Uber is turning its app into an AI training ground (www.theverge.com)

Uber announced a US pilot that lets drivers and couriers earn extra pay by completing “microtasks” in the app to train AI models—examples include recording voice samples in local dialects, photographing cars, and uploading documents or menus in other languages (one Spanish menu prompt can pay up to $1). The move expands earlier trials in India and follows Uber’s acquisition of Segments.ai and its stated human-in-the-loop approach, positioning its global network of gig workers as an on‑demand data-labeling force that could compete with incumbents like Scale AI and Amazon Mechanical Turk. For the AI/ML community this is significant because it changes where and how labeled data can be sourced: closer-to-edge, geographically localized samples (dialects, regional menus, street-level images) can improve model robustness and localization, while scale could reduce costs and speed iteration. Technical implications include potential gains in coverage for underrepresented languages/dialects, but also challenges around data quality control, privacy, consent, and incentive design (micro-payments may affect label reliability or invite gaming/data poisoning). The move raises labor and ethical questions too—worker pay, classification, and oversight—and signals platform-level vertical integration of labeling, model ops, and productization that may reshape the downstream supply chain for AI training data.