CCMSAI 2026

This article reflects on “Duo Squared” and “Mirage,” two improvisational works that pair human musicians with AI-generated versions of themselves. Both works use Somax2, an interactive music system that recombines analyzed slices of audio corpora in response to live performers. Their bespoke corpora are drawn from the H2H Music Improv Dataset, specifically recordings of the participating musicians improvising together. Consequently, the systems reproduce not only each performer’s characteristic sounds and gestures, but also musical responses originally created in dialogue with a particular partner. Through practice-based reflection, the authors examine whether musical AI should resemble human collaborators, generate unfamiliar behavior, or operate as a reflective tool through which improvisers can interrogate their own creative practices. Performance experiences reveal moments in which distinctions between human and AI partners become difficult to hear, complicating perceived boundaries between performer, corpus, and system. The article argues that the value of bespoke-corpus improvisation lies not solely in successful human imitation, but in its capacity to expose what musicians recognize, expect, and value in interaction. It concludes by identifying further questions concerning corpus temporality, creative advancement, and the social dimensions that may be lost when AI substitutes for human collaborators.