Negotiating Autonomy and Trust when Performing with an AI Musician

We report a practice-based exploration of developing and performing with an autonomous AI musician called LOERIC that improvises folk tunes in response to a human partner. We follow an artist-led methodology involving a subjective, first-person reflection on our own practice of iteratively developing LOERIC through a series of public performances. A key feature of LOERIC is the manner in which autonomy is negotiated. Prior to a performance, surrender and looseness of control are preconfigured as playing styles. During a performance, the human and LOERIC interact solely through the single parameter of musical intensity to enable the human musician to loose awareness of explicit control as part of a flowful musical experience. This separation also helps the human establish a baseline of trust in how the system might potentially behave beforehand, within which they can safely experiment and improvise when playing live.