Only a decade ago, preparing a lecture meant going to the university slide library to retrieve slides from an array of drawers and assemble them into a lecture on a light table. Today, most historians of architecture assemble digital images in presentation software and display them using digital projectors. Our first stop for gathering images is, likely as not, either Google Images or Yahoo!'s Flickr, the two search services dominating the global commons of images on the Internet.1
Although more specialized databases such as SAHARA or ARTstor exist for architectural historians, Google Images and Flickr have certain advantages. They have no barriers to entry; they allow users to restrict searches to Creative Commons––licensed images that can be legally reused and even modified; and, above all, they offer access to huge numbers of images——Google Images attempts to index the entire web while Flickr contains some four billion images uploaded by users. Although Yahoo! has an image search engine and Google owns Flickr's chief rival, Picasa, thus far Google Images and Flickr dominate.2
The fundamental difference between Google Images and Flickr stems from their origins. Google Images was created in 2001 as one of the first extensions of Google's web search service. Just as the mission of the corporation is "to organize the world's information and make it universally accessible and useful," the mission of Google Images is to "organize the world's images."3 Notwithstanding Google's informal corporate motto "Don't be evil," the stakes in this are high: in 2002 Google founder Larry Page told a Stanford class "If we solve search, that means you can answer any question, which means you can basically do anything."4
During the last decade, Google's chief advantage was its familiar interface, the number of images it indexes, as well as the speed and intelligence of the search engine that not only picks images from across the web but associates them with terms from the site it finds them on. With competition from image search services at Yahoo! and Microsoft Bing rising recently, Google introduced new ways to restrict results. After clicking on a "show options" link, users can restrict images to a certain size, type (face, photo, clip art, or line drawing), or specific color. Underneath some images, a "Find similar images" link permits users to take advantage of image recognition technology that Google developed to restrict results based on visual qualities. In addition, Google recently introduced Image Swirl, an experimental interface for classifying related images together (Figure 1). Conducting an Image Swirl search for a well-known building such as the Villa Savoye creates a number of stacks: a stack of plans, a stack of photos of models, a stack of exterior shots of the building, even a stack of book covers with the building on it. Clicking each stack reveals a swirling cluster of associated items: in this case plans, models, exterior shots, or book covers, respectively.
For its part, Flickr is as much a social networking site as a photo-sharing service, a characteristic it owes to its origins in Ludicorp's 2004 Game Neverending. Using Johan Huizinga's idea of Homo Ludens as its departure point, Ludicorp thought of Game Neverending as a vast multiplayer online platform for fostering interaction among users. To help achieve this, Ludicorp developed an instant-messaging chat room in which discussions would be based around photographs. Realizing that this interface would be more marketable than the game, Ludicorp shut down the game and opened Flickr. In building Flickr as a social networking site, the developers eschewed the photo-album metaphor that had previously dominated photo-sharing sites. The result gave users more flexibility, permitting them to make comments about photographs in threaded discussions, make friends, exchange mail, follow the photo streams of other users, tag images, and join and create groups.5
Groups can be very broad——the "architecture" group, for example, has 34,361 members and 531,298 items in its image pool——or very specific——"the real BAUHAUS" has 77 members and 216 photos, all dedicated to images of the building shot from the same vantage point of the west side of the faççade. Many group administrators seek out images to add to their group pool. Thus, even if the image owner doesn't add it to an appropriate group pool, a visitor will find it and add it to the pool.
