Maziar Kanani

Title: Executable Design: AI Tools for Music

Supervision Team: James McDermott, UoG / Seán O’Leary, TU Dublin

Description: This project is about new AI-enabled creative tools for musicians and new creative ways for music consumers to interact with music. Modern AI methods can generate plausible-sounding music, at least in some cases. However, AI has not achieved anything like “understanding” of the internal structures, relationships and patterns which are deliberately designed by human composers. This understanding can best be represented by short programs, since (by the Church-Turing thesis) no other representation can be more powerful. Short programs are those which best capture all possible regularity and structure. We will use modern metaheuristic search methods such as genetic programming, and neural program synthesis, to automatically create short programs which, when executed, output given pieces of pre-existing music. Simple manipulations of these programs will give natural variations and extensions of the music, enabling exciting new tools for creativity.