Very nice to see this expressed, thank you! This is a trap so many folks fall into when starting out with generative work IMHO. There's this kind of natural instinct to anthropomorphize the data -- where the assumption is that some platonic representation exists that can be expressed with "the right choices" or some neutral choice -- while little attention is paid to the impact of the choices themselves on the representation.
FWIW Andrea Polli nicely shines a light on this during her Sonic Antarctica album (on the track "A Model is a Cartoon") where she superimposes her own sonifications & field recordings with an interview exploring the limitations of modeling data in a scientific context... https://www.gruenrekorder.de/?page_id=342