Trying to use ML to get rid of programmers will just replace them with ML experts who also have to be programmers to implement the models and munge all the data. These people will in turn have to be paid more than the original programmers were.
The goal of ML isn't so much to get rid of programmers as to get rid of specialized programmers. Right now if you want to solve a problem, you need someone with a deep understanding of that specific problem to develop a solution. For a complex problem like diagnosing cancer patients, you are talking about a team of people with decades if not centuries of combined experience in oncology on top of the actual programming expertise to implement the tool. The holy grail of ML is to reach the point where someone who is an expert on making ML systems can apply the same (or substantially similar) tools and expertise to a wide range of problems - the same team that makes a cancer diagnosis system could also make a legal text search system or a protein folding system. Realistically we'll probably never get to the point where zero domain knowledge is required, but even if the requirement is just substantially reduced to the point where the ML expert can learn what they need in months instead of years, that would be revolutionary.
Yeah I think a very few but flashy situations, ML is worse results but even fewer programs, so worth it in some sense. The idea the non-programmers should be doing a little something is also very good. Just too the embodiment of that is the mess that is non-programmer Python.
