ABOUT THE TEAM

We are a group of members at Universidad Privada Boliviana in La Paz, Bolivia currently studying AI under the supervision of Gissel Velarde. We are curious about where creativity stems from and how it can be mirrored in machines. The aim of this project was to reflect, in a cooperative nature, how machines and humans can work together in the future to create new and exciting things. Some of us practice music in our free time, so this project was a perfect match for our interests and goals, both academic and recreational ones. 

ABOUT THE SONG

Scaling Pilot is a song produced on Reaper Studio using MIDI outputs generated on different executions of our Genetic Scale algorithm. It is worth mentioning that we tried to produce pop rhythms (more commercial today), using joyful modes that can be selected as generation parameters. We have six modes to generate, from which we selected the happier ones: jonic, doric and lydian, in tempos between 100 to 150 beats per minute. About the progressions, there’s a preference for eleven notes since these all compose the selected modes, being better scored by the algorithm. Therefore, the concurrency of the F note might be limited on the final result. 

We produced a multi-instrument reverb track, with the drums signaling changes in the patterns, making it highly enjoyable, even without a leading voice. In fact, it is ideal for setting scenarios that go with a feeling of peace and joy. 

ABOUT THE HUMAN-AI CO-CREATION PROCESS

Even though this was our first collaborative interaction with a genetic algorithm, we knew the "how" we wanted to do it long before. From the beginning, we wanted our generated MIDI files to sound good. For us, the optimal way to do it was to generate music from scratch. We aimed at parameterizing the algorithm using modes in a tempo range that we considered suitable to convey positive emotions. 

Our algorithm does not take a long time to compose. It can generate 30,000 tunes in less than twenty minutes on Google Colab. It returns the top five results in the mood it was set to compose. 

We ran the algorithm several times until we obtained interesting results. We selected the most joyful excerpts to produce the final track. We used Reaper Studio and assigned different virtual instruments to each generated excerpt, for example synth and bass sounds. We used a pop beat with reverb settings. 

It was actually a simple process, requiring no deep knowledge of music theory, just creativity. Still, we recognize that some of us are semi-professional music producers, and therefore our ability to arrange and produce the final track together with our 10 selected algorithm-generated excerpts, explain that our song sounds pleasant to us.