Google has similar goals. So do Microsoft, Yahoo, China’s Baidu and other companies. IBM poured more than $1 billion into its Watson computer system, which uses artificial intelligence and competed on “Jeopardy.”
Already, Google has used artificial intelligence to improve its voice-enabled search and Google Now, as well as its mapping and self-driving car projects. Google wouldn’t discuss the details of its projects, but CEO Larry Page showed his enthusiasm at a TED technology conference in March.
“I think we’re seeing a lot of exciting work going on, that crosses computer science and neuroscience, in terms of really understanding what it takes to make something smart,” Page said. He showed videos from Google and DeepMind projects in which computer systems learned to recognize cats and play games — without detailed programming instructions.
Google and Facebook both hope to do more with “deep learning,” in which computer networks teach themselves to recognize patterns by analyzing vast amounts of data, rather than rely on programmers to tell them what each pattern represents. The networks tackle problems by breaking them into a series of steps, the way layers of neurons work in a human brain.
The approach was pioneered in the 1980s by a handful of scientists including Hinton, Ng and LeCun. But researchers say its potential has exploded in recent years because they now have access to more powerful computing systems and bigger sets of data.
That technology could help companies build systems that go beyond recognizing words or phrases, to understand the intended meaning of written texts and conversational speech — so instead of typing the keywords “weather forecast San Jose,” users can simply ask: “Do I need an umbrella today?”
It may also help companies like Google and Facebook analyze individuals’ posts and preferences to tailor the ads they see, though neither company tends to highlight such uses.