Participants 2022

We decided to present our idea in the form of a pop song with a clear storyline, which would reflect both the socio-cultural as well as the creative opposition of human and machine. First, we defined the structure as a dialogue between the two players (a human and an AI), represented in alternating verses, and wrote our lyrics accordingly. To achieve some musical separation between these parts, we used search algorithms to find complementary musical keys for the parts of the two opponents and the shared choruses. Then, formal grammars were applied to generate chord progressions and melodies for the two opponents, based on the chosen keys and the lyrics’ rhythmic structure, constrained by our vision. In contrast to that, a multitude of other open-source AI tools developed for pop music generation helped us co-compose further elements for the final musical score through a back-and-forth exchange of ideas. Among others, we chose Magenta.js for keys, drums and general instrumentation inspiration, Groove2Groove for guitars and accompaniment, and a pre-trained T5 model for additional lyric ideas. Parts of the song were later modified depending on which player they represent. “Human” parts had more of our intervention, whereas we gave more freedom to the AI generation for the “AI” parts. We made the final revisions of the score according to our musical intuition and the session musicians helped us to put everything together with their expressive performance.