Originally Posted by
lucille
Unmanned flights of increasing complexity are being demonstrated every day. Helo flying, the last bastion has been breached!
https://newatlas.com/military/black-...-without-crew/
Garmin have an emergency autonomous autoland function for certain GA aircraft. It requires no pre-programming of destination and runway choice. Press the button and next thing you know you’re sitting at the end of a runway at an airport the system has chosen, with park brake on and engine shut down.
https://www.garmin.com/en-US/blog/av...nd-passengers/
There is probably no technical reason today why any aircraft can’t be retrofitted with equipment which would allow unmanned operation from gate to gate. Engine start, to shutdown.
As for emergencies? For the most part we are robots… Fly the aircraft, identify the problem, call for the checklist..etc.
Agreed there has been the occasional outlier which required inspirational problem solving by the crew outside the remit of the checklist. You can be sure researchers, designers and engineers are working feverishly to eliminate the likelihood of such unimagined combinations of failures.
I think that once 100% reliable, redundant and robust two way data comms can be guaranteed, the era of manned airliners will come to an end. I wonder how close they are to achieving this.
Your story describes the ideal world, correct for some 95% of the time.
Unfortunately, the flying public, nor the electoral public, will accept a 95% perfectness in real life (and the rest more or less ending up in disasters of varying seriousness).
Ever seen robots, spinning around in vain, when outside their pre-programmed capabilities or having a defect ?
AI is NOT human ingenuity. Feed an AI engine a lot of horse pictures, to recognize in the future horses on pictures and find out later on, the AI network "learned" to recognize the logo on the training pictures. AI (nowadays) is no more than a super-multidimensional recognition of acceptable solutions (and in practice fed with additional sensory input to validate/control the AI output/feedback), used to operate in the same super-multidimensional environment as its training. Change something in the environment, and suddenly the Teslas start stopping for the moon. If the electronics of a Tesla fails, it strikes, and the Tesla is frozen. Or road recognition algorithms/AI completely choke on just a piece of white paper behind a window opposite the long leg at a T-crossing.
When things go haywire (stuff fails, though unknown when), human ingenuity (based on interpretation of significantly more "training" results than can be fed into AI) is required, to solve the situation in an acceptable way. Even, as
Pilot DAR writes, in situations considered "normal", it turns out the learning/algorithms aren't covering the complete super-dimensional circumstances to be considered for directing towards a safe outcome. We should not forget, when humans are "training", they embed (!) those training items in their already existing knowledge/capabilities outside the currently offered training items, not something AI-networks do on training.
Reliable comms will never exist (ever heard of solar flares?). Reliable GPS will never exist (think about GPS jammers, used in wartime or just for fun). The more vulnerabilities are build into society, the easier things can go wrong in unforeseen circumstances. Remember the Tesla that went on an extremely wild road-warrior crusade, just recently in China, in the end, killing 2 people. Is this really "ready in 2 years" for full autonomous driving ?
And, for the one-person cockpit occupation: When things go haywire, 2 (highly trained) people are needed to keep situational awareness and resolve the issue into a non-issue.
Back to the ideal Instagram world......