This paper investigates how jazz improvisers engage with artificial intelligence (AI) to reorganize and internalize new musical vocabularies. Grounded in Adorno’s concept of estrangement and Fumi Okiji’s reading of jazz as critique, the project treats improvisation as a method of alienating the familiar to reveal underlying structures. Using manually segmented phrases from constrained improvisations, we extract audio descriptors to train a set of simple autoencoders, projecting each phrase into a latent space. This space is explored through dynamic modulation, visual navigation, and descriptor-based retrieval from a corpus of real recordings. Instead of generating new audio, the system selects and recombines existing materials, preserving expressive detail while encouraging disorientation. These reorganized outputs are internalized and reintroduced into improvisation, generating subconscious 'slippages' that reflect embodied learning. Rather than replacing creativity, the system reframes it, mirroring jazz's historical absorption of external influences. The result is a recursive feedback loop between performer and system, where algorithmic mediation becomes a tool for critique, and improvisation retains its power to estrange, subvert, and expose.