Baidu’s Push into AI Hints at Why Google Loves Robots

One of the big mysteries in the tech industry is why Google has acquired eight robot companies. The activity of Baidu—Google’s main rival in China—in artificial intelligence suggests that tech companies’ interest in robotics could be motivated as much by software as hardware.

Baidu believes that in five years, half of all Internet searches will be done using speech or images, rather than text. That shift will make artificial intelligence strategic to the Internet’s future, said Andrew Ng, a Stanford University researcher who joined Baidu earlier this year.

Ng, who spoke at the EmTech conference yesterday, is an expert in deep learning, an artificial intelligence technique loosely inspired on how the brain works. When he worked at Google, he started the Google Brain project that used thousands of interconnected processors to discover patterns. For example, by viewing You Tube for a week, the system was able to identify people’s faces and, without any prior training, learn what a cat was from the high volume of cat videos.

“To the extent that AI, or perhaps deep learning, is the critical technology for helping computers understand speech, images, and text, many days when I wake up I think that whoever wins AI wins the Internet,” Ng said.

About 10 percent of searches done in China are done through speech, typically on a mobile phone. Often people will dictate a full sentence, such as “when will those noodles I had last week go on sale,” which makes the queries much harder to decipher than text. Machine learning—teaching a network of computers to recognize patterns by feeding it large amounts of data, rather than giving it explicit instructions—is better suited to answering these types of search and image-based queries, Ng said.

A person could, for example, dictate and send a text into a phone while driving or take a picture of clothing to search for similar items. “As the world moves to mobile devices, I think speech will be incredibly important. I would love to redesign your cell phone around speech recognition,” he said. “Speech and images are just much more natural ways for people to communicate.”

Baidu, who heads a lab in Silicon Valley, said deep learning could enhance a head-mounted computer that Baidu is working on. A person could go to a museum and the smart glasses, which have look similar to Google Glass, would be able to recognize images of the artwork and provide information to the user through the earpiece.

Google, Apple, and Facebook also have ongoing research in artificial intelligence to improve how their services can handle images, video, and speech. A search on LinkedIn shows that Apple, for instance, has a number of job openings related to computer vision, machine learning, and writing algorithms that can recognize patterns from sensor data.

How does this relate to robotics? Robots need artificial intelligence software to better understand and move around their environments. In fact, Ng is also a robotics expert and spent years at Stanford applying machine learning to robots.

Google’s interest in robots is deeper than other tech companies, although it’s intentions are not totally clear. The company disclosed last year that it had acquired eight robotics companies and put former Android chief Andrew Rubin in charge of its robotics effort.

Google already has developed self-driving cars but people have speculated that it could break into other related areas, such as robots for manufacturing and warehouse fulfillment or drones to deliver packages. The company also acquired a deep-learning company earlier this year called DeepMind, giving it technology that could be used to improve Web search or, perhaps, the performance of robots and connected devices.

One company Google purchased is Boston Dynamics, which is making a series of robots that try to mimic the motion of animals. Under contracts from DARPA, it made four-legged robots that can carry hundreds of pounds of gear for 20 hours and one that can run more than 30 miles per hour.

Marc Raibert, the CEO and founder of Boston Robotics who was also at EmTech, said DARPA was a great benefactor for the company because it had lots of money and vision, which he sees continuing under Google. “Google has more money and lot more ambition and vision,” he said.

Trending on Xconomy