Generative modular music is a very valid concept and what it really defines is the process of setting up a modular system in such a way as to generate ideas that continually evolve and change over time. Sometimes these things can just play themselves and evolve over hours without any interference but you can also steer what happens too by making adjustments to various things along the way.
(I used to bring in non musical people such as photographers and ask them to randomly twist things. All I told them to do was only make fine or very subtle changes) Back in the 80's and 90's I had quite a large Korg modular setup myself and got into this very thing. It is quite interesting and fascinating.
Except where some people get this wrong is they can easily think that it is confined to timbral changes only but this is very wrong. A lot of people who set these things up including the guy in the video may have a lack of understanding as to how to setup a pure melodic concept using this approach. I have done this where key centres and notes all play in relation to each other
(at random as well) and it is possible to setup parameters in such a way as to not so much of a timbral change taking effect but a melodic one. I have done both and while the timbral change concept is very cool
(and some of the sounds in that video were also cool) you can also do it purely melodically as well.
But it takes some serious knowledge of music and theory and harmony concepts to be able to do it. The problem with many people who fiddle with modular systems is that they lack this knowledge and hence most often the results are timbral rather than melodic.
I have setup quite extraordinary setups which play only melodically and that include bass parts and chords as well. And yes a great percentage of the results can still be rubbish too but there are always moments of glory in there too. The trick is to identify them and then develop those short lived gems further yourself.
Jan Hammer used to do this when he was scoring Miami Vice soundtracks. I think at one stage he was doing like 5 episodes a week and that is tough. He used to set up things like this and let them run overnight and then quickly scan the results the next day and pick out the good bits and then develop and refine.
There is more than one approach to it that is all I am saying. The musical way though is much harder.
What is cool though is this thread is interesting
(many Sonar users may not agree) and I thank the OP for mentioning it. The real challenge now though could be figuring out a way to do it all now in software. I might need that Arturia Moog Modular after all!