A look at a new labor force using microtask platforms like Remotasks to create and edit data sets for self-driving cars, located primarily in the Global South
To master the roads, autonomous vehicles need lots of data. Workers everywhere from Kenya to Venezuela are providing it.
Context & Ripple Effects
This 2021 Rest of World report was the early map of a labor market that later coverage kept returning to: autonomous-vehicle firms outsourcing the creation and editing of road-scene datasets to gig workers in Kenya and Venezuela through platforms like Remotasks. Within a year, MIT Technology Review documented how Appen and other data-labeling firms scaled the same playbook, concentrating annotation work in crisis-ridden economies where wages are lowest.
What began as AV-specific labeling has since widened and destabilized: by 2025, Venezuelan annotators reported that jobs they had depended on grew scarce as generative AI reshaped demand, while Rest of World's interviews with Kenyan labelers revealed Chinese firms routing hiring through opaque WhatsApp middleman networks that insulate platforms from accountability.
First-order effects
- Self-driving car developers get human-labeled training data at Global South wage rates, while workers in Kenya and Venezuela gain a dollar-denominated income stream that is often their main access to the AI economy.
Second-order effects
- Rivals like Appen replicate the model and compete on locating ever cheaper, more distressed labor pools, pushing pricing for annotation work down and intermediaries' margins up.
- Because platforms avoid direct employment relationships, accountability gaps open — the structure that later let Chinese firms hire Kenyans through informal middlemen with no recourse for workers.
Third-order effects
- Annotation demand proves cyclical rather than durable: when generative AI shifted what data buyers needed, Venezuelan annotators saw the work dry up, exposing an entire workforce built for one AI era to obsolescence by the next — a precarity now extending into new frontiers like Indian workers paid to wear cameras for robot-training video.
The trend: AI's training-data supply chain is becoming a globalized gig labor market whose geography follows wage arbitrage and whose demand swings with each shift in what models need to learn.