It's all pretty much a moot point, at least for the near future. The FAA is on record as stating that they will not certify any airborne software that incorporates AI (and I believe EASA has said something similar).
The issue is pretty basic - all the testing done to certify airborne software requires that the outputs for a given input are always the same - and AI doesn't always do that.
Now, at some future point, I expect that AI will progress to the point where the regulatory authorities will need to re-access that position - and it will become apparent that the advantages of AI outweigh the disadvantages.
But I don't see that happening for a long time - probably decades.