a human will ultimately be able to drive better than a computer.

This incident is a good thing. Accidents are going to happen, and every time they happen they will enable the changes necessary to prevent them from happening in the future. This one happened in Beta, instead of to the general public. I'd say that means it will never happen again.

There is nothing special about humans that makes them more capable of operating machinery than a computer.

GameboyRMH wrote:

Keith Tanner wrote: Interesting article on Jalopnik about a possible blind spot in the Tesla's sensors that makes it less likely to pick up a tractor trailer straddling the road. http://jalopnik.com/does-teslas-semi-autonomous-driving-system-suffer-from-1782935594

Huh, this suggests that the color of the trailer was unimportant as the car does indeed behave as if it's height isn't much more than that of the car's hood...that's a huge and basic design flaw, it's hard to believe Tesla made a mistake that huge.

Maybe there was I typo in the programing. Instead of driving like a Lambo it was set up to try limbo.

I'm just shocked that the car depends that much on a visible light camera. I'm quite certain it'll be fixed but that doesn't help this fella.

I'm also surprised they don't limit the speed to the speed limit when in autopilot.

SVreX wrote: There is nothing special about humans that makes them more capable of operating machinery than a computer.

I think you're underestimating just how hard AI actually is.

Today's "autopilot" and "driver assistance" features are impressive and cover a lot of the core driving tasks, there are also a lot of other things they don't do, and won't be doing anytime soon. Things that require actual thought to process. Things like a police officer standing in the middle of a malfunctioning intersection, directing traffic and waving you into the oncoming traffic lane to get around. Detour signs for road construction, or roads that have been converted to one-way recently and the map database hasn't been updated yet.

Computers are very good at well-defined, carefully-delineated tasks. Chess is a great example -- it's a well-defined game with simple, easy-to-explain rules, and all of the complexity emerges from the combinations of those rules. The guys at IBM wrote some great software, did some very clever things, but the writing had been on the wall for a long time that the reign of the human chess grandmaster was coming to an end. Computers are getting faster and faster (following Moore's Law), so even without the work they did on Deep Blue it's likely that the following 5-10 years of speed improvements would render the dumber algorithms capable of brute-forcing their way to victory.

The real world is not well-defined, you can't brute-force all the possibilities. Computers don't think, mainly because we don't really have any idea how "thinking" actually works. The allure of human-level AI is huge, it would make so many things work so much better (well, ignoring the SkyNet questions), but it is not an evolutionary process from the computers we have today. Real AI is fundamentally different, and we don't know how to do it.

In reply to SVreX:

I love it when people think they can out think a computer!

Codrus just covered it, but you wrote "think," when you should have written "calculate." Computers hands down can calculate a set of variables much faster than a human. If all of the variables are identified and defined. But if one is missing- we will get incidents like the one we are discussing. There are a lot of variables out there. Computers are getting more powerful. But if autonomous vehicle had even a fraction of the computing power of a human brain, it would need a flatbed trailer with "Wide Load" banners to haul it around, and it would take more juice to run the computer than the car itself. So yes, people can out think a computer.

The question is, can a computer do the job of driving a car good enough to surpass humans? I'd say the short answer is no. The average human has the ability to drive from A to B without incident. The computer's advantage is that it would be more consistent mile after mile. But when something goes wrong, it can go very wrong in ways that you wouldn't expect to happen with a human driver.

Communication between cars adds another layer of advantages and disadvantages. On paper, it's a great idea, and has the potential to vastly increase efficiency and safety. But it adds more variables. Variables supplied by sources outside the car itself. Consequences could range from one can erroneously transmitting flawed data to a group a cars, to someone doing the same intentionally. They would need to lock that down, no easy task.

You don't need full AI to drive a car with a computer. You just need to be able to handle the car driving part. Your car won't be able to beat a chess grand master or write a symphony.

Things like detours will require real-time traffic updates. That's not very difficult, it's basically already happening with apps like Waze and Google Maps using both user-submitted input and cellphone tracking. Your cop directing traffic into a different lane could be replaced with a series of cones to close a lane, or they could be equipped with a signalling device that tells local autonomous vehicles to change to a different lane. The smartphone in that cop's pocket already has the processing power and networking capability to perform this task.

This is the sort of thing Google's been working on with their little buggies. They're driving the cars around with a human driver and exposing them to all the random things that happen, and training the car how to respond. As noted earlier, once one car learns it they all learn it.

Sure, there are new and exciting ways this could fail. But any of us who looks into other cars around us already knows of the new and exciting ways that humans fail. I've seen people flossing their teeth. People get tired, they get emotional, they get distracted. I wasn't joking when I said the computer only had to be slightly better than the average driver at an average time in order to increase safety, it's a matter of averages.

In reply to Keith Tanner:

I guess the real test will be how this is handled legally. I'm interested in how insurance companies will respond. There is a big step between Tesla saying it's a feature, but the driver is responsible at all times Vs. no driver required. That is as much a leap legally as it is technologically.