Since Flickr hosts images on its site, it requires users to own the rights to images they upload and lets users specify licensing terms, thus making Creative Commons searches more accurate on Flickr than on Google, which merely assumes that if a page has such a tag or badge on it, it is licensed under the Creative Commons, even if the image is included not because the site's owner has the rights to it but because the site's owner is using it under the terms of fair use. Moreover, the use of Flickr by amateur and even professional photographers means that it hosts many highly technically proficient images. Downloading images from Flickr is just a matter of clicking on the appropriate image size. With Google Images, on the other hand, searchers seeking more than a thumbnail must visit the original website to download the image. Still, Flickr's requirement for ownership restricts the sort of material that can be hosted, and what Google Images lacks in a general level of quality it makes up in sources. It links to images from official websites dedicated to buildings or architects, periodicals, and books that are generally not available on Flickr. Thus, Google Images almost always returns better drawings and plans than Flickr. At the same time, the images on Flickr form a visual record of the public engagement with buildings, giving the opportunity to conduct research that uses this data empirically. For example, a recent study by researchers in Cornell University's Department of Computer Science determined that the Apple Store on Fifth Avenue is the fifth most photographed site in New York City, ahead of the Empire State Building.6
Flickr makes such research easier with a more open Application Programming Interface (API), which allows developers to build applications for the web, desktop or tablet computers, smart phones, and other devices. Such applications can upload images to Flickr or act as "mashups" providing alternative means of searching content as well as filters, and remixes of content on the site. Dozens of applications have been built, allowing users to generate authorless films from Flickr content, explore Flickr over Google Maps, view images in a screensaver, and so on. Thus far these have been largely devoted to entertainment and are of little use to scholars, but it seems only a matter of time before scholars devise projects that draw on the data on Flickr.7
Where present, Flickr extracts metadata about the image from the uploaded file. Often users can access not only the time an image was uploaded but the time it was taken, together with information such as the camera used to record the image, the focal length of the lens, the length of exposure, even the software used to process and upload the image, information of moderate utility for photographers.
Flickr lets users browse for images on a world map. Camera phones, a handful of GPS-enabled cameras, and Eye-Fi Explore SD cards automatically add geographic coordinates to image metadata, but users can also manually add tags to images that do not have embedded geographic data. Assuming the data is accurate, the Flickr map has the potential to give a better sense of the context within which a building is located.
Instead of a controlled vocabulary like the Getty's Art and Architecture Thesaurus, Flickr employs a folksonomy system, in which users——not experts——tag images with metadata annotations of their own choosing. Once tags are established by users——for example "villasavoye" (spaces are not allowed)——they allow similar images to be tagged and retrieved easily. Registered users can "follow" or subscribe to photo streams of tags that interest them. Flickr extends this with "clusters" that suggest related terms. In the case of "villasavoye," the system returns "lecorbusier," "poissy," "internationalstyle," "architecture" and "france" as related tags. Advocates of folksonomy argue that it is more efficient, more flexible, and more democratic that a taxonomic hierarchy. Still, folksonomy tags can sometimes be arbitrary or applied inappropriately.
For its part, Google is also starting to add labels to images through Google Image Labeler, a game in which users are matched with other anonymous participants. When both players agree on a tag for an image, they win a point. The result, Google hopes, will be less arbitrary user-generated tags, but the anonymous nature of this system suggests that it may be less successful than Flickr.
The nature of image retrieval has irrevocably changed. The walls of the slide library defined a canon, selected by visual resource professionals together with members of the faculty, and painstakingly reproduced, mounted in glass, and hand labeled. Beyond its aspirations toward a universal canon, each collection reflected an interpretation of the canon within the institution where it was situated. Further context emerged from the taxonomic system used to classify the slides. Based on some combination of architect, building, geography, and building type, these systems were often unique to the collection.
No matter how problematic, the canon still provided a frame of reference and gave a system of value to the images. If taxonomic classification was restrictive, it was a tool that visual resource professionals built to search and make associations. In both Google Images and Flickr, this has been lost. Images are disconnected from any context, floating on the glowing white space of the web browser. Take Google's Life magazine photo archive, for example, which is composed of some ten million photos. Even though Google now has the entire run of the magazine from 1936 to 1972 scanned into Google books, there is no link between the photo archive and the magazine.
