Attempting to Code the Human Brain

The Wall Street Journal

Startups, Tech Giants Expand World of Artificial Intelligence; Software With an 'Imagination'

By EVELYN M. RUSLI 

Updated Feb. 3, 2014 7:59 p.m. ET

Vicarious's technology created a series of images of cows, using the software's 'imagination,' after being shown one cow. Vicarious

Somewhere, in a glass building several miles outside of San Francisco, a computer is imagining what a cow looks like.

Its software is visualizing cows of varying sizes and poses, then drawing crude digital renderings, not from a collection of photographs, but rather from the software's "imagination."

The technology is the work of Vicarious FPC Inc., a quasi-secretive startup backed by early Facebook Inc. FB -0.80% employees and investors that is part of the rapidly expanding world of artificial intelligence. The company is weaving together bits of code inspired by the human brain, aiming to create a machine that can think like humans.

Such powerful software is still several years away from being fully developed, if at all, and raises all sorts of ethical questions. But the potential applications—such as masterfully translating foreign languages, identifying objects in photos and directing self-driving cars through busy intersections—are so compelling that technology giants like Facebook and Google Inc. are investing heavily in artificial intelligence.

Last week, Google said it purchased a small startup similar to Vicarious, London-based DeepMind, for more than $500 million, according to two people with knowledge of the matter. Facebook was reportedly interested in DeepMind, and two months ago the social network tapped Yann LeCun, a New York University professor who is considered one of the top experts in the field, to lead its new artificial intelligence lab.

The idea of creating smarter computers based on the brain has been around for decades as scientists have debated the best path to artificial intelligence. The approach has seen a resurgence in recent years thanks to far superior computing processors and advances in computer-learning methodologies.

One of the most popular technologies in this area involves software that can train itself to classify objects as varied as animals, syllables and inanimate objects.

The field remains so specialized that Vicarious shares an investor with DeepMind—Founders Fund, run by Facebook investor Peter Thiel —and the two startups briefly discussed creating a singular company in 2010 before going it alone, according to Vicarious co-founder D. Scott Phoenix.

Vicarious has since raised about $16 million from Founders Fund and several early Facebook employees, including Facebook co-founder Dustin Moskovitz and former chief technology officer Adam D'Angelo.

The company is shrouded in mystery, a point often cited by its skeptics. Like DeepMind, it has yet to release any products and may be several years away from doing so. And some of Vicarious's investors, such as Aydin Senkut, the head of Felicis Ventures, have never seen its lab. The founders say it is located somewhere in the South Bay, keeping it secret to prevent malicious hackers from breaking in.

Vicarious was founded by Mr. Phoenix and Dileep George, a Stanford Ph.D. graduate who studied hierarchical models of the brain. Their premise was to focus on the sensory aspect of the brain, particularly vision's critical role in the early stages of human development. It has tried to further differentiate itself from its peers by designing a system with a high degree of interactivity between the basic visual receptors of the software, its eyes, and the higher-level, information processing parts. Such a feedback loop allows the machine, for example, to imagine the missing contours of a cat that is partially hidden behind a box.

Like an infant, Vicarious's software started with the basics, first learning to recognize simple shapes such as text. Now it is beginning to understand texture and lighting. Eventually, Vicarious's researchers hopes the software will learn how to move within the physical world and understand cause-and-effect relationships.

Vicarious's team of eight is best known for claiming to break the online Captcha test, or the Completely Automated Public Turing test to tell Computers and Humans Apart. The test, which typically appears on websites before registration or payment, shows a series of slightly jumbled numbers and letters that make it difficult for a computer to scan. Last October, Vicarious announced that its software can break Captcha 90% of the time.

The announcement irked some in academia who questioned the strength of the software and a lack of verifiable data.

Mr. LeCun, of Facebook's artificial intelligence lab, wrote in an online post that Vicarious's announcement was a "textbook example of AI hype of the worst kind." He says Vicarious needs to release more information about its technology through academic papers or to test its algorithms against widely approved data sets.

Dr. George stands behind the results. He says the team is conservative about how much information it discloses because of competition concerns and to prevent malicious actors from replicating the software.

Beyond Captcha, Vicarious's visualization software still needs work. In the example of the cows, the images are pixelated and in grayscale. While the software successfully created cows in varying positions—by pulling not only from its knowledge of a cow's image but also how other animals it has seen behave, move and distribute body weight—some cows still came out distorted. One it drew, for example, had a very long neck.

"We don't have the perception problem solved," Mr. Phoenix said.

Though the research is still young, tech giants are already dreaming up a big future for artificial intelligence.

In a recent earnings call, Facebook CEO Mark Zuckerberg said he is interested in artificial intelligence that will help Facebook better understand users. Such technology could help Facebook understand the objects inside users' photographs, such as handbags or food, which could lead to more targeted advertising.

London-based DeepMind, another secretive startup whose technology has been demonstrated with computers playing videogames, could theoretically help Google improve search results.

In the more distant world, one could imagine Jetsons-like robots that could run medical tests or fix damaged nuclear reactors.

The companies are sensitive to societal risks. DeepMind, as a condition of its acquisition, pushed Google to create an internal ethics board. Vicarious was purposefully incorporated four years ago as a flexible purpose corporation, which requires it to have a civil purpose. Vicarious founders, for instance, say they will never create or sell software that would be used in military attack drones.

For now, such dreams are far off. Vicarious said it may need another five to 10 years and more engineers. But if it can graduate beyond pixelated cows, the payoff could be huge.

"If you invent artificial intelligence, that's the last invention you'll ever have to invent," Mr. Phoenix said.

—Rolfe Winkler contributed to this article.

Write to Evelyn M. Rusli at evelyn.rusli@wsj.com