Ghostcrafting AI: Hidden Labor in AI

• Ghostcrafting AI is the practice of leveraging invisible, adaptive labor from marginalized workers to sustain, repair, and enhance AI systems.
• It encompasses diverse tasks from data annotation to infrastructural improvisation, revealing tactical workarounds and communal strategies.
• The concept highlights systemic invisibility via opaque platform policies, emphasizing the urgent need for equitable labor protections and design reforms.

Ghostcrafting AI refers to the practical, material, and algorithmically mediated processes by which hidden and marginalized platform labor not only enables but also sustains contemporary AI systems, particularly through “ghost work” and survival repertoires that are rendered institutionally and technically invisible. Developed through an eight-month ethnography of Bangladesh’s platform labor industry, this concept situates ghost labor as both a foundation for AI development and a complex field of tactical improvisation, exploitation, and infrastructural workaround (Rahman et al., 25 Dec 2025).

1. Conceptualizing Ghostcrafting AI

Ghostcrafting AI expands upon the notion of “ghost work” by elucidating how workers in Global South contexts not only undertake hidden microtasks—such as data annotation, content moderation, and logo design—but also materially sustain, repair, and adapt AI deployments through craftwork and peer learning. The concept foregrounds not just invisibility but the process of active erasure and obfuscation of labor by platform governance architectures.

Key distinctions include:

• Material and situated labor: Beyond mere labeling or microtasks, workers construct and maintain necessary infrastructures (e.g., negotiating shifts at cyber-cafés, leveraging pirated software, orchestrating VPN and ID-masking strategies).
• Communal improvisation: Survival depends on the collective pooling of resources, informal peer mentoring, and shared troubleshooting within localized and often restricted digital/physical geographies.
• Structural specificity: Contextual challenges such as fintech exclusion (e.g., lack of PayPal/Stripe access), language barriers (ad hoc use of multilingual visual tutorials and translation tools), and geographically targeted bans are intrinsic rather than incidental.

This framework moves from diagnosing invisibility to analyzing the mechanisms by which it is produced and maintained at scale via platform policies, payment infrastructures, and legal-compliance protocols (Rahman et al., 25 Dec 2025).

2. Methodological Foundations

The characterization of ghostcrafting AI is grounded in an eight-month ethnographic study conducted across urban and peri-urban Bangladesh (Dhaka and Jashore) from May 2024 through August 2025. Empirical data collection involved:

• In-depth observation: 18 sessions (~60 hours) in cyber-cafés and training centers, capturing detailed workflow practices and infrastructural arrangements.
• Semi-structured interviews: 34 participants (ages 19–39; educational background from primary school to master’s degree; majority male) active across major global platforms (Upwork, Fiverr, Freelancer.com, CrowdGen, 99designs, Toloka, PeoplePerHour). Tasks included data annotation, UI/UX design, translation, web development, and training.
• Biography making: Three extended life-history narratives capturing diverse trajectories, from urban migration to family agency formation.
• Thematic coding: Inductive open coding and clustering of ~160 pages of fieldnotes, >200 images, and 22 hours of recorded audio.

This methodological rigor supports cross-role, cross-platform, and intersectional analysis of ghostcrafting practices, with emphasis on both systemic structures and granular tactics (Rahman et al., 25 Dec 2025).

3. Economic Structures, Workflows, and Precarity

Ghostcrafting AI is embedded within a precarious socio-economic system characterized by low and volatile wages, unreliable payments, and exploitative platform fee regimes:

• Wage regime: Microtasks (image labeling, data entry) typically pay USD 0.20–1 per item; mid-level gigs (logo design, transcription) range USD 5–20; complex projects may reach USD 30–50. Platforms routinely extract 20% commissions, layered with withdrawal fees (3–10%) and adverse currency exchange rates (5–10%).
• Payment insecurity: Withholding of payments is common, attributed to ambiguous algorithmic suspensions and non-transparent dispute arbitration. For instance, account lockouts and escrow freezing (e.g., $45 lost from a flagged university-assignment gig) were recurrently reported.
• Aspirational mobility: Despite evident precarity, platform labor is a channel for rural-urban mobility, social prestige, and entrepreneurial agency formation (e.g., hiring subordinates for collective gigs).

