Naoto Fujie

Artificial intelligence is usually analysed through the problem of labour displacement: which occupations will be automated, augmented, or reorganised by computational systems. This article reverses that perspective. It asks how artificial capability is produced through forms of human activity whose labour status, authorship, temporal burden, and political claims are systematically weakened or erased. Existing research has documented free digital labour, microwork, ghost work, content moderation, data annotation, expert evaluation, and global AI supply chains. Yet these literatures have not fully theorised the institutional process by which persons who are not legally or socially recognised as workers become productive contributors to AI capability while remaining excluded from labour rights, attribution, collective bargaining, and control over technological direction. I call this process ghost-workerization. The article distinguishes three layers of human contribution to AI: contracted ghost data labour, platform-embedded unpaid data labour, and ambient data extraction. It does not treat every data trace as labour. Instead, it proposes six conditions under which data-producing activity acquires a labour character: institutional capture, productive integration, purpose asymmetry, control deficit, return deficit, and bargaining deficit. It further defines “unpaid” as the absence of effective rights over price formation, benefit distribution, attribution, use, and collective negotiation, and “continuous” labour as the absence of institutionally protected off-time governing data capture, retention, derivation, and reuse. Drawing on Terranova, Gray and Suri, Irani, Casilli, Fuchs, Tronti, Alquati, Marx, Braverman, and Pasquinelli, and engaging recent research on expert annotation, AI supply chains, worker positionality, performative accuracy, and labour standards, the article argues that AI capability is accumulated and reorganised human judgement rather than an autonomous technical substance. It treats the AI ghost data worker not as an already unified new class but as a class-formation process: technically integrated across supply chains while politically dispersed. Finally, it develops a non-aggregative FED-DLR audit framework covering labour visibility, human judgement attribution, unpaid data-time exposure, generative outcome return, self-replacement exposure, contestability, collective dialectical capacity, institutional re-entry, and directional legitimacy. The normative question is not only whether AI labour is fairly compensated, but whether the producers of artificial capability can participate in determining its purposes, ownership, substitution boundaries, and future direction.