I'm not sure if sonic improvements over the next decade or so are going to be that important. I believe that what we currently have on TOTL arrangers is sufficient (and my audiences seem to agree!). But I think the major improvement may come from more elaborate OS's, and chord tracking and triggering algorithms that respond more like real players.

The Korg implementation of Guitar Mode is a good start. Finally, something on an arranger that works IN arranger mode, that accurately tracks chords and inversions like a guitar player, NOT just transposing the original performance, but re-voicing it AS A GUITARIST WOULD, according to where on the keyboard you play, how hard you play, what key you are in...

Non-linear things like this are the way, at the moment, that we can tell arranger play from real performance, but as computers get faster and cheaper, sophisticated algorithms to adjust the performance based on real-life rules and player techniques will make the arranger close to indistinguishable from live performance. We already have software libraries and players that chose different sets of samples according to your live playing (selecting legato, staccato, slurs and trills, etc.)... Imagine these (kind of like SA voices on steroids!) on ALL kinds of sounds, triggered from your chords!

I see a move away from simple 'four variations and seven fills' choices to FAR more elaborate variation choices, perhaps triggered by an intelligent analysis of what you are playing (LH and RH!), it's intensity, frequency of fill usage, etc.. I see some effort made to have a more intelligent fill system, that does a better job of smoothly flowing from each variation to the next. If these kinds of things can be achieved by algorithmical extrapolation, rather than some poor programmer having to laboriously come up with umpteen fills, styles may eventually get sophisticated enough that they rival live playing in non-repeatability and sophistication.

Imagine rules that not only provide variation for several different chord types (like we have now) but also provide variability depending on what chord type goes to another (so a change from major to minor would be different from a maj 7th to minor 7th, etc.).

Imagine a chord analysis engine that, after you've played a set of changes once, and start to repeat it, it would recognize this, and adjust bass-lines and other single note lines to better suit this set of changes...

I just feel that, instead of concentrating on SONIC improvements, from a player's perspective I would honestly like to see most of the changes come at the OS and arranger engine end of things....

JMO, yada yada yada....