Uber’s efforts to stay one step ahead of the competition by investing heavily in robotics and AI research are showing signs of trouble.
In recent months, Uber has lost several senior members of its Advanced Technologies Group, a self-driving car project headquartered in Pittsburgh. And the head of its new AI lab, Gary Marcus, also stepped down from his role after just a few months in charge. These are part of a bigger picture that highlights the challenges involved with commercializing technology that remains extremely complex and cutting-edge.
Uber created its AI lab in December after acquiring Geometric Intelligence, a startup headed by Marcus, a cognitive scientist from New York University. Marcus, who remains an advisor on AI to Uber, will discuss the challenges that remain in artificial intelligence today at EmTech Digital, a conference organized by MIT Technology Review.
The newest setback from Uber came last week, when it was forced to halt testing of its self-driving vehicles in Arizona after one car was involved in an accident with another vehicle. There is no indication yet that the self-driving car was at fault.
As Marcus will explain, making computers as smart as humans in critical situations such as driving remains a formidable challenge. Self-driving cars cannot yet react to any eventuality they might encounter on the road, and they require huge amounts of data to learn.
Uber has rushed to develop automated vehicles for fear that the technology could easily disrupt the taxi industry. The company got up to speed quickly, and has self-driving cars on the roads of several cities. But as MIT Technology Review discovered, these systems do not yet work perfectly, even in ordinary driving situations.
There are significant engineering challenges, too. For example, it isn’t clear how to make self-driving cars cope with degraded sensors, or how active systems like lidar, a type of laser system, might interfere with each other if lots of self-driving cars were on the roads (see “What You Need to Know Before Getting in a Self-Driving Car”).
Marcus has been an outspoken critic of what he sees as an overreliance on neural-network-based machine-learning approaches in artificial intelligence. He founded Geometric Intelligence, in 2014, to explore alternative approaches (see “Can This Man Make AI More Human?”).
Among other things, Geometric Intelligence sought to find more efficient ways for machines to learn. While a human can learn to recognize a new traffic sign very quickly, a computer requires many thousands of examples using today’s best machine-learning approaches.
Other companies working on automated driving have also found progress slower than they might have hoped. Google has spun out a company, called Waymo, out of its self-driving car project, but its technology is not yet available commercially.
DeepMind’s cofounder: Generative AI is just a phase. What’s next is interactive AI.
“This is a profound moment in the history of technology,” says Mustafa Suleyman.
What to know about this autumn’s covid vaccines
New variants will pose a challenge, but early signs suggest the shots will still boost antibody responses.
Human-plus-AI solutions mitigate security threats
With the right human oversight, emerging technologies like artificial intelligence can help keep business and customer data secure
Next slide, please: A brief history of the corporate presentation
From million-dollar slide shows to Steve Jobs’s introduction of the iPhone, a bit of show business never hurt plain old business.
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.