A World Without W*ork

[tulak]

I recently listened to a Planet Money podcast called "The Big Red Button."

The podcast touches on Air France Flight 447. What I found most disturbing is if it wasn't for human intervention, the plane wouldn't have crashed.

Unfortunately, humans don't like to give up control even if it's for our benefit. But technology pushes us in that direction regardless. Tying it back to work, there's not much difference. Apparently it's easier to make machines "smarter" than it is humans.

Here's a link to the podcast for those that are interested: http://www.npr.org/sections/money/20...big-red-button

[harley]

Quote:

Originally Posted by Amethyst This reminds me of something I read recently [where? Scratches head] concerning self-driving cars. Faced with the choice of an inevitable collision with a) a child running into the street or b) elderly couple driving on the other side of the road, how does the self-driving car know (as a real driver would) that it needs to endanger the old people and save the child, not the other way around? After all, it makes more sense (in terms of the safety of the car's occupant) to hit the softer target. Conversely, if a teenager were driving, and the person running into the street were 90 years old, how do we program the car to "know" what a person would do? No doubt, most people would choose to hit the oldster...

Not a problem. When faced with a unsolvable situation like that the car will simply blow up in place, saving both external people and stopping the car passengers from having to deal with a lifetime of guilt. Simple.

[tulak]

Quote:

Originally Posted by harley Not a problem. When faced with a unsolvable situation like that the car will simply blow up in place, saving both external people and stopping the car passengers from having to deal with a lifetime of guilt. Simple.

The problem is that you can't design a 100% fool proof system. This reminds me of countless work meetings where everyone argues about the edge cases when a new proposal is presented that is better than what's currently in place. I also think that's human nature.

The reality is self driving cars only need to better than humans driving cars. With the data that we currently have, that seems to be the trend.

Google has been publishing their self driving car stats and at last check, it's self driving cars at 0 accidents, humans 14. I suspect someday their car will fail - it's only a matter of time - but I doubt they'll catch up to humans.

[NW-Bound]

Google still has a human driver onboard when the computer "needs" help. We do not know how often the human has to step in.

About hardware failure, how long do your PCs or laptops last? And when they fail, what happens?

If the computer in your car fails, is it going to accelerate hard, or stomp on the brakes? Does it swerve hard left or right? Keep coasting straight at the same speed? Or is it going to be different and unpredictable for every failure? I guess eventually the car will stop when encountering sufficient obstacles.

I worked on a fault-tolerant autoland autopilot 35 years ago. Then as is now, the only way to handle failures is to have a triplex, sometimes quadruplex system, and use majority voting. This means 3 computers to fly the plane while only one is really needed. If one channel says the plane should go down, while 2 others say the plane should go up, the majority wins. The failure is logged, and when the aircraft lands it is not permitted to fly again until the problem is fixed.

This redundancy is carried throughout the plane. Even the cockpit switches and dials are redundant (2 or 3 built into 1), so that a bad button will not cause a wrong input without being detected. Actuators to move the plane control surfaces are triplex, so that two good ones will override the bad one. The control surfaces are often redundant themselves. Jetliners like the DC-10 have upper and lower rudders, inboard and outboard ailerons, left and right elevators, etc... Ditto for triplex hydraulic systems, electrical power sources, etc... Heck, we used to have 3 human pilots in the cockpit, and still have 2 today. Redundancy, redundancy...

Granted, car systems do not have to be so elaborate, but certain safety aspects are still needed and must be defined. In the wake of the Toyota infamous "unintended acceleration", the NHTSA has been criticized as not being as knowledgeable in regulating auto electronics, to the same level as the FAA in establishing safety standards for airplanes.