Google is doing WHAT?

July 7, 2008 (Computerworld) With a skyrocketing stock price, fanboy hysteria and -- most importantly -- really useful products, Google Inc. is the prima donna of tech for the new millennium.

The company is so active that it's hard to keep track of everything it does. And, just when you get a good handle on its litany of Web applications, promising lab innovations and unheralded research projects, it seems to turn on a dime -- a difficult move for a $167 billion company with 19,000 employees -- and invent something new. Who would have thought a search site company would get involved in laying a fiber-optic undersea cable between the U.S. and Japan?

Of course, not everything has worked out for the company, as these flubs, flops and failures illustrate. JupiterResearch analyst Michael Gartenberg, for one, isn't put off by the wide range of directions the company has taken and occasional miscues.

"The whole Google empire started as a research project, and it's a core in their DNA to try and discover new things and figure out how to monetize them," he says. "When you have a market cap like they do and the cash cow in the guise of paid search, they can keep experimenting. You need the financial wherewithal to support these projects, and plenty of smart people to carry them out. Google does not seem short on either."

Truth and rumors

Here's an update on some of Google's most interesting projects, including some new details about Android, energy initiatives, language translation and a new facial recognition search technology. Also, the Web is rife with wild rumors about clandestine Google projects, so we asked the secretive company to comment on some of the more prominent ones to try to find out what's really going on.

Android

Although the "gPhone" never materialized, the company has been planning something better: an operating system for phones called Android. It's partly a direct competitor to Windows Mobile and partly an experiment in open-source development. Recently, the company held a contest for third-party developers to create innovative apps for Android. 1,700 programmers took up the challenge.

Street View on an Android phone suddenly becomes much more powerful because you can use it when you are standing on a street corner, trying to find an address.
Click to view larger image.

Examples from the contest include wayfinding apps that tap into the handheld's Global Positioning System chip. One application lets users find a taxi based on where they are. Another app lets users find their friends' locations and what they're doing and lets them create plans with them, with all the information tracked in real time. Some of these apps sounds a bit theoretical at this point -- the platform and phones will ship in the second half of 2008 -- but Google did post a PDF that shows the top 50 winners in the first round of the challenge, along with screenshots.

Erick Tseng, Android product manager, says it's a massive shift in thinking from the phone dictating what you can do to the device being open to any kind of content, service, provider and media.

"There are clear benefits to the ecosystem, not just [for] the users, but [also for] developers, carriers, providers," Tseng says. "Whatever phone you use today, think about the difficulty of getting content -- Android has unfettered access to content. You never have to think about, because I am on this service or this provider I can't get certain content."

Not everything has gone smoothly for Android, however. Charles Golvin, a Forrester Research Inc. analyst covering Android, says "I think the Android platform is a long-term play, and its short-term hiccups are no surprise. Google is intent on reaching consumers wherever they can, and it's clear that, while Internet use on mobile phones is still limited, it is the next venue where Google can expect to interact with its customers."

Facial recognition search

After measuring the facial characteristics of an image, you can find all versions of that image, including the most common photo and all variations.
Click to view larger image.

Image search is a burgeoning market that is woefully untapped. Today, when you type "Paris Hilton" at Google.com, you'll find images that other users have tagged. Yet tagging is a tedious process. At Flickr.com, for example, many images are left untagged, making it impossible to find them by searching. The more images stored without tags, the harder it is to find them.

At Google, new facial recognition technology will make it easier to find untagged images. Unlike the technology used for biometrics -- where you can pass through a security checkpoint when a video camera confirms your identity -- this image search is purely for finding the information you want.

"What Google did for text, we want to do for vision," says Shumeet Baluja, a Google research scientist. "We want to make images just as searchable and accessible as text."

Imagine this scenario: Five years from now, when all of your digital photos are stored online, you decide you want to search for pictures of your grandmother. With Google facial recognition technology, you might start with a source scan that measures the distance between the eyes, arrangement of nose, ears, eyes and other data. In seconds, you find every image you ever uploaded -- and any image stored anywhere online.

Language translation

The actual language-translation interface looks simplistic, but it is based on thousands of language-pair rules that require high processing power and complex programming techniques.
Click to view larger image.

Translation has been around for years, especially as part of search engines such as Alta Vista. Google has made progress with the vast number of languages it has made available for translation, including Russian, Arabic and the recent addition of Hindi. Another innovation is in researching the rules applied to machine translation based on cultural phenomena of languages, which requires a great deal of computer processing.

"The more rules used, the better the quality of the translation," says Franz Och, a Google machine translation research scientist. "If you want to perform an English-to-Hindi translation, for example -- which has a small subset of the language pairs [matching words] of French or Spanish -- the smaller the language, the more important machine translation becomes. Finnish is a challenging language because of the morphology. One word could have all kinds of information inherent to it. Other language translations are more complicated because there are so many differences between the languages. Nice languages with historic roots and similarities are easier, like French to English."