← All student projects · Theoretical and critical projects

Who's leading? Measuring agency in human-AI performance data

Can information-theoretic measures detect who is leading in human-AI musical performance?

Level: Honours, Masters
Prerequisites: Strong mathematics or statistics (e.g., probability and information theory) and Python.

When a musician plays with an intelligent instrument, who is leading? Our qualitative work suggests that the answer shifts from moment to moment. In this project you’ll test whether quantitative measures such as transfer entropy or Granger causality can detect these shifts in logs of human and AI gestures from IMPSY performances. You’ll start with synthetic data, where the right answer is known, and then apply the measures to real performance data. Most existing IMPSY logs record only the human side, so you’ll record new sessions with AI prediction logging turned on (we’ll help), across IMPSY’s interaction modes. Modes where the human and AI play at the same time are the most interesting, because in call-and-response mode the turn-taking is fixed by design.

• One semester: a validated analysis toolkit, tested on synthetic call-and-response data and applied to one set of performance logs.
• Two semesters or Master: a comparison between quantitative measures and qualitative accounts of the same performances (your own reflections, or interviews with other performers, which need ethics approval), and a discussion of what each approach can and can’t capture.

Tags: #theory #information-theory #data-analysis #agency #IMPSY

How to apply

Contact Charles Martin with your CV, your unofficial transcript (if you are an ANU student), and a brief statement (200 words) explaining how you would approach this project. Before applying, read the Join page and our project expectations.