Artificial intelligence remains key as Intel buys Nervana

With Intel’s acquisition of the deep learning startup Nervana Systems on Monday, the chipmaker is moving further into artificial intelligence, a field that’s become a key focus for tech companies in recent years.

Intel’s purchase of the company comes in the wake of Apple’s acquisition of Seattle-based Turi Inc. last week, while Google, Yahoo, Microsoft, Twitter, and Samsung have also made similar deals. Those purchases – large tech firms have bought 31 AI startups since 2011 – according to the research firm CB Insights, also underscore a shift.

While AI was once thought of as a sci-fi concept, the technology behind it has come to propel a slew of innovations by hardware and software companies that both attract attention – like self-driving cars – and ones that often go unnoticed, like product recommendations on Amazon.

“Intel’s acquisition of [Nervana] is an acknowledgment that this area of deep learning, machine learning, artificial intelligence, is really an important part of all companies going forward," says David Schubmehl, an analyst who focuses on the field at the research firm IDC.

Intel also purchased Saffron Technologies, which aims to solve complex problems using “cognitive computing,” last fall.

But, Mr. Schubmehl tells The Christian Science Monitor, “I think this [recent purchase] is noteworthy because it provides [Intel] with both software and hardware technology.”

Nervana’s focus, deep learning, builds on advances in machine learning, the idea of training a computer to “think” and recognize patterns.

Deep learning expands so-called neural networks, which are modeled on the human brain, to a much larger scale, allowing a computer to perform tasks such as recognizing an image of a cat without necessarily having prior knowledge of what a cat looked like. 

For Intel, which has focused on cloud-based computing systems, machine learning represents a necessary step to contend with the reams of data they generate.

 “There is far more data being given off by these machines than people can possibly sift through,” Jason Waxman, an Intel vice president, told Recode.

Working with Nervana also offers Intel an opportunity to expand machine learning capabilities in its own chips, rather than simply having software that works on top of them, Recode reports.

For many users, these technologies often go unnoticed, Schubmehl notes. 

“Do you really spend time thinking about when Amazon says ‘Hey, other people like this product, would you be interested in this product?’” he asks. 

“That’s a product of machine learning, but you don’t think of it that way. You just look at as ‘Gee, Amazon actually knows me a little better than other people, and they’re recommending products that I’m actually interested in,’” he adds.

In other cases, users have been quick to point to concerns about potential biases hidden in such algorithm-driven technologies.

Last year, some complained when they noticed that the first result for a Google search for “female CEO” was a Barbie doll. In June, a video showing that a Google search for “three black teenagers” produced a series of mugshots, in sharp contrast to a search for “three white teenagers,” went viral.

Civil libertarians and researchers have also expressed alarm about what happens when such technology is combined with personal data about users, as in tech firm Samsung’s acquisition of the “big data” company Joyent.

“I think that a company that is involved in manufacturing devices that people hold so close to their hearts, it behooves them to tread very carefully,” Jay Stanley, a senior policy analyst at the American Civil Liberties Union's Speech, Privacy, and Technology Project, told the Monitor in June.

But at this point, many companies' focus on machine learning shows little sign of slowing down, Schubmehl says.

For businesses making hardware, whether Intel, the carmaker Tesla, or the power titan GE, the technology offers the ability to integrate their products with software in a variety of ways. Self-driving buses, or robotic manufacturing equipment, he notes, are two of the results.

"At some point it will change enterprise software, and technology in general, so that everything is machine learning-enabled in one fashion or another, so it will slow down at that point," he says. "But, I think for the foreseeable future, there’s a lot of what I would call ‘dumb’ software out there that has to be made smart."