Cubdriver

That would be disturbing if true but casual surfing provides credible enough sources on record saying drones have about the same reliability as other developmental manned military aircraft did in their day.

(Raining drones? CNN Security Clearance - CNN.com Blogs)

These recent high-profile crashes of U.S. drones raise questions about the reliability of the crucial unmanned aircraft. Unmanned aerial vehicles, or UAVs, have about the same mishap rate as the F-16 manned fighter jet did at a comparable stage of development, according to retired Lt. Gen David Deptula, who ran the Air Force drone program until he left the service in 2010. An Air Force chart obtained by CNN confirms Deptula's assessment, showing as flight time increases, mishap rates for drones drop, just like they do for the piloted F-16 and F-22 fighter jets. When looking at the total number of flight hours, the mishap rates for "Predators (MQ-1s) are a bit higher," said Deptula. "Reapers (MQ-9s) and Global Hawks (RQ-4) are a bit lower."

No aircraft can, not a good question the way it is posed. Autonomous aircraft will be able to land in any winds human aircraft can for landing and TO. Ice detection is not a difficult problem.

Or can a human? Another vaguely-defined question. If computer #1 fails you get instant transfer to computer #2 just like when the human PF fails to act and the PNF takes over in human crews.

As much as any human flight crew can. Computer #1 craps out and #2 takes over and flies to the limit of its abilities.

No, and some oversight will probably always be in place by experts humans for this reason. We are not close to true artificial intelligence and humans will remain the best decision makers. However, looking at the crash history there are many examples of accidents that were made worse by human intervention and management, so assuming a human will fare better is not always true. It's the routine flight events including stock flight emergencies that can be replaced by computers and those that fail to fall into standard grouping will definitely faze the computer because they do not really think. The task is to make this scenario so rare that it is allowable. For example, one in 40 million hours, or something like that.

Probably not because I assume extraordinary decision making skill went into Sully's decision. However, if it was a logical thought process, and I assume it was, computers are supremely logical and usually fast enough to compute this sort of thing. The question is, does the source code provide for enough contingencies.

When I was a teenager I loved chess. My parents bought me a fairly expensive toy computer to play which knew a lot of strategies and tactics. It was very fast and it beat me without mercy for months until I learned all the standard tricks and chess plays, and it was daunting. I clearly recall one point when I started winning against it though, and it was because I could simply look ahead farther than the computer could looking for possible combinations. I had begun thinking like a chess player and had more computing power of my own. So if you want to make a good criticism of computers, it's that they really do not think, they compute, and they have limited power. The task is to reduce the amount of thinking they need to do to some absurdly small and accordingly safe level. There are no guarantees and no absolutes.

This one is a moving target kind of question. Like existing computers, drone computers will require regular updates to their security code in order to be safe. The question is cost, is keeping them safe worth the cost of the R&D required to find new threats and counter them. But as with any security issue, you cannot have 100% security, you can only hope to stay a bit ahead of the threat. We do this with humans too, before we use a human pilot we make sure they pass an FBI check to get the job, and they must go through scanners and so on each time they fly. The system is good but not perfect. All it has to be is slightly better than the threat it is designed to prevent.