BY CADE METZ
Richard Socher carries a resume that would seem to make him rather attractive to the giants of the internet.
He just finished a PhD at Stanford University, where he explored a form of artificial intelligence called “deep learning,” teaching machines to recognize images and understand natural language using software that operates a bit like the networks of neurons in the human brain. In recent years, the giants of net—including Google, Facebook, Microsoft, and Baidu—have seized on deep learning as a path to the future of automated computer systems, and they’ve been hiring researcher after researcher from the relatively small community of academics that specialize in this rather complicated technology.
|Richard Socher. MetaMind|
Socher says the big names have knocked at his door—“I had some very, very attractive offers”—but he turned them down. He wanted to start his own company, a company that would build deep learning technologies anyone can use, not just the internet giants. That company is called MetaMind, and it’s backed by $8 million in funding from Saleforce.com CEO Marc Benioff and big-name venture capital fund Khosla Ventures, with Khosla’s resident chief technology officer, Sven Strohband, serving as the new company’s CEO.
“They’re doing some amazing work—Google and Microsoft and Facebook and so on—and their work is impacting a lot of people,” Socher says. “But I felt like there’s a lot more potential if you give those tools to the remaining Fortune 500 companies—or to people on the internet, just to let them play with them on their own.”
MetaMind is just four months old, but its website, launched today, provides a taste of the technology the company will offer to businesses large and small—not to mention anyone else on the net. You can see how its deep learning tools can, say, recognize particular images or understand the meaning of particular sentences. If you drag and drop a few chocolate-chip-cookie photos onto one MetaMind tool, it can then automatically identify other images of chocolate-chip cookies. If you type “bald man on a horse,” it can show you images of bald men on horses.
These are neat party tricks. But if used on a larger scale, this kind of thing can be remarkably effective inside online businesses—i.e. practically any business. It’s certainly useful to Google and Facebook—they’re using similar deep learning technology to better understand search queries and identify images on their own online services—and Socher says MetaMind is already working with a wide range of businesses, including everything from companies with an interest in identifying food photos to medical outfits looking to automatically examine things like body scans and X-rays.
The startup is just one of many created to bring this type of advanced artificial intelligence to the larger online universe. Though several have been bought up by the likes of Google, Facebook, Yahoo, and Twitter, others remain independent, most notably a startup called Clarifai.
But whereas Clairfai focuses on image search, MetaMind aims to offer a rather broad set of tools, including natural language processing. At Stanford, Socher specialized this field, striving to build systems that can understand not just words but sentences or even entire paragraphs.
The jury is still out on MetaMind’s particular tools. But the company has at least pinpointed the important area of research. Though many companies have deployed deep learning tools that recognize images and speech, the “next frontier” is a breed of computer system that can truly understand language, says Yann LeCun, the deep learning founding father who now runs Facebook’s AI lab.
|Sven Strohband. MetaMind|
Yes, he explains, tools like Siri and Google Now can understand words you say, but not necessarily the meaning of those words. The hope is that deep learning will help drive machines that can learn to understand language as they go along. One of the technology’s key attributes is its ability to train itself on certain tasks, and this, LeCun says, is where many believe it can help with natural language processing.
This is the kind of thing being explored by another MetaMind tool, where you can type two sentences and it will tell you how similar they are. If you key in “surfers ride big waves” and “big waves are ridden by surfers,” it will tell you they mean the same thing. It’s the sort of technology businesses can use to, say, automatically answer questions from its customers. “A customer can ask a question a myriad of ways—even though they all mean the same things,” says Socher. Or it could help analyze what customers are saying about a company on social networking services like Twitter.
MetaMind—which currently spans only 10 employees—will act as a kind of deep learning consultant, but it will also offer its own deep learning services and software to businesses. Running across hundreds of machines loaded with tens of thousands of graphics processors, its online service will let business run deep learning tasks without setting up their own hardware. But if a business prefers to run their own deep learning systems, MetaMind will provide the software—and the expertise—needed to do so.
The company’s pitch is rather broad. It’s positioning itself as a catch-all deep learning company, and in all likelihood, it will hone its efforts in the coming months. But for Adam Gibson, the brains behind another deep learning startup called SkyMind, MetaMind is an outfit worth following, mainly because of Socher’s previous work. “They will occupy a niche,” he says, “if only because Richard knows what he’s doing.”