For Google, a leg up in the artificial intelligence arms race
By Verne Kopytoff
Google’s executives have long dreamed of solving one of the technology industry’s biggest riddles. How do you predict what people want — hockey scores or new Ugg boots, for example — before they even ask for it? Reading user’s minds, or at least seeming to, would make Google’s products that much faster and more convenient. It could also help the company fend off rivals.
Last week, Google (GOOG) took its biggest step yet to ramp up its predictive powers. It acquired DeepMind Technologies, a British startup focused on artificial intelligence, for a reported $400 to $500 million. DeepMind develops algorithms that learn as they comb through mountains of data. The company had been working on technology for simulations, e-commerce, and video games, although the exact details were not disclosed.
The addition could help Google’s perpetual efforts to improve its search results and make its ads more relevant. Helping tag photos and improve the accuracy of voice recognition are among the other possibilities.
“Artificial intelligence can become part of the fabric for all of Google’s products,” said Colin Sebastian, an analyst with R.W. Baird. “Google has a knack for adapting and innovating. For a company its size, that’s a requirement — or you risk becoming a Yahoo or AOL,” two companies that are struggling because of an aversion to change.
Artificial intelligence isn’t new. Researchers have worked on the technology for decades. But it has been slow-going because computers haven’t been powerful enough to handle the number-crunching that’s necessary. Recent advancements in technology have sped up the progress.
“Powerful algorithms and computers are able to solve problems that were intractable just a few years ago,” said David Bader, executive director of high performance computing at Georgia Institute of Technology.
Google has been working on artificial intelligence almost since its founding 16 years ago. Its efforts focused on improving search results, translation, and filtering spam. Google Now, a digital personal assistant within Google’s mobile search app, is among the most ambitious of its efforts to date. It guesses what information users want based on their past search history and location, and then gives it to them.
“Google Now can do simple things like give you directions to work or figure out that you have a flight,” Bader said. “But to really predict what you’re really going to do or to understand your lifestyle, and be able to do that in multiple languages and cultures requires this new technology.”
Google already collects a huge swath of personal data about its users including their search histories, locations when using their smartphones, and any purchase history made through Google products. Google’s engineers — or rather algorithms — have plenty of information to crunch so as to create user profiles and gauge user intent.
DeepMind, which was founded four years ago, has been trying to get computers to learn much like the human brain does. Algorithms search data for patterns and then draw conclusions from it in what’s known as deep learning. The company, which has around 50 employees, has yet to come up with a commercialized product. But a recent paper published by some of its researchers provides a window into its progress.
The paper describes a computer that learned to play seven Atari 2600 games. Feeding the computer with information about how to play specific games was unnecessary, the researchers said, a sign that the system learned to play from first-hand experience.
Google’s rivals are also hard at work on artificial intelligence. Amazon (AMZN), Microsoft (MSFT), and Facebook (FB) all have big efforts underway. It’s a new arms race in the Internet industry. As the DeepMind acquisitions shows, companies are wiling to spend huge amounts of money to keep up.
Analysts said it’s impossible to tell whether artificial intelligence will lift Google’s profits. Rather, they said any improved products that come from the technology would help the company defend its leadership position.
“It’s all about staying ahead of the competition,” said Jan Dawson, an analyst with Jackdaw Research. “You won’t be able to point to a specific product and say ‘A.I. contributed this.’”
Artificial intelligence could play an especially big role in Google’s new business lines. For example, earlier this month, Google said it would pay $3.2 billion to acquire Nest Labs, the maker of Internet-connected devices for the home. A thermostat the company already sells infers when homeowners come and go based on usage patterns. Any future appliances that come out of Nest may also depend on artificial intelligence — at least in part — to control the settings or, in the case of a smart television, recommend shows to watch.
Google is also pushing into robotics through the acquisition of a number of companies including Boston Dynamics. A separate project to build self-driving cars is running parallel. Both initiatives require sophisticated technology to be able to maneuver independently of humans. Artificial intelligence is therefore a given, but it’s unclear whether it would involve DeepMind.
Although artificial intelligence is a huge opportunity, Google will have to tread carefully. Privacy concerns and the “creep factor” may rule out certain uses. Imagine if people wearing Google Glass, Google’s science fiction-inspired eyeglasses, could use facial recognition technology to identify everyone passing by them on the street and to pull up detailed information about them. Public opposition would be intense.
Another potential problem is if an algorithm were to draw an inaccurate conclusion based on your personal data. Google might wrongly conclude that you’re seriously ill, for example. Or it could accurately deduce as much and then inadvertently share that information with someone who borrowed your computer by suggesting all sorts of drugs to take.
“Artificial intelligence can be really scary when it crosses the line,” Dawson said.