Uber branches out into AI data labeling

Uber no longer offers just rides and deliveries: It’s created a new division hiring out gig workers to help enterprises with some of their AI model development work.

Scaled Solutions grew out of the company’s own needs for data annotation, testing, and localization, and is now ready to offer those services to enterprises in retail, automotive and autonomous vehicles, social media, consumer apps, generative AI, manufacturing, and customer support. Its first customers include Aurora Innovation, which makes self-driving software for commercial trucks, and game developer Niantic, which is building a 3D map of the world.

The skills Uber requires its new gig workers to have vary, Chris Brummitt, the company’s senior director for communications in Asia-Pacific, said via email. “It depends on the task, but some might need language skills, some programming, some no specific skills.”

No pay guarantee

The company has begun signing up contractors in India, the US, Canada, Poland, and Nicaragua. Workers must agree to a comprehensive, non-exclusive contract, and are paid monthly based on tasks successfully completed. Uber has not revealed pay rates, merely saying in the contract that it varies according to the task, and that contractors are not guaranteed they will be paid the statutory hourly minimum wage for their region.

Uber is also recruiting corporate positions for the division in San Francisco, New York, and Chicago.

This kind of business process outsourcing (BPO) isn’t new. Data labeling in particular is a growing market, as companies rely on humans to check out data used to train AI models. One vendor providing these services, Scale AI, claims to be valued at over $13 billion, although it has run into criticism for the way it treated outsourced data annotation workers.

Data as a product

Irina Sedenko, principal research director at Info-Tech Research Group, said Uber’s entry into the AI data labeling market will allow it to leverage its experience working with massive datasets and technology platforms to support data processing and model development.

“This move allows Uber to capitalize on the growing demand for AI development services and tap into a new market. It is also an example of a company launching a new data-driven business capability and defining data as a product,” she said.

CIOs will want to think carefully about whether to trust Uber — or any other data labelling outsourcer — with their data.

 “If an organization brings its data to Uber’s platform so that it can use Uber’s workforce and technology to label its data, then it needs to work with Uber to understand how the data is going to be stored and used, who will have access to data, and so forth,” she said. The requirements will vary, depending on whether the data to be labeled is proprietary or public, and whether there are any special considerations around privacy or security.

Uber’s senior genAI services advisor, Nate Carson, addressed this issue in a post on LinkedIn, writing, “We are not just working with ‘gig workers’ […] We also have secure/internal networks when dealing with highly sensitive data.”