Originally Posted by
ams6110
And yet Tesla, an automotive company which presumably follows this process, still has an "auto pilot" software function that on more than one occasion drove a car into a stationary object at 70mph.
That is a regulatory failure (or the regulatory process lagging far behind the technology), just like the MCAS issue, neither MCAS or the Tesla autopilot should, IMHO, have the safety rating they do (ander any sane safety analysis) . . . Secondly, the automotive safety standard is nowhere near capable of dealing with this kind of technology . . . it's from the nuts and bolts save lives era . . .
This is not a software problem, the software, certainly in the case of the Boeing product, could have been specified to be compliant with a higher safety case under well established processes, but it wasn't, and so that wasn't the way it was created, that decision would have been made months or years before a single line of code was written. The code monkey writing the code has no input, whatsoever, to that process, single AOA, Dual voted AOA, Triple AOA, inverted AOA on one side, monkey holding the AOA while standing on a unicycle, it's all decided and specified at the system level, miles above the tedium of people writing code.
Tesla (and the automotive industry) actually have a more difficult problem, their autopilot software uses deep learning AI which by its very nature produces indeterministic outputs, so the standard safety approaches and mitigations do not work (indeterminism in a safety critical system is not really allowed), this is another regulatory failure and the safety industry is struggling to understand how to approve these kinds of systems . . . I would not let one drive me.
Fd