Uber pays idle drivers to train its AI

Uber has been running the pilot in 12 top cities in India for about two weeks, said the people quoted earlier, who spoke on condition of anonymity.

“The most common model, especially in pilots, is to pay workers a flat wage per task. Cash remains the preferred form of incentive as drivers view this as tangible income rather than a bonus,” said one of the people cited above. Payments generally vary as follows: ₹10– ₹50 for annotation tasks and ₹5– ₹20 for uploading photos, usually calculated per minute or per click. These are sometimes added to the driver’s weekly earnings along with ride or delivery revenue.

Uber has around 1.4 million driver partners in India, and the company makes money from the time its fleet sits idle. Platforms with physical touchpoints such as fleets, IoT devices, connected vehicles take advantage of idle time to create valuable labeled data that can be used to train internal AI models or be sold to third parties.

Global demand for high-quality labeled data is predicted to reach $17 billion by 2030, according to estimates from policy research and advocacy group Consumer Unity & Trust Society (CUTS) International. A 2021 report from Nasscom estimates that India is expected to serve more than $7 billion of the global disclosure market by 2030.

As drivers mark lanes, intersections or sidewalk rules, those inputs are fed into Uber’s training systems, which will help the AI ​​”see” the city more like a human, the first person previously quoted said. The payoff is better routing, sharper ETAs and newer maps, as drivers spot roadworks, new one-way streets or closed storefronts long before official releases, the person said.

Labeling and description

While uber announced The company confirmed on October 17 that it would offer drivers in the US a similar study when they’re not on the road. Mint It is stated that the model is currently being tested in India.

Tagging goes beyond road features to include images and other datasets, but details remain limited. It’s also unclear whether Uber developed its own proprietary disclosure tools or relied on third-party platforms.

“This is a venture into microdata and an attempt to monetize it,” one of the people quoted earlier said, adding: “The end goal is for it to go to external customers and leverage third-party players.”

This means that over time, Uber could package this tagged data (or the insights derived from it) into a commercial service for frequently refreshed geodatasets for external customers like businesses, AV developers, logistics firms, or fact-finding city agencies. This could be channeled through Uber’s AI Solutions unit, which already offers data labeling services to third parties.

Don’t take out more

Uber’s move comes after its US arm acquired Belgium-based Uber segments.aiData tagging initiative , as platforms with large physical touchpoints transition from passively collecting microdata to actively curating and monetizing it, as early as October of this year.

“For a mature platform like Uber, growth is now based on extracting value from the same network rather than adding vehicles,” said Farheen, an analyst at the Center for Critical and Emerging Technologies. “If drivers can use downtime to collect and tag data, the company gets more out of every minute they work.”

From service to knowledge work

Uber’s move reflects a shift in the internet economy, where the lines between service and knowledge work are blurring.

Telecom operators Bharti Airtel is delivering precisely targeted campaigns to 320 million users by monetizing subscriber data allowed through Airtel ads. MapmyIndia monetized detailed geospatial datasets by licensing APIs (tools that allow two software to communicate) for navigation and analytics. Delhivery’s OS1 platform transforms logistics data from billions of deliveries into address verification and routing services, while Ola leverages terabytes of platform data daily for pricing, route optimization and personalization.

Tesla relies on human reviewers to review car images around the world; Amazon pioneered microtasking with Mechanical Turk; And JD.com It tightly integrates artificial intelligence into warehouses.

India’s grade economy is already worth it, according to CUTS ₹2,000 crore and is expected to go beyond that ₹4,000 crore in appliances alone by 2030 and a CAGR of around 29%. India accounts for approximately 7.9% of the global market, with approximately 70,000 reviewers (approximately 50,000 freelancers, 20,000 full-time). Approximately 60% of its revenues come from US customers and it remains predominantly export-oriented.

business account

“Uber clearly sees business upside; its AI Solutions unit already sells tagging services to third parties,” said Sohom Banerjee, senior research fellow at CUTS International. “But the key is labor economics. Turning downtime into jobs can boost a driver’s income, but it also risks deepening precarious piecework, with millions of cloud workers in the Global South already tagging data as underpaid with weak protections.”

Farheen signaled a compromise. “For drivers, this means their business continues to grow even though rewards are not increasing,” he said. Each new microtask expands the platform’s reach without hiring more people, he said.

Uber India reported in FY24: ₹3,860 crore revenue generated ₹It maintains its lead over its rivals with a loss of 89 crore. Ola’s ride-hailing arm hangs ₹1,761 crore revenue generated ₹10 crore lost while logging into Rapido ₹648 crore income but much steeper ₹371 crore lost.

Is it worth the effort?

Not all experts are convinced. “Google and Apple control base map ecosystems with satellites, proprietary sensors and decades of behavioral data,” said Abhivardhan, president of the Artificial Intelligence and Law Society of India. “Uber’s driver descriptions are incremental inputs, not architectural advantages. This is cost arbitrage on labor, not technical differentiation.”

Banerjee highlighted the challenge: “Even after spending half a billion dollars building its own maps to reduce dependency, Uber still lags behind Google’s decade-long advantage in scale and detail.”

In India, foreign mapping rules that limit resolution to one meter and force onshore storage through local partners such as MapMyIndia limit any proprietary advantage. Uber’s integration with open network policies like ONDC further reduces exclusivity because they are interoperable by design.