I wonder how hard it would be to train an AI to catch differences from a reference model? Because human eyes don't seem to be sufficient.
Garbage in, garbage out. The AI would still need reference points to make up for the bad practice, which would have to be designed in as well. Without a 'red dot matching red dot and a green dot matching green dot' on identical connectors, it would be hard for anyone to know the difference, much less an AI, as apparently happened here. Why not simply spec a part that's easily mistake proof?
The real issue here is that NOWHERE in aerospace in the last 50 years(or rail, or ships, or general industry, for that matter(I've worked in all of them)) does anyone use
identical connectors that aren't keyed differently. That's a lesson that humans learned a long, long time ago.......except in this case, they chose to re-learn it.
There is an entire aerospace industry devoted to nothing other than connectors, pins, wiring harnesses, and the design, installation, and repair of such. There are literally tens of thousands of externally identical connector shells that can be ordered and produced as different master and sub key slots, male and female, on top of differing pin configuration layouts.....ALL as catalog, Commercial Off The Shelf units, already built, bagged, tagged, and ready to ship.
The critical design review process is specifically set up to catch such flaws as interchangeable duplicates as has been reported happened here. Changes after design review have to be vetted in the same manner as the original design review.
One of the key features that aerospace(and similar industries) design uses in through the Bill of Materials (BOM) engineering audit
before anything is ordered or built. It's an actual engineering/logistics job that is usually staffed by a team of SMEs in each part of the BOM (materials, wiring, fasteners, etc).
The final, as proposed BOM goes through a step in critical design review where LIKE ITEMS flagged in the BOM are flagged for specific attention and engineering design review to make sure that they're NOT physically co-located unless engineering controls and design dictate so. THEN, a failure mitigation designator is usually put in place, such as a RED square being placed on a bulkhead connector, and the corresponding wiring harness gets a RED nylon overwrap AND a quality assurance inspection step added to ensure proper installation and compliance with controls.
The final step to catch it is on the item itself where assembly engineers look over each part as Bill and Ted are putting it all together to catch these potential failures, and call an "ALL STOP" until mistake proofing measures are put in place and complied with.
Absolute last chance is when the item goes to testing. Again, engineering controls are usually put in place, with multiple QA sign offs, to ensure that
everyone agrees on the definition of "Up, down, left, right"...usually because it's ALSO spec'd with a mistake proofing step as "Triangle/Square/Circle/Star" or some such manner that can't be confused because you're operating it facing North and I'm inspecting it looking East and we both fail to communicate.
Again, that's nothing new, and even tiny little local electrical firms practice such controls.
It's been said that most humans only learn things 4 ways:
1.Read it
2.Write it
3.Repeat it
4.Traumatic experience
and that some humans add a 5th way:
5.Repeated traumatic experience due to a failure of reading and/or writing