Let's take an example: CFIT, 1995, American Airlines 965, Colombia. That was a "perfectly flyable airframe" - although stated by this SLF/attorney that's not incorrect or controversial. So, the "argument" is that today's algorithm techniques and related processes could have conducted that flight operation without a CFIT accident with many fatalities. But not at that time - the capacity assumed by the "argument" didn't exist yet in 1995. Logically the "argument" addresses most (maybe nearly all) of the accidents involving perfectly flyable airframes via a counter-factual.
What about today, when the capacity is said by some to be at the brink of adequacy? Let's assume it is ready. But what about the other side of the equation advanced by the "argument"? Is it not correct that aviator training, safety management systems, and a variety of related processes have reduced the likelihood of a similar CFIT event to very low levels? Okay, so in some regions or countries, this level of fidelity to training requirements and methods may be less, or woefully less. What makes it true that in a place where a flight crew conducts a flight operation such as the one that resulted in the recent accident in Karachi, that the care and feeding of an autonomous/algorithmic system won't also be inadequate to the task? So the comparison advanced by the "argument" isn't actually valid.
Not least, there certainly still are lots of arguments (no sarcastic quotes this time) about whether the performance and reliability of autonomous/algorithmic systems have or have not reached a sufficient level, in the judgment of professional people with the requisite high level of knowledge and authoritativeness. Or, the jury is still out -- in fact, the court has not even given the jury its instructions yet. (The recent issuance of a position paper by IFALPA on these issues says a lot more than an SLF/atty should try to articulate . . . a thread started on it was moved over to Ground & Other Ops (Safety, CRM etc.)).
It's not just electrical failures. I'm waiting for the algorithmic whiz kids to provide an algorithm that would fly American 191 out of Chicago O'Hare on 25 May 1979, safely. Oh, that's a lousy example, because the faulty engineering of the hydraulic lines, and the maintenance issue with the engine removal and replacement, were unknowns at the time of the flight operation? Right then, and we can therefore be assured that the algorithms and their related processes won't have any errors in their derivation and iteration, or in the upkeep of the overall autonomous system either, I see?