Ryan Avent sees them on the horizon:
The world is now blanketed in sensors, most of which are connected to the internet. The machines of the future will be able to draw on that information (or a lot of it, anyway) and use it to inform themselves about their surroundings. They will be able to talk to, learn from, and add to a wealth of data on What Things Are and What To Do With Them. And if they get stuck, they’ll be able to talk to humans. The stray object in the path that might once have immobilised a robot won’t immobilise it if the robot can ring up tech support and informed that it’s just a stray object, go around dummy.
It’s very hard to design a machine that can improvise when confronted by the unfamiliar or reason its way through most difficulties—just as it’s rare to find a human who can seamlessly navigate his way across all of America’s public roads, large and small, without some sort of guide. But just as any regular joe with access to Wikipedia can do a passable impression of someone with enormous intellectual powers, the extended mind of the cloud could lead to impressive improvements in robot capabilities.