I began with short improvisations recorded with a MIDI keyboard connected to my computer. For instance, I’d improvise arpeggiated chords that could work well as an intro or rhythmic patterns that could be worked into a snare drum part. Using a machine learning toolkit in Max, a visual programming language for music, I used these recordings to train a model that predicts the next most probable musical event in a sequence based on the current situation. This model was then let loose to create new recordings that followed the probabilities in my recordings. I sifted through these generated recordings to create the harmonic, melodic and rhythmic framework of the work.  The lyrics were generated using ChatGPT-4 with a message to write 50 poems based on parallels between pharmacotherapy for mental illness and AI-generated content in regards to feeling like you no longer own, control or have access to your true emotions. Exceptionally expressive as well as intriguingly ambiguous results were pieced together and occasionally modified to form the lyrics.   These lyrics, as well as the music generated in Max, was used to map the vocal parts using the VOCALOID:AI vocal synthesizer, which learns and models the characteristics of a real singer’s voice using deep learning. These components were then all brought together in Logic, a Digital Audio Workstation, and built upon to finalize the result.  AI tools used for lyrics and music  GPT-4 Machine Learning Toolkit for Max by Benjamin Day Smith MIDI-Trained Markov Chain patch for Max by Samuel-Pearce Davies VOCALOID:AI (Free Trial)