Identity in the Loop Cognitive Ergonomics of the Mediated Voice and the Future of Service Work

Abstract

Real-time AI systems that modify a worker’s vocal, physical, or behavioural presentation during live interactions are entering service-work infrastructure at scale. Existing cognitive-ergonomic frames — automation, augmentation, human-in-the-loop supervisory control — handle this class of intervention poorly because the AI acts neither on the worker’s task nor on the informational resources used to perform the task, but on features of the worker’s embodied output that carry identity information before the output reaches the receiver. This paper makes three contributions. First, it defines identity-layer AI as a distinct locus of human–AI intervention characterised by this intervention point. Second, it introduces a three-party mental-model asymmetry framework in which worker, deploying enterprise, and receiving customer hold partial and non-overlapping role-relative models of what is being modified, by whom, and toward what target profile; the framework is formalised in terms of role-adequate mental-model alignment, and the design and governance corollaries (the Identity Modification Transparency Protocol) are derived from it. The framework speaks directly to the trust, transparency, and deception concerns that real-time voice modification raises. Third, it translates the framework into an empirical research agenda specifying four cognitive-ergonomic study designs that measure disclosure salience, agent workload, perception gap between produced and delivered voice, and longitudinal occupational-identity effects. AI-powered accent modification in global contact centres serves as the leading deployment case; publicly available worker-experience reporting provides a preliminary — and explicitly stratified — evidence base. The paper concludes with implications for the future of service work: the dissolution of embodied locality, longitudinal risk of occupational identity drift, and the measurement challenges that arise when AI modifies rather than replaces human output.