[snom380]

Aircraft accidents are extremely rare in the first place, so that's a given. The question should rather be whether computers would do a better job than pilots in the same situation, and currently the answer is no. The actual flying of the airplane which autopilots do today is just a small part of the pilots job, and automating the rest of the tasks a pilot performs is non-trivial.

[lisper]

Are you a pilot? Because I am, and I'm telling you from firsthand experience that you're wrong. Except for takeoff and landing, there's next to nothing I have to do. And the only reason I have to do the landing is because my plane is a small GA aircraft without autothrottle or autoland technology.

[snom380]

Yes, I am a pilot, albeit not an instrument rated one. Are you telling me you don't do anything? Do you bring a book to read instead of monitoring the instruments? You don't talk to ATC, you never have to make a decision regarding a route deviation? You never take into account weather information and make a decision underway on whether to press on or to find another place to land?

[lisper]

No, of course I'm not saying that. I'm saying all those decisions could be automated, not that they have been. Except for ATC communications, all the information I use to make in-flight decisions is already available in digital form. All the engine parameters are digital. I get en-route METARS via XM. I have an WAAS GPS coupled to the autopilot. The only thing standing in the way of making my aircraft completely autonomous is a throttle actuator and the right software. And no, writing that software would not be trivial, but neither would it be impossible.

[snom380]

For normal operations, I think it's possible to make it _mostly_ autonomous today (TTS and voice recognition for ATC, and some heuristic metar/weather radar analysis might work). I don't think we have the technology today to make such a system safe enough to not have a human ready as a backup. And for an accident scenario, I think it's completely impossible today since we would need to integrate audio, video and smell sensors and AI software to rival humans in situational awareness. This would mean exceptionally complex software.

The way we have solved reliability in autopilots and FBW systems today is to make them as simple as possible and to give them sensible fallback modes (like the Airbus FBW removing stall protection when missing certain inputs), and even then we have had real life accidents because of programming errors or design errors. So if you think pilot automation is mainly a question of politics as you said earlier, I think you are being overly optimistic (which, of couse, is not uncommon in the software industry).

I think a better approach to removing pilots is to see the strengths in both computers and humans and design systems where the advantages of both are maximized.

If you are interested in reading more on software safety, http://sunnyday.mit.edu/ is a good starting point.