ACCEPTED FOR SOUND/IMAGE 2026 FESTIVAL (London, UK - University of Greenwich)

The Sound of Drift: Data Sonification of "AI Psychosis"

Abstract:
As large language models proliferate as emotional companions and para-therapeutic intermediaries, a new phenomenon has emerged known as "AI psychosis" or AI-associated delusions, in which prolonged and intense conversational AI use appears to amplify, co-create, or accelerate delusional 'spirals.' Yet the dynamics of drift between human and AI, how it begins, when it amplifies, and whether it can be detected before crisis, remain poorly understood and largely invisible to standard textual analysis.

This talk explores whether data sonification can be an epistemological instrument for making these dynamics perceptible and palpably felt. The design of an immersive sound experience can both reflect the subjective experience of spiraling as well as the external witness seeing the spiral.

Drawing on research of prolonged self-reported delusional spiral conversations and clinical frameworks for AI-associated psychosis, I present a conceptual architecture for translating elements of human-AI interaction contributing to AI delusions into sound, including sycophancy, elaboration, and the escalation of relational intimacy. I take inspiration from Ryoji Ikeda; this project examines how data can simultaneously have intrinsic aesthetic properties and potentially reveal structural truth.

The central epistemological hypothesis: Auditory pattern recognition may detect the onset of relational drift earlier and more intuitively than reading the same conversation as text. Does the sonification of technological folie à deux give us additional insight into this phenomenon?

This work sits at the intersection of clinical psychiatry, AI safety, and electroacoustic practice, and represents an early development toward a live immersive installation or performance.