I know we can't compare current generation automation with next gen, but even the other day while in a managed descent, the aircraft suddenly thought it was 9999' high on profile and increased rate of decent to 6000fpm to try re capture said profile, speed rocketing towards VMO.
A non event for two human pilots looking at it and rolling our eyes, as we simply pulled for open descent and the aircraft came to its senses and eased off, but it really doesn't inspire me with confidence that whatever autonomous aircraft they come up with won't have some similar issue and will be able to correct itself, if even the current generation aircraft which have been flying for 35 years haven't resolved simple computer brain-farts like this despite multiple avionic upgrades since its inception.
Not to mention that flight path control should be the simplest part of the flight to automate, it's the more nuanced issues, legalities, failures etc that should pose the complex automation task. Things like disruptive passengers, load sheet errors, birds ahead, fuel decisions on marginal weather days, assessing whether to commit or divert, airport's closed; do you hold or divert. Flock of birds hits the "artificial vision system", or AOA probes, ASI etc, how does the autonomous aircraft now cope?
I haven't done many auto-lands in my career so far but the ones I have done, particularly when there's any sort of wind element are always quite inconsistent. How will an advanced autopilot cope on days when it's gusting 60 knots? I've had the autopilot trip out on my numerous times over the years on stormy nights when it just says I'm done, over to you. And this is an aircraft again, that has been around for 35+ years and had numerous avionic updates since its inception.
CPDLC even now seems to have delays of 1-2 minutes sometimes to send messages, so these aircraft can't be remotely controlled if you have to factor in a 2 minute lag, they will have to be totally autonomous.
Far more problems to solve than can be mitigated just by paying two guys to sit up front and adapt to situations as they arise.
I'm not a luddite, in fact I've a background in software engineering before flying, but AI will really have to come on a lot more before it can cost effectively replace the flexibility provided by having two humans operating the aircraft, sat up front making decisions on the fly, if you'll pardon the pun.
I've no doubt we'll see aircraft in the next 15-20 years that could get from gate to gate with no intervention required from the humans on boards, the basics of managing the aircrafts flight path and taxi-route is not rocket science and in fact should be easily automatable. However, I can't see how this can consistently be done on every flight, we all know there are days where the thing runs on rails, these days could easily be automated. Then there's days where we're managing passenger issues, spotting dangerous goods irregularities, doing maintenance resets at outstations, changing taxi-routes on the fly, taxiing around potholes and other aircraft that have gotten stuck, picking fuel on marginal weather days off runways with marginal performance and then managing the options enroute, this will all still have to be done by humans.
So do I think we'll see an aircraft that can go from gate to gate with little or no intervention, absolutely. Do I think it will be possible to do this everyday, not a chance. And if you're going to need someone there 2-3 days a week, and if you can't tell in advance if today will be the day that HAL needs to have the CB reset, then invariably you'll need a crew on board every flight.