I've actually tried all of the approaches that you've mentioned over the years, and - for my needs - they're not that compelling at the end of the day.
Sure it might be cool to use cellular automata to generate rhythms, or pick notes from a diatonic scale, or modulate signals, but without a rhyme or reason or _very_ tight constraints the music - more often than not - ends up feeling unfocused and meandering.
These methods may be able to generate a bar or two of compelling material, but it's hard to write long musical "sentences" or "paragraphs" that have an arc and intention to them. Or where the individual voices are complementing and supporting one another as they drive towards a common effect.
A great deal of compelling music comes from riding the tightrope between repetition and surprising deviations from that scheme. This quality is (for now) very hard to formalize with rules or algorithms. It's a largely intuitive process and is a big part of being a compelling writer.
I think the most effective music comes from the composer having a clear idea of where they are going musically and then using the tools to supplement that vision. Not allowing them to generate and steer for you.
-----
As an aside, I watch a lot of Youtube tutorials in which electronic music producers create elaborate modulation sources or Max patches that generate rhythms and melodies for them. A recurring theme in many of these videos is an approach of "let's throw everything at the wall, generate a lot of unfocused material, and then winnow it down and edit it into something cool!" This feels fundamentally backwards to me. I understand why it's exciting and cool when you're starting out, but I think the best music still comes from having a strong grasp of the musical fundamentals, a big imagination, and the technical ability to render it with your tools and instruments.
----
To your final point, I think the best example of this hybrid generative approach you're describing are Autechre. They're really out on the cutting edge and carving their own path. Their music is probably quite alienating because it largely forsakes melody and harmony. Instead it's all rhythm and timbre. I think they're a positive example of what generative music could be. They're controlling parameters on the macro level. They're not dictating every note. Instead they appear to be wrangling and modulating probabilities in a very active way. It's exciting stuff.
When you learn to use it you can throw a lot of intention into it, knowing the output even before you hit play. Yes, you can go the other way and "subtract" your way out of a chaos, but you can also intentionally piece together the components and produce an output you imagined beforehand. The missing pieces here for this format, my instinct tells me, are layers of abstraction or additional UI elements that will help in composing a final piece, using code for the fundamental components plus something else that hasn't been invented yet or noone has thought of glueing it together.