The availability of images in unprecedented and unmanageable numbers is both a blessing and a curse. For example, Google Images returns 31,500 images for "Villa Savoye" while Flickr returns 6,848 images. In the case of Tatlin's Monument to the Third International, Google Images returns 3,140 images and Flickr only 44, some of which are non sequiturs. As there is no controlled vocabulary, a savvy searcher will need to try multiple search terms: abbreviating the search to "Tatlin Tower" returns 14,700 images on Google and 116 on Flickr, including a series of photos of an unusual wedding cake shaped like the monument.
But this sea of images is not without an organization of its own, produced by the vast social media game that makes up the internet. If the endless trade of images suggests the undoing of the canon, we should keep in mind that networks generally follow a power-law curve: most links go to only a few sites. Should this hold true in image search, then the starchitecture of the last fifteen years may still have legs.8 If, not so long ago, the success of a commission was judged by its appearance in newspapers, perhaps in the near future the success of a commission will be judged by the number of images that an online search retrieves.
There is little question that this endless flow of images removes us further from the physicality of the building. And yet, isn't this the longue duréée of modernity at work? Yet again all that is solid melts into air. Having had a century-long run, the slide is itself history now. Instead of either lamenting or blindly rushing into the new medium, our task is to follow Heinrich Wöölfflin's example and identify the methods of historical analysis immanent in it.
According to ranking site Alexa.com, which does not provide separate statistics for Google Images, Google is the most popular website in the world and Flickr is number 33. Alexa.com, "Top Sites. The Top 500 Sites in the World," http://www.alexa.com/topsites/global;1 (accessed 30 July 2010).
Google no longer publicly discloses the size of its indexes. I performed a search query for -'.$&%!@?* on Google Images and retrieved a result of 1.3 billion images. Even if some images on Flickr are marked private, such a low number would suggest that either Google is unable to index all the images from Flickr or that there is a limit to the maximum number of images it will retrieve. I derived this method of estimating the size of the database from Loren Baker, "How Big is Google's Index," Search Engine Journal,http://www.searchenginejournal.com/how-big-is-googles-index/4174/ (accessed 30 July 2010). See also this statement from Google: "We Knew the Web Was Big," http://googleblog.blogspot.com/2008/07/we-knew-web-was-big.html (accessed 30 July 2010). On the size of Flickr's database see Christina Warren, "Flickr: 4,000,000,000 Photos and Counting," Mashable. The Social Media Guide, http://mashable.com/2009/10/12/flickr-4-billion/ (accessed 30 July 2010). Warren concludes that the social media website Facebook is the largest photo database on the web, with 15 billion as of April 2009.
Larry Page, quoted in Ken Auletta, Googled: The End of the World as We Know It (New York: Penguin, 2009), 94.
Google Friends Newsletter, 28 July 2001, http://www.google.com/googlefriends/jul2001.html (accessed 30 July 2010), Jesse James Garrett, "An Interview with Flickr's Eric Costello," Adaptive Path, http://www.adaptivepath.com/ideas/essays/archives/000519.php (accessed 30 July 2010) and ITC Conversations, "Caterina Fake; Co-Founder, Flickr; The History of Flickr," http://itc.conversationsnetwork.org/shows/detail1755.html# (accessed 30 July 2010).
David Crandall, Lars Backstrom, Daniel Huttenlocher and Jon Kleinberg, "Mapping the World's Photos," WWW '09: Proceedings of the 18th International Conference on the World Wide Web (2009), 761––70, http://doi.acm.org/10.1145/1526709.1526812 (accessed 30 July 2010).
Similar work on the analysis of very large datasets of imagery from the history of art is already undertaken by Lev Manovich at the Center for Software Studies based at the University of California, San Diego. See "The Software Studies Initiative," http://lab.softwarestudies.com/2007/05/about-software-studies-ucsd.html (accessed 30 July 2010).
See Clay Shirky, Clay Shirky's Writings about the Internet, "Power Laws, Weblogs and Inequality," http://www.shirky.com/writings/powerlaw_weblog.html (accessed 30 July 2010). I am indebted to Gideon Fink Shapiro for this insight.