In reply to codrus and Boost_Crazy:

AI is completely unnecessary to have automated vehicles. I think you are underestimating just how advanced the current status is among virtually ALL manufacturers.

Current projections:

So, if you are just in the mood to argue, it's not me you are arguing with. I suggest you take it up with basically the entire automotive industry.

So what if it's hard? The largest number of the greatest engineering minds in the history of the world are all working on the problem. It's GONNA happen (quickly). We can choose to sit around and argue about it on the internet, or we can choose to make productive decisions on how we are gonna handle it.

I'm placing a deposit on a Tesla. That's my decision. What are YOU gonna do?

Keith Tanner wrote: Ah, but autonomous cars can identify all those things - every other moving vehicle and pedestrian in the vicinity. Their velocities can be monitored and trajectories calculated. If one starts to move in the wrong direction, the car can react. That's really all people do. People can't monitor all of them simultanously, though, and may go through a moment of denial ("that guy can't possibly be turning!") that the computer cannot. People can also be distracted, even if it's a matter of adjusting the volume on the stereo or seeing a billboard. Computers can't. It's the 98% of drivers that consider themselves better than 98% of all other drivers that will have trouble accepting this. But just like ABS can modulate brakes better than any driver because it can modulate individual wheels, even the best driver can't pay full attention to everything all the time. It's not physically possible. You have to meter out your attention between what you have calculated to be the biggest threats. If one of the other things in your vicinity turns out to be a bigger threat than your expected, you're taken by surprise.

Keith, I teach driver education and am not afraid of the day I'm not needed. You are correct in your assessment not just of drivers but of the autonomous systems. They don't get bored, drowsy, distracted, annoyed or overwhelmed. Anyone who thinks they're better doesn't understand the scope of just how much information they are actually missing while driving. There are no perfect drives and no perfect drivers. But, autonomous systems will do for the motor vehicle death rate what the lack of better and more stringent education has failed to do.

In reply to

AI is completely unnecessary to have automated vehicles. I think you are underestimating just how advanced the current status is among virtually ALL manufacturers.

We are having this discussion because a car didn't "see" and 18 wheeler and drove into the side of it, killing the "driver." A new, well maintained car, with nothing "wrong" with it. How can you say with a straight face that there is no arguement to be had? There are still significant technological, legal, and societal hurdles to be overcome. Of course auto manufacturers are investing in the technology. There is a chance that it can be developed and pay off, and they don't want to be left on the sideline if it does. That doesn't mean that they are going all in. Auto manufacturers have historically invested in numerous technologies that haven't come to fruition. Autonomous cars are making progress. But all it takes is one bad software update that leads to a rash of accidents, and it's over.

I'm placing a deposit on a Tesla. That's my decision. What are YOU gonna do?

Good for you. I'm sorry that you chose a lifestyle that requires that you drive 70,000mi a year, but you don't like to drive. I'll continue to drive myself, and enjoy every mile of it. The automobile is a symbol of freedom and independence to me. I believe autonomous cars are a large step away from that. I really hope I'm wrong, but I can see a lot of senarios that result in those of us that enjoy driving our cars losing that privledge. As others have mentioned, there are problems with human driven and autonomous cars sharing the road. People like you will be complaining "but, that's not what I wanted. I just wanted to take a nap on my commute." But it will be to late. I hope you enjoyed your naps, because your freedom was taken away while you were sleeping.

In reply to Boost_Crazy:

You are exceptional at making dramatic statements beyond what was said.

I didn't say there are not hurdles to overcome. I said AI is completely unnecessary to have automated vehicles. The capacities we have now are sufficient- as the AI develops, the capabilities will expand.

I also didn't say I don't like to drive. I love to drive. I just happen to have 70,000 miles per year of commuting (which I did not choose, but that's a different story).

You are welcome to drive. No one is stopping you. I plan on doing it as often as I can. But an automated commuting pod will be a welcome addition to my stable, and I will earn money from it. In fact, I will own several, and they will all earn me money.

1 man died in Beta testing. How many test pilots do you think have died so you can enjoy your commercial airline flight from coast to coast? (The answer is about 1 per week in the 1950's).

Autonomous cars will not limit my freedom. They will expand it. Like the internet, air travel, or the Interstate highway system. But only if I learn how to embrace the changes and opportunities, instead of standing in the corner like a Luddite with my fingers in my ears crying, "I don't wanna give up driving, I don't wanna give up driving!"

Im not one prone to conspiracy theories, but I can legitimately see a future where "for safety" we prohibit the use of non autonomous cars on public roads. The profit available to both require replacement of every vehicle in the US, and of services like UBER and Google and the trucking industry to remove the paid driver will provide impossibly large pools of money to lobby with. It very much will be good-by to your freedom to move around the country. Your autonomous car will be required to tell the DOT where it is at all times and it will be illegal to drive yourself. Sure the extremely wealthy will be able to have license to maintain user operated cars but the costs will be to high for normal people.

I really hope I am wrong but the needle seems to be pointing that way.

In reply to SVreX:

If I'm being overly dramatic, it's because you are being overly simplistic.

Don't you remember saying this?

I love it when people think they can out think a computer! There was a time when chess Grand Masters could win against a computer. But the machines consider billions of variables in every move, and can no longer be beat. Garry Kasparov was beaten, and he was the best in the world. Some of the variables they consider are "fuzzy logic"- moves and decisions made by opponents that are not logical, or based on their personalities, culture, or age, not the progression of moves. Similarly, a driverless car will keep getting better. A kid bouncing a ball will not be an unpredictable event- it will be as normal as counting. There is no way a human can ever make the number of decisions a computer can, and it is just silly thinking a human will ultimately be able to drive better than a computer. This incident is a good thing. Accidents are going to happen, and every time they happen they will enable the changes necessary to prevent them from happening in the future. This one happened in Beta, instead of to the general public. I'd say that means it will never happen again. There is nothing special about humans that makes them more capable of operating machinery than a computer.

Some of what you described is AI. You said think. That is AI. Outputs based on sensor inputs is not thinking, it's programming, big difference. Reactive Vs. Proactive. Chess is a horrible example, and not applicable to the discussion. With chess, every variable can be programmed in, there are no unknowns. To be comparable, you would need to program a car with every possible situation ahead of time. They are still working out "don't drive through a semi trailer." I'd say that qualifies as a long way off.

Do you know what is special about the human brain? Scientists recently created a similation of human brain activity. For one second. It took rooms full of computers 30 minutes to load that one second of activity, and it took a tremendous amount of power. To simulate something that fits in our head and uses the energy equivalent of a light bulb. So a comparison between the human brain and a computer that fits in a car is laughable, but for the opposite reason than you think.

Boost_Crazy wrote:

We are having this discussion because a car didn't "see" an 18 wheeler and drove into the side of it, killing the "driver." A new, well maintained car, with nothing "wrong" with it. How can you say with a straight face that there is no arguement to be had?

You ever lost another car in the sun? I have. That's essentially what happened to the tesla. The sensors and computing will never be perfect. But they don't have to be. They only have to be better than the average person to result in a net increase in safety. And that's a pretty low bar. Even in this early pre-production beta attempt, Tesla has logged FAR fewer fatalities per mile than human drivers at large.

In reply to nocones:

Im not one prone to conspiracy theories, but I can legitimately see a future where "for safety" we prohibit the use of non autonomous cars on public roads. The profit available to both require replacement of every vehicle in the US, and of services like UBER and Google and the trucking industry to remove the paid driver will provide impossibly large pools of money to lobby with. It very much will be good-by to your freedom to move around the country. Your autonomous car will be required to tell the DOT where it is at all times and it will be illegal to drive yourself. Sure the extremely wealthy will be able to have license to maintain user operated cars but the costs will be to high for normal people. I really hope I am wrong but the needle seems to be pointing that way.

That is definitely one of my fears, and I think that had a legitimate reason to be part of the discussion. Most likely, driving your own car would become prohibitively expensive through taxes, fees, etc. to discourage the behavior without having to outright ban it. My fear is that there would be a push towards that, even if autonomous cars are not what they are promised to be. How good is good enough? No more crashes? Not likely. So a 50% reduction? That's a lot of lives. 25%? How about 10%, that's still progress. Will we remove the steering wheels for 10%?

In reply to Boost_Crazy:

The capability exists NOW. If you'd like to call that overly simplistic, so be it.

You won't score any points with me fighting a semantics argument because I used the word "think" improperly. Yes, I know what AI is.

Your fears will not change what is happening. I intend to enjoy the opportunity to see the future unfold in front of my eyes and actually be a part of it. You can too.

...or, you can just lease your transportation needs from people like me when your insurance rates hit $5K per year. Your choice.

In reply to SVreX:

Dude, you could replace "driverless car" with "flying car" and this conversation would have been from 60 years ago.

...except that cars were not flying 60 years ago, and the opening post of this thread shows that they are now driving.

What are we disagreeing about? The fact of whether or not it will happen? The timeframe? The scope? Whether we like it or not?

I am not understanding the fight.

Boost_Crazy wrote: Do you know what is special about the human brain? Scientists recently created a similation of human brain activity. For one second. It took rooms full of computers 30 minutes to load that one second of activity, and it took a tremendous amount of power. To simulate something that fits in our head and uses the energy equivalent of a light bulb. So a comparison between the human brain and a computer that fits in a car is laughable, but for the opposite reason than you think.

In the 1940s it took that same "rooms full of computers" literally a week to solve a quadratic equation. That's something that can be done in milliseconds by your phone right now. Microprocessor development has followed a logarithmic path and we're way up the hockey stick from the 1940s. It'll be a few years before that room full of computers simulating the brain fits on a desktop and a few more years before it's small enough to fit in a car but it'll likely be within our lifetimes - speaking as a recently turned 40 year old. Assuming, of course, that the simulation actually achieved what they thought it achieved; I'm not familiar with that research specifically and not too confident that they've achieved that. My point is that what's in the lab today is in the office tomorrow and in your pocket the day after that.