Workers navigate these conditions by exploiting off-platform networks, optimizing low-cost access to shared computer infrastructure, and leveraging informal financing to maintain contingent engagement (Rahman et al., 25 Dec 2025).

4. Tactical Repertoires and Peer Infrastructures

A defining trait of ghostcrafting AI is the extensive array of tactical (often extra-legal) practices developed to circumvent or mitigate algorithmic exploitation and infrastructural exclusion:

• Identity masking: Workers deploy VPNs, remotely rent U.S. phone numbers/bank accounts, and borrow national IDs to bypass age or geography-based restrictions.
• Fee bypassing: Contact information is concealed within gig images or embedded via codeword puzzles (e.g., numeric strings spelled out in text) to move clients off-platform and avoid commission/skimming by platforms.
• Pirated tool maintenance: Stable “safe” cracked software versions (e.g., older Photoshop) are propagated through informal mentorship and WhatsApp groups to guarantee workflow continuity under platform detection regimes.
• Portfolio workarounds: Due to non-disclosure agreements or platform bans on attribution, workers maintain shadow portfolios or claim authorship post contract expiration.
• Rating management: Rapid accumulation of positive ratings is achieved by accepting dozens of underpaid microtasks before pivoting to higher-value work, compensating for internally opaque status algorithms.

Peer networks (class WhatsApp groups, mentor-apprentice arrangements) serve as the backbone for troubleshooting technical issues, emotional resilience, and real-time intelligence on platform behavioral shifts. Survival and success are contingent upon continuous adaptation of these repertoires (Rahman et al., 25 Dec 2025).

5. Algorithmic Governance, Bias, and the Production of Invisibility

At the infrastructural level, platform governance systematically enforces invisibility through:

• Opaque algorithms: Search and ranking visibility is increasingly pay-to-play; ratings systems penalize anything less than perfect five-star reviews, often overriding actual client feedback.
• Suspension and dispute systems: Account deactivations or payment freezes are automated and rarely transparent, with negligible recourse for direct human appeal.
• Platform exclusion: Absence of mainstream fintech integrations (e.g., PayPal, Stripe) forces reliance on inefficient and fee-heavy intermediaries.
• Legal and policy erasure: NDA clauses, anti-attribution mandates, and automated plagiarism checks erase authorship claims for the bulk of platform labor, enforcing a regime where ghostcrafted contributions cannot accrue to worker reputation or future opportunities.

These structural mechanisms function less as oversight and more as disciplinary infrastructures, optimizing surplus extraction while minimizing the risk of worker organization or collective bargaining (Rahman et al., 25 Dec 2025).

6. Policy, Design, and Governance Implications

The ghostcrafting AI phenomenon exposes both the dependency of contemporary AI pipelines on invisibilized, improvisational human labor and the urgent need for multi-level intervention:

Design recommendations:

• Incorporate fairness and accountability by default in workflow algorithms (e.g., delayed blind mutual ratings, accessible appeals processes).
• Integrate translation APIs and real-time support for multilingual workers.
• Develop recognition and provenance mechanisms (portfolio-safe fragments, standardized badges, temporary NDA carve-outs) to enable sustainable credentialing.

Policy recommendations:

• Enact cross-border labor standards that recognize online gig work as labor with legal protections (e.g., anti-defamation, wage, and safety protections).
• Mandate transparency and non-retaliatory dispute arbitration.
• Regulate platform fee structures and restrict pay-to-play visibility to mitigate rent extraction.

Collective and institutional direction:

• Formalize peer-to-peer learning platforms for skills accreditation and joint advocacy.
• Leverage NGOs and local training centers to co-design language-appropriate, accessible resources.
• Redistributive obligations: establish regional training funds, emergency-income supports, and third-party credential portability frameworks.

Research trajectory:

• Pursue comparative, multi-country longitudinal studies to observe evolution of ghostcrafting tactics and platform governance structures under changing regulatory regimes (Rahman et al., 25 Dec 2025).

Ghostcrafting AI compels a revision of the boundaries of AI development, foregrounding hidden human infrastructures and pointing to the necessity of regulatory and design reform for long-term sustainability and fairness in the platformized AI economy.