Uber’s AI Ambition
Using Artificial Intelligence to Transform Urban Transport
A year ago, Uber acquired artificial intelligence and machine learning start-up Geometric Intelligence to create a new division, Uber AI Labs. Its ambitious goal is to gather top scientists and engineers to create the next generation of machine learning algorithms that will improve the lives of millions of people worldwide. Berenice Baker asks what this could mean not only for AI research, but for future cities and transport
Uber is the fairy godmother app that brings us home even when we’re not sure where we’ve been, and UberEATS delivers food from our favourite takeaways. And all of this without awkward conversations about arrival time, estimated cost or stopping off at a cashpoint.
The service improves year-on-year thanks to the masses of data parent company Uber Technologies Inc. has gathered over more than five billion rides across the 633 cities in which it currently operates.
With the launch of Uber AI Labs, Uber is now waving the magic wand of artificial intelligence and adding a pinch of machine learning (ML), which in the short term will make your ride or food order even more efficient and better value. Further into the future it could make visions of self-driving cars, urban aviation and optimised cities a reality, while making streets safer for road users and pedestrians.
Cutting edge of AI research
To ensure Uber AI Labs could hit the ground running with its research, Uber acquired Geometric Intelligence. Before joining Uber AI Labs full time, Geometric Intelligence co-founder Ken Stanley was an associate professor of computer science at the University of Central Florida specialising in AI and machine learning, specialising in an area called neuroevolution.
“Our vision when I co-founded Geometric Intelligence with Gary Marcus, Zoubin Ghahramani and Doug Bemis in 2015 was to create a world-leading commercial research lab for AI and ML,” says Stanley.
“To do that, our idea was to combine the expertise and insights of diverse areas within these fields that do not normally interact closely. In other words, we saw an opportunity to capitalise on the diversity of ideas from experts from disparate research areas. We understood that such a combination of expertise would naturally lead to inventions and discoveries with real value, which is why the time was right to start such an effort.”
“Uber has the resources and existing expertise to support and grow a genuinely world-class research lab.”
Stanley says that Uber’s desire to compete at the cutting edge of AI research was an excellent match for Geometric Intelligence’s ambitions.
“It allowed us to take the team that we built and embed it in an environment with numerous problems suited to ML, but also with real recognition of the long-term value in progress in the methods of ML itself,” he says. “Uber also has the resources and existing expertise to support and grow a genuinely world-class research lab.”
Day-to-day, Uber AI Labs engages in both applied and fundamental research.
“We are applying ML techniques to the challenges of Uber and inventing and investigating new human knowledge on ML, which can in turn feed back into addressing Uber’s business challenges,” Stanley explains.
“These challenges include problems from autonomy to logistics and run the gamut of machine learning applications.”
Uber’s AI tools and techniques
Uber needs to gather masses of data to make predictions about market demand, find optimal routes for drivers, respond to support issues in natural language, update its knowledge of changing roads and even detect and respond to potential fraud.
Applications like autonomous driving may use specific instruments such as using vehicle-mounted cameras to collect information, while others respond to data fed over networks.
But when it comes to pinning down a timescale by which this could translate into ambitious applications like self-driving – or even flying – cars, Stanley admits it’s difficult to speculate with specificity that far in advance.
“We are going to be part of the story of how Uber addresses these ambitions and challenges,” he says. “I think you will increasingly find our work under the hood in all these areas, often a piece of the puzzle but not necessarily the whole product. You can already see this process starting with the lab’s recent release of Pyro and some of our projects are already having impact addressing Uber’s core business needs.”
“I think you will increasingly find our work under the hood in all these areas, often a piece of the puzzle but not necessarily the whole product.”
Pyro is a general purpose tool for AI developers that can be applied to making complex predictions and inferences in the face of uncertainty. Uber recently made it open-source.
“It makes building the models to make these kinds of decisions easier and makes collaboration among researchers interested in these types of problems easier by providing a common platform,” says Stanley.
“It represents the kind of positive contributor Uber AI Labs aims to become within the larger AI community. Pyro will benefit practitioners and researchers across a wide range of industries and Uber AI Labs is happy to be able to make that kind of positive impact.”
Despite its ambitious aims and achievements, Stanley clarifies that while AI is a priority, Uber is fundamentally a people-first company.
“We recognise the preeminent importance of AI and ML to our business and to the wider technology industry in general,” he says. “We’re committed to recruiting the best and brightest in AI and ML to help create a better experience for drivers, riders, and cities.”
Artificial intelligence and the data dilemma
Professor Noel Sharkey, emeritus professor of Artificial Intelligence and Robotics & Public Engagement at the University of Sheffield, cautiously welcomes the launch of Uber AI Lab and its research.
“It looks very exciting in its efforts to combine a number of probabilistic and machine learning methods like deep learning, evolutionary algorithms and Bayesian techniques,” he says.
“They are making this all open source so that the AI community get to use the fruits of their labour although this is a clever idea to get a lot of free development of the techniques.”
“We have no idea what our data will ultimately be used for.”
However he cautions that a potential downside arises because all of these methods are data hungry and that means collecting as much data about its customers as possible.
“Uber already collects massive data about our locations even when we are not using their app,” he warns. “Other apps allow ‘use location data only when app is open’ but with Uber it is either on or off. With data comes power and we have no idea what our data will ultimately be used for.”
Sharkey says there has been concern about big data and learning algorithms causing gender and ethnicity discrimination in a wide range of applications from mortgage decisions to predictive policing, for example.
“Uber's work can be greatly beneficial to the future development of AI but it is operating in an environment that urgently needs regulation,” Sharkey concludes.
Once appropriate guidelines are in place, with some of the best minds in AI and ML working on it, it’s only a matter of time before your Uber ride arrives without a driver behind the wheel.
Image courtesy of Uber