Akarsh O. 2023 | BASIS Independent Silicon Valley
- Project Title: Procedurally-Generated Heavy Metal: A Fun Tool or a Copyright Disaster?
- Basis Independent Advisor: Jon Noble
With the creation of Midjourney, diffusion models, and more, AI-generated art has become one of the largest technological movements. However, it often utilizes digital and physical art posted online without the artists’ explicit permission. Critics raise the following question: “Is AI-generated art art?” My paper aims to create a generation model that can recreate facets of music (key, chord progression, style, rhythm, and melody) from famous musical artists. Through this process and copyright analysis for the US and EU, I will determine whether these types of algorithms highlight an ethical problem within AI generation of art. Researching methods on AI-assisted music generation, most of them that use waveform generation have subpar quality, so I will be using a method utilizing MIDI data of 10 songs from 10 different artists of different genres. Using a model pre-trained with MIDI data from a dataset of MIDI transcriptions, the selected songs can be input and a symbolic modeling transformer can form event predictions that include the melody, chords, etc. After my data is generated, the MIDI music will be matched to the sound choice of the artist. To gauge a proper subjective impression of the generation, I will interview 50-60 students from BASIS Independent Silicon Valley to compare the generated songs and the artists’ songs as a single-blind experiment. To gauge whether this is a threat to copyright and ethics, I will also be analyzing the Fair Use Act of the DMCA for the U.S. and an equivalent clause from the EU.