Uber expands AI “data factory” idea by turning driver downtime into labeling work

Uber is moving beyond rides and into AI training data production, telling investors and readers that its drivers can become a scalable workforce for data labeling. In a recent report, Uber said its Uber AI Solutions are now live in 30+ countries, and that the company can draw on its millions of drivers to generate annotated data needed to build and improve AI systems.

Uber’s approach is essentially a shift in where AI training data comes from. Instead of paying independent contractors or using third-party crowdsourcing platforms to tag images, transcribe audio, or classify text, Uber is testing programs that route certain “digital tasks” to drivers who have spare time in the app. The goal, according to the reporting, is to make data production cheaper and easier to scale as AI demand grows.

The concept has been tested in India. Computerworld reported that Uber ran a pilot that let drivers earn extra money by labeling data for AI during downtime, and said the program was designed to compete with traditional data-labeling providers. The pilot covered drivers across 12 cities and included a menu of task types such as image classification, text analysis, object counting, audio transcription, and receipt digitization, according to a statement attributed to Megha Yethadka, Uber’s Global Head of Uber AI Solutions.

Computerworld said Yethadka described the program as giving drivers “more choice, flexibility, and ways to earn,” and added that Uber reported early engagement. The same report noted that some drivers interviewed in New Delhi had not yet seen the feature in their app at the time of the interviews, pointing to a staged rollout rather than a fully open launch.

The underlying business logic is straightforward for Uber and similar platform companies: Uber already has a large, geographically distributed workforce that is routinely offline and online in predictable cycles, and it sits inside a mobile application that can route work to users. If enough drivers choose the tasks, Uber can produce large volumes of labeled or digitized content in-house, potentially improving margins versus buying that work from external vendors.

Still, important details remain unclear in the limited public reporting. Uber has not, in the materials available here, provided specifics on the percentage of completed tasks that come from drivers versus other sources, the average cost per labeled item, quality-control methods, or how performance is measured when the “workforce” is mobile and voluntary. The degree to which tasks are automated or reviewed by human quality teams is also not described.

Looking ahead, the key question for the market is whether Uber can scale these driver-powered workflows without introducing accuracy issues that undermine AI model performance. Observers will also watch for indicates on expansion speed beyond early pilots, whether Uber discloses more about task economics and data-quality governance, and how enterprises evaluate outcomes from Uber AI Solutions compared with established labeling providers.

],

keyFacts 2=[],

keyFacts2 placeholder