Forget ride-sharing, Uber AI Labs wants machine-smarts everywhere

Chris Davies - Dec 5, 2016
Forget ride-sharing, Uber AI Labs wants machine-smarts everywhere

Uber has just revealed its latest push into artificial intelligence, opening Uber AI Labs and giving it an instant team by acquiring a specialist startup in the field. The company, best known for its ride-sharing platform, has been pushing into AI as part of its self-driving car research for some time. Now, it’s setting those efforts up in a standalone division.

Like most of the transportation-focused research centers founded by automakers, Uber AI Labs will be based in the Bay Area: specifically in San Francisco. It’ll be “dedicated to cutting-edge research in artificial intelligence and machine learning,” Uber says, but go beyond just vehicles that drive themselves. Instead, the goal is to figure out how computers could better navigate a world – and deliver Uber’s services, naturally – when surrounded by generally unpredictable humans.

That might be dispatching a self-driving cab alternative to wherever you summoned it from your phone, navigating between other vehicles and figuring out the most efficient route to pick up your UberPool co-traveler. However, it also has implications for collecting and delivering UberEATS orders, or helping autonomous vehicles in dealing with accidents. Uber suggests the technology could be just as much at home embedded in aircraft or “new types of robotic devices” as it is in cars.

The first into the new lab will be the fifteen members of Geometric Intelligence, an AI startup which Uber has recently acquired. One of numerous companies working in the field, Geometric Intelligence differentiated itself in how it attempts to educate a smart AI. Rather than feeding it the results of hundreds of thousands of prior examples, from which the AI could intuit patterns and such, the team instead took the angle of teaching a novice computer in a similar way to how human children learn.

In that way, the AI could hopefully learn the process of learning itself. That has huge implications in figuring out how to safely travel from A to B through a city, for instance. While Google and others are feeding their self-driving algorithms with millions of miles of road experience and extrapolating best practices from that, Geometric Intelligence’s approach could be more like a teenage driver learning and then taking a test.

NOW READ: How AI can provoke horror

Uber AI Labs will be directed by Gary Marcus, CEO and founder of Geometric Intelligence, and also a Professor of Psychology and Neural Science at NYU. His team has a collective background in IBM Watson, neural networks, evolutionary biology, robotics, crowdsourced algorithms, and particle physics, among other fields.


Must Read Bits & Bytes