An ocean of writing, artwork, and graphics is being created and stored on "web 2.0" sites. Much of it exists to realize our desire to define ourselves. We self-identify (and self-promote) not only through the "bios" and "abouts," but also the various bits of media and social connections we collect.
Self-Portrait (http://www.turbulence.org/Works/self-portrait) makes literal the social networking tendency to define oneself in relationship to others. The project is an on-going search across Flickr.com (a photo sharing website) to find faces that match my own:
The search is done using facial recognition technology (a data mining and biometric analysis evocative of "homeland defense" surveillance techniques). During the past twenty-two months, 6.2 million photos have been analyzed by Self-Portrait (and 1.2 billion photos uploaded by Flickr users). The project's automation takes what is already a mechanical process (photography) and extends the machine’s reach to include editorial selection. If the facial recognition algorithm judges that a photo is a match, I appropriate it as a kind of self-portrait:
Fig. 4 Self-Portrait ('Erin Shows Me The Office' by Jennifer Daniel), 2008
There's something powerful about a random, live experience. It is this lure of the immediate that often has me wasting my time, flipping through the mediocre offerings of the half-dozen, static-filled TV channels I can get through my bunny-eared antennae even though there's a perfectly good Netflix DVD waiting to be viewed. There's no rewinding that broadcasted Jennifer Aniston movie--if I miss a line of dialogue, it's my own fault for having a too-weak bladder.
Similarly, I occasionally find myself mesmerized watching photos being analyzed on the Self-Portrait website. The experience has the attraction seeing images come and go with time's arrow, as well as the interactive pleasure of being able to dive in deeper if I spot something intriguing and can click quick enough.
Clicking a photograph on the site opens a window to the photographer's Flickr photo-stream and possibly satisfies the curiosity that caused the click: Where was that photo taken? Are there more photos of that cute person?
When I selected the Self-Portrait above (choosing it because it is a recent match and I liked how it is a photograph of a photograph, adding yet another step removed to the self-portraiture) I clicked on it to get the attribution information. In Jennifer Daniel's (the photographer's) profile, I noticed that she has the coolest URL I have ever seen:
Following that link, I found a graphic on her home page warning that blogging about your cats with your 37,000 Facebook friends is fake intimacy. Jennifer created a similar illustration (this one cautions that sitting in front of a computer encourages exhibitionism) for a New York Times book review of Lee Siegel's "Against the Machine." The review (http://www.nytimes.com/2008/02/03/books/review/Lanchester-t.html) quotes Siegel as writing that the Internet invites people to "carefully craft their privacy into a marketable, public style."
Privacy, exhibitionism, and voyeurism were among the issues I was contemplating when working on Self-Portrait. Flickr (with its comments, contacts, and "faves") certainly is a social network, but for many of its users the site is mainly a convenient place to store photographs and share them with friends. There probably isn't an expectation that the images will be systematically analyzed and appropriated.
Self-Portrait was intended to first search the Creative Commons (CC) licensed photographs on Flickr before moving on to ones that have All Rights Reserved. Creative Commons allow content creators to selectively waive some copyrights while retaining others. For example, Jennifer Daniel's image above is being used under an "Attribution-NonCommercial-NoDerivatives " license.
Jennifer Daniel is conscious of the panopticonical nature of the Internet. Her website includes the request, "I realize the Internet is infinite, but please do not reproduce in pixel or print without asking first." With that appeal in mind (and the fact that it's questionable whether my usage is really non-derivative), I contacted Jennifer to double-check before using her image for this essay. She graciously consented. However, my email may have caused her re-think the blanket permission given by Creative Commons licenses. Shortly after our email exchange, Daniel switched to having All Rights Reserved for future downloads of her Flickr images.
Several other self-portraits have had their Flickr files made "private" or their licenses withdrawn (since I originally downloaded the photos when they were CC licensed, I can continue using them). But due to a software glitch, there was one match made on an All Rights Reserved photograph in an early phase of the project:
Fig. 5 Detail of the website showing Self-Portrait ('DL 2005 136' by David Collier), 2006
I've always been particularly fond of this self-portrait and hesitated to ask for permission to use it and risk being told "no." But I don't really want to appropriate against a creator's wishes, and felt that the fact the subject is a child increased my obligation to ask. David Collier was happy for the photograph to be used and volunteered that he would make it Creative Commons licensed. I was struck by how my emails to Jennifer and David resulted in them taking opposite actions.
We're still grappling how the Internet makes our lives public and work acquirable. This very morning I came across my name and (no longer in service) telephone number listed on a website (http://www.userinstinct.net/address-18790-4294.htm). I filled out a form to remove my information and was told it would take 4 weeks. I was also directed to a paper (subtitled When Data Democratization Verges on Privacy Invasion) that explains why the listing is legal: http://writ.news.findlaw.com/ramasastry/20050512.html.
Fig 6 Self-Portrait ('IMG_9747.JPG' by Tyler Tate), 2006
The "brains" (soul?) of Self-Portrait is Verilook, a facial-recognition software development kit donated by Neurotechnology. Verilook determines matches by using pupil distances and face feature ratios. Or something like that. I have consciously avoided gaining too much insight into how the matches are determined.
Golan Levin once described generative art as an artist creating the illusion of an algorithm having control (CIAC Magazine, 2004). As much as possible, I want that illusion to be reality. I'd like the machine to escape my control and to surprise me with its creations. So I loosened up Verilook's standards to be more generous in making matches (mistakes are interesting) and embraced it as a black box into which photos are submitted and judgments (of "match" or "not a match") are retrieved.
Fig. 7 Self-Portrait ('Ritmo Latino' by Jedi RC), 2007
Perhaps there is an illusion here, but I don't think so. The illusion of control is when the stimulus is a MacGuffin--when it could really be anything, when the triggering data is really just a minor factor in comparison to the impact of how the programmer/artist frames the process for realizing the outcome. Self-Portrait's facial-recognition algorithm isn't creating, it's editing. The industrial nature of the facial-recognition is what makes it real. The algorithm is making sincere choices despite my hampering of its abilities (the equivalent, perhaps, of plying a photographer with several rounds of cocktails before having her/him do a final proof selection).
Fig. 8 Self-Portrait ('HDR Sunrise Walk' by Dale Furutani), 2006
I had assumed there would be an inhumanness to the matches--that perhaps the point of the piece would be the arbitrariness of our body images. It isn't uncommon for people to have difficulty distinguishing individuals