I much prefer uniqueness and character as a vocal-synth trait to realism. Some AI vocals are quite characteristic and have a unique sound (e.g. KAFU and Chis-A), but many of them aim so hard for realism that they lose some distinctiveness. Miku is (likely intentionally) pretty unrealistic for the most part, but I love using her and I love her unique sound. Also, has anyone else noticed that AI vocals often sound less "full" than the more standard ones, maybe a bit more thin/tinny?
I'm also not as hard-wired toward power vocals as much of the fanbase. Shouty vocals sound awesome, it's true, but while everyone seems to be reaching for whoever they can make scream the loudest, I've been using Miku. (Making Miku Solid shout is fun though.) My point is, soft and "neutral" (e.g. Miku Original, Kaito Straight) vocals are just as appealing, and many singers will have both powerful and soft parts in their songs. (Unless you have V5 to effectively make her sing softly, making Flower sing a soft piano ballad probably isn't a good idea.)
Androgynous vocals are cool and all, but very feminine and masculine vocals are too! I'd really like to see another Big Al-type vocal someday...Youthful-but-still-masculine vocals like Yohioloid and Kevin are cool too, as are deep-but-still-feminine ones like Eleanor Forte and AVANNA. Sometimes I kinda like knowing the gender of the singer I'm listening to...
If they do make a Miku AI, since it likely wouldn't sound Miku-like enough with the straightforward path, I wouldn't mind if they took the approach that the Quadimension "Plus" vocals do with training her on tuning from her regular voicebank. I think that'd be the best option for her, because her entire thing is her unique and characteristic sound and without that she might just be "Saki-loid".