Human Labor Behind AI: From Ghost Workers to Platformized Experts

• In Event: Organizations, Science, and Expertise

Abstract

This paper examines the evolving role of human labor in the development of Artificial Intelligence (AI), tracing a transition from invisible “ghost workers” performing simple, low-wage microwork to a new class of “platformized experts”. As large language models (LLMs) approach a “data wall” - having already consumed much of the high-quality data available for scraping - AI labs increasingly rely on domain-specific judgment for tasks such as evaluation, safety, and alignment. Building on scholarship on microwork, human-in-the-loop systems, and data colonialism, the paper argues that AI supply chains are being re-stratified rather than “automated away”. Empirically, it draws on a case study of Mercor, a platform intermediary that brokers short-term, hourly contracts between AI labs and professionals (e.g., radiologists, lawyers, finance specialists, engineers, and multilingual consultants). Based on a snapshot analysis of 223 Mercor job postings (December 7, 2025), the paper shows that expert data work is often well compensated (average $74/hour), yet remains organized through contingent, platform-mediated relations that reproduce familiar vulnerabilities: opaque governance, uneven access to tasks, and regional pay disparities. The central paradox is that contemporary AI development simultaneously pursues the reduction of human involvement (e.g., through AI-mediated feedback) while intensifying dependence on human expertise to handle ambiguity, values, and accountability. As an extension of this analysis, I am currently conducting semi-structured interviews with domain experts (such as senior engineers, linguistic experts) who participate in this form of platform-mediated expert data work. These interviews examine how experts interpret their own role in the AI supply chain- whether they experience the work as automating their own jobs away, or as contributing to domain development, or as something more ambivalent?

Author

• Sayam Siddha, The New School