Leading AI-data firms frame human expertise as a low-cost, extractable input to power models and pitch institutional ‘liberation’ to unlock talent for AI — a narrative that could reshape white-collar labour markets and the value of expertise.

Main Finding

The expert data annotation industry frames a future in which “expertise” is commodified and cheaper to produce via hourly-paid gig labor for AI training. Industry messaging (from Surge AI, Scale AI, Handshake AI, Mercor, and Turing) presents: AI expertise as low-cost but not yet fully realized; human expertise as an extractable, measurable input whose value is judged relative to whether it can already be encoded in models; and institutional expertise as a legacy asset needing “liberation” or reform so its knowledge can be captured for AI. This vision implies broad revaluation of skilled white‑collar labor, institutional power over knowledge, and the price/availability of expert inputs to AI systems.

Key Points

• Industry narrative (nine themes, grouped):
  • AI expertise: (1) cheap (higher ROI than paying humans), (2) needs recontexualization (real‑world evaluations/environments), (3) unrealized in the very near term (thus continued need for human demonstrations).
  • Human expertise: (1) decoupled from the expert (treated as harvestable data), (2) judged relative to AI (valuable where not yet encoded), (3) elastic (value contingent on labs’ immediate needs).
  • Institutional expertise: (1) in need of liberation (institutions hoard knowledge), (2) performative (credentials less reliable than annotation performance), (3) inverted by AI augmentation (those embedding AI gain status/power).
• Context: Meta’s June 2025 49% ($14.8B) investment in Scale AI catalyzed competitive positioning by multiple firms to supply expert demonstrations to top AI labs, creating a rapidly expanding “expert gig economy.”
• Scope of analysis: 1,254 X (Twitter) posts and 29 CEO podcast appearances totaling ~20:47:14 of audio/video from June–Dec 2025.

Data & Methods

• Data sources:
  • Public X (Twitter) posts by five organizations and their CEOs (June 1–Dec 31, 2025).
  • Transcripts of 29 podcasts featuring the CEOs (total >20 hours).
  • Organizations studied: Surge AI, Scale AI, Handshake AI, Mercor, Turing (selected via TIME and Forbes coverage; Appen/Invisible excluded for lack of public X feed).
• Methodology:
  • Directed content analysis with an inductive–deductive coding approach.
  • Built a codebook with three primary codes (AI expertise, Human expertise, Institutional expertise) and inductive subcodes.
  • Applied codes to social media posts and podcast transcripts, then performed thematic analysis to derive nine themes summarizing the industry’s vision.
• Key empirical facts from the dataset:
  • Total X posts reviewed: 1,254.
  • Podcast appearances: 29 by CEOs; combined duration ≈ 20 hours 47 minutes.

Implications for AI Economics

• Labor markets and returns to skill:
  • Potential downward pressure on prices for specialist human expertise as firms commodify annotation labor; this could reduce returns to formal credentials and alter career incentives for knowledge workers.
  • “Elastic” demand for expertise may create episodic high wages for narrowly required skills but lower long‑term security and fewer stable career paths.
  • Risk of skill depreciation: repeated extraction of expertise to train models can reduce future demand for human experts in those tasks (partial automation).
• Market structure, rents, and concentration:
  • Firms that control high‑quality expert datasets may capture substantial rents; vertical integration (e.g., platform + data + model access) raises concentration risks and bargaining asymmetries with both clients and annotators.
  • The industry framing of institutional knowledge as a resource to be “liberated” encourages competitive strategies (buying/contracting access to institutional data/experts) that could externalize value from universities, firms, and professions.
• Productivity and innovation externalities:
  • Lower input costs for expert-labelled data can speed model iteration and lower barriers to deploying specialized AI capabilities—raising aggregate productivity but also accelerating displacement dynamics.
  • Quality vs. cost tradeoffs: “Cheap” expertise risks lowering annotation quality or privileging standardized/replicable annotations over richer, contextual knowledge—potentially degrading downstream AI performance in complex domains.
• Measurement and empirical research priorities:
  • Track wages, contract terms, and hours for domain experts across fields and geographies (heterogeneity by domain complexity and credential level).
  • Measure price elasticity of expert annotation supply and its effect on returns to formal education/certification.
  • Estimate firm-level margins on annotated-data services and the value of proprietary expert datasets.
  • Study causal effects on expert labor demand, occupational mobility, and human capital depreciation when models are trained on expert demonstrations.
• Policy and institutional responses (economic levers):
  • Strengthen labor protections and collective bargaining options for expert annotators (minimum standards, transparent contracting, benefits).
  • Clarify data/property rules around institutionally generated expertise (IP, contract terms for expert contributions, compensation when institutional knowledge is reused).
  • Support reskilling and transition assistance for displaced experts; fund research on human‑AI complementarity to preserve high‑value human roles.
  • Consider competition policy scrutiny for vertical integration and market power (platforms that both supply annotated data and host/consume models).
• Broader welfare tradeoffs:
  • Gains: cheaper access to specialized AI capabilities, faster innovation, broader diffusion of services.
  • Costs: precarious/episodic employment for white‑collar experts, credential devaluation, concentrated rents, and potential quality externalities in high‑stakes domains (medicine, law, safety‑critical systems).

Suggested next empirical steps for economists: - Collect panel data on expert annotator earnings, job stability, and career trajectories pre/post engagement with annotation platforms. - Quantify value contribution of expert-annotated datasets to downstream model performance and firm revenues. - Analyze market concentration trends among annotation providers and vertical relationships with major AI labs.

Summary takeaway: The paper documents an industry narrative that treats expertise as an inexpensive, extractable input for AI—an economic framing with important consequences for labor markets, institutional incentives, market power, and social welfare that merit rigorous empirical study and policy attention.