What happens when the software develops a bug? Be it in the middle of rush hour traffic, or hurling down the highway at 70mph?
For all of the great things we know self driving cars can potentially bring to the table, how do we account for the issue of buggy software?
I could say the same about mechanical or electronics of a vehicle
"I've always been leery of vehicles. They run on mechanical parts. Mechanical parts are only as good as the mechanical engineer that designed them and the quality of material. What if a newly introduced part was not well tested and developed a stress failure due to design or material deficiency. Be it in the middle of rush hour traffic, or hurling down the highway at 70mph? How do we account for the issue of buggy mechanical parts or electronics "
See how that works? Everything other than your own two legs is man made. It's just the level of maturity of the process and testing and time that eventually gives you enough confidence to start trusting any human made device, whether it is software, mechanical, electric, electronic etc...
Software is as good as the best team we can find to write it, and then it can get constantly improved - unlike meat drivers, whose firmware cannot be easily flashed. Moreover, when switching from human control to software you're trading the need for quick evaluation and reaction of driving conditions repeated times number of drivers for a long, concentrated time spent on development and testing. It seems to me like a good trade-off. You can spend all the time you want writing, testing and fixing software up until you reach the point it drives better than humans.
And then you have usual, obvious advantages - self-driving cars won't get tired, emotional, distracted by texting and definitely won't be driving in blatant disregard of traffic laws because "they're pointless and that's how real drivers do things".
There are lots of examples of not being able to stay in lane, pulling out, not seeing stopped traffic, not paying attention to surroundings. Humans can't even get the trivial simple things right in broad daylight.
(Yes I am aware that you are in the 97% of drivers who are absolutely convinced they are above average. That doesn't make it true. Yes the video mainly has clips from Russia - you can find USA based ones too. And a 9/11 happens on US roads each and every month. People are dying in large numbers because of human drivers.)
Edit: added warning about video content
Plus it's a huge security risk. Cars are internet-connected these days.
I believe that self driving cars will primarily be owned by companies with fleets of them, rather than individual ownership we have today. It doesn't make sense to have them sitting idle 23 hours a day. The owner companies will have a very strong incentive for safety since even rumours would be bad for business. They will also be in a better position to do something about it due to having multiple vehicles.
Not if you're in a highway.
A plane can't just let go of all control surfaces. Heck it can't even leave them where they are. It must do something and always must provide some degree of control. A flying plane is moving at several hundred miles an hour and cannot stop.
Well, I agree. That's why I expect that, along human drivers, the very concept of car ownership will disappear. It's much more efficient to have those cars available on call; they can serve someone else while you're not using them, and they can be parked&recharged at night with much better space-efficiency than cars are parked today.
I guess chess computers can never beat a human then. Because the software is only as good as the guy writing it.
> What happens when the software develops a bug? Be it in the middle of rush hour traffic, or hurling down the highway at 70mph?
What happens when a human is inexperienced/texting/drunk/sleepy? Humans have plenty of "bugs" but you don't notice them.
> For all of the great things we know self driving cars can potentially bring to the table, how do we account for the issue of buggy software?
Trials. Testing. Every accident (which is inevitable) will be scrutinized and used to improve the system (unlike the 40k annual deaths in the US where that unfortunate experience helps nobody, except perhaps installing a traffic light or barrier).
Software already runs our lives, and worrying about the quality of the underlying software is a strawman. When was the last time a code bug took out a car or an airliner? Some industries have very stringent requirements and QA processes to make sure the code is bug free and can't kill people.
Also, Google's self-driving car is already a far better driver than a majority of the drivers on the road today http://theoatmeal.com/blog/google_self_driving_car
Personally I can't wait until the human aspect of driving is completely removed from day-to-day travel.
Apparently infrequently but it does happen a lot of people end up dying.
https://en.wikipedia.org/wiki/Aviation_accidents_and_inciden...
"The crash of Japan Airlines Flight 123 on August 12, 1985 is the single-aircraft disaster with the highest number of fatalities: 520 died on board a Boeing 747. The aircraft suffered an explosive decompression from an incorrectly repaired aft pressure bulkhead, which failed in mid flight, destroying most of its vertical stabilizer and severing all of the hydraulic lines, making the 747 virtually uncontrollable."
"On March 3, 1974, Turkish Airlines Flight 981, a McDonnell Douglas DC-10, crashed in a forest northeast of Paris, France. The London-bound plane crashed shortly after taking off from Orly airport; all 346 people on board died. It was later determined that the cargo door detached, which caused an explosive decompression; this caused the floor just above to collapse."
"On February 19, 2003, an Iranian military Ilyushin Il-76 crashed in mountainous terrain near Kerman in Iran. The official report says bad weather brought the aircraft down; high winds and fog were present at the time of the crash.[21]"
"On May 25, 1979, American Airlines Flight 191, following improper maintenance and the loss of an engine, a McDonnell Douglas DC-10-10, lost control and crashed near O'Hare International Airport in Des Plaines, Illinois. The crash resulted in the deaths of all 271 passengers and crew on board, as well as two people on the ground."
"On July 11, 1991, Nigeria Airways Flight 2120, a Douglas DC-8-61 aircraft operated by Nationair Canada, crashed in Jeddah, Saudi Arabia after two tires ignited upon takeoff, leading to an in-flight fire. All 261 people died."
I could go on, but I think you get the point.
If you're looking for something that could be software related try Air France Flight 447, though even that is ultimately pilot error. You could also use Ariane 5 flight 1, where a software problem caused vehicle failure.
Try diagnosing someone suddenly getting pissed off and cutting off another driver which sets off a chain reaction of the events leading up to a multi-vehicle collision. The driver may attempt to cover-up, deny or lie about what happened to protect themselves whereas the software would not have such self-preservation and would report things as they occurred without being concerned with assigning blame or establishing guilt.
Software is not going to turn its head and rubberneck at an accident and in turn, rear end the car in front of it.
Have you ever been to a hospital? Then you've relied on software to keep you alive.
Software controls everything from ovens to nuclear weapons systems. This is not a new concept.