As we speak, the new visual search tool it launched yesterday, Google Goggles, is scanning and rescanning a vitamin bottle I am trying to search. A blue scanning bar seems to be stuck re-reading the product. And from the looks of the results on the last dozen searches I performed upon loading the app onto my DROID, I must have broken something.
I was surprised to see Goggles launch even in a Google Labs beta. Recently under wraps altogether, the app was featured last week in a CNBC special on Google. In the context of this report, it seemed that Google itself was holding the visual search engine back from release because it suffered a lot of bad hits. My guess is that they decided the Android crowd is more amenable to beta testing an obviously imperfect product than the rest of the population -- and so they let it into the wild.
Whether that was a good idea remains to be seen. It is hard to tell of Goggles teases us effectively with the incredible promise of mobile visual search -- or if its many false hits and limitations does long-term damage. My vitamin bottle returned true Google gobbledygook. A scan of the UPC code on a Dasani water bottle returned links to a Wikipedia entry on the painting Venus and Amor. A head-on shot of the same bottle rendered the text "pnce cd Ughi" and offered links to a Beethoven piece and "Ughi" hotels. Just for kicks, and because augmented reality is one function of the program, I scanned the recent "Augmented Reality" issue of Esquire magazine and got in return "sired reality BEI" and links to a museum exhibit and MP3s with "reality" in the title.
And just because I love screwing with the Google engine when I get the chance, I scanned my Apple TV remote as well as the iPhone itself. Goggles surrendered on both counts and admitted it was not good at recognizing many object types. And it seems really bad at recognizing the competition.
Maybe I am being unfair. For straight-on scans of the cover of the "Abbey Road" album, the search results got me to the CD. For a Sims3 box, likewise it got me to an Amazon purchase page and to Electronic Arts' dedicated site. I was a little miffed it couldn't get magazine covers. Even Amazon's image search engine can do that.
My own reservations about the fugliness and the standardization woes over 2D coding naturally led me to wish for a purely visual search engine answer to the problem of using phones to knit the physical world with digital data. So far, visual search engines like the ones used in mobilized magazine campaigns and in the Amazon and Barnes & Noble app have benefited from their aiming for a narrow target. There is a limited database of images that the engine has to recognize. I am very happy with the results both apps give me in-store, and they have become a part of the shopping habit.
Google Goggles makes the unforced error of promising to index the world. Of course the program offers loads of caveats: that it is better at locations, books, contact info on a business card, logos, etc., than it is on general objects. Regardless, it leaves the user with a Forrest Gump box of chocolates. You never know what you are going to get.
Goggles is more interesting right now as an augmented reality engine. If I aim the DROID in landscape mode and rotate in place, it picks up about 20 locations, including the nearby shopping mall, some restaurants and some major franchises. Actually, Yelp!'s "Monocle" augmented reality mode has a better stock of local mom and pops, but ultimately a Google engine should be more versatile. Imagine being in a large shopping plaza and being able to hold up your phone and filter the augmented reality view of nearby store both by type and even by product. While the index right now is sparse, how long will it take for Goggles to leverage all of those street view images so that the locations of most major thoroughfares can be recognized?
The promise here is simply massive. The peril comes in alienating people from the concept of free-form visual search because the engine is running only on one cylinder.