Does Software Understand Complexity? (2018)
michaelfeathers.silvrback.com
michaelfeathers.silvrback.com
Modern programming as we know it is always a case of a range of given 'knowns' such as input data, program algorithms and programs' application of iterative mathematics and basic logical operators (if-then-else type techniques etc.) to produce an output/the end result. Of itself, no programming language has an a priori concept of 'complexity' built in.
If we move on from traditional software to AI we may be able to tackle the complexity issue to some degree but that's not to say that we understand how AI arrives at a solution. I see this as a deeply unsatisfactory outcome and thus it is only part of the solution. It's why many others and I believe that the term Software Engineering is a misnomer in that the software/programming profession has yet to come of age. Those arrogant enough to say that it already has I believe are just fooling themselves.
I put this case to you as a simple example: take the software that ran the 737 Max and crashed those planes. Let's assume Boeing's engineers knew what they were doing (and that's open to question) and they had essentially programmed the MCAS system correctly but with some bugs. Then I would contend that any system that was capable of adequately handling complexity—here processing parameters outside the range of program's normal design parameters—would have automatically defaulted to a Do what is necessary to save the plane fallback algorithm.
As such tragic cases demonstrate, no such software exists—as software has just not reached the stage of development where users can be certain that their software does not let them down when unforeseen problems arise.
It seems to me software innovation of this type is still years—likely decades—away.