Embodied AI and the future of accessible music-making: a review of developments in AI-ADMIs post-Frid

This paper examines the evolution of Accessible Digital Musical Instruments (ADMIs) since 2019, a period marked by the increasing integration of Artificial Intelligence (AI). We refer to these emerging instruments as AI-ADMIs. Emma Frid's 2019 systematic review [1] identified key design priorities for ADMIs, including adaptability, user-specific customisation, multisensory feedback, and participatory design. However, Frid's study notably lacked discussion of AI or machine learning, reflecting the limited adoption of these technologies at the time.
Since then, AI-ADMIs have rapidly expanded, incorporating camera-based gesture recognition, interactive machine learning, and adaptive deep learning systems. This paper provides an overview of these post-2019 innovations and critically assesses how they build upon or diverge from Frid’s foundational concerns. Drawing on the primary author’s doctoral research into an AI-driven, gesture-controlled instrument, we explore current challenges such as algorithmic bias, normative assumptions in datasets, and the absence of user-specific training data.
In response, we propose an Embodied AI Design Principles framework to guide the creation of inclusive, disability-led AI musical interfaces. We argue that an embodied approach, one that foregrounds the lived experiences of disabled musicians and the physicality of musical interaction, is essential for developing equitable and accessible AI-ADMIs. We conclude by reflecting on future directions for the field.