I think the airplane load of passengers at the mercy of an autonomous computer is still some distance in the future.
A very long way...
And not because of the lack of technical capabilities (let's face it, already most large commercial aircraft out there, if properly set up, could be landed by anybody with the simple instruction 'don't touch it'...), but because the balance of risks and associated expenses.
I'm not a pilot but one of those usually referred to as engaged in doing arithmetic involving the seeds of a certain legume species. In the past I have been involved in designing and operating a very complex automated data extraction systems. One of the key realisations during the project was that you can design a good robust software that handles 99% of the envisioned scenarios (lets not speak of the un-envisioned ones), but to automate the remaining 1% will take 9x the effort so far (directly translating into cost), and the result will still have an unacceptable error rate (which was defined as less than one error for every 1,000 processing events). The solution was to design a system with a human operator to handle and decide upon exceptions, with automation aiding to spot errors and correct manually. This achieved an error rate better than 1 in 10k events.
The scenario in aviation is similar. We have reliable automation to handle 99% of scenarios, and the pilots are there to cope with the remaining 1% and monitor what the automation is doing and take control if needed. That is not going to change, simply because the potential error rate (and the associated costs) with a remote pilot or full automation is too high to be acceptable. Military drones is another story - there the increased payload and reduced complexity for 99% of missions provides more savings than the cost of losing 1% (or less) of expendable airframes.