The significant aspect of this demonstration was that it used vision-based technology; equivalent to a piloted visual takeoff.
Future developments might lead to an autonomous aircraft, which together with machine learning could provide a satisfactory model of 'manual flight'.
It is unlikely that a machine version of pilot would replace the pilot in the short term, but it could be a model for acceptable performance and for alerting deviation - similar to EGPWS being an 'error detector'.
A takeoff module could improve weather dependent FD commands or intervene with autoflight assistance, e.g. auto rudder / roll correction quicker than piloted response, like existing rudder input with engine failure
Further development for all flight phases could be the basis of a 'pilot assistant' for single pilot operations, and a solution for single pilot incapacitation.
As per
mjohansen, AI is far from intelligent and learning normally requires baseline data so that comparisons can be made and performance improved. With human piloted aircraft, a challenge is to judge which data is representatively good or that not so; technology has to learn the boundaries of good piloting performance before it can alert or help.
A more immediate problem is that we all 'think' that we are good pilots; so in exceptional or extreme conditions when the machine suggests assistance how might we react - 'shut up gringo'; or a polite, CRM 'thank-you'.
P.S. '
A human-centric approach to AI in aviation'.
https://www.easa.europa.eu/sites/def...admap-v1.0.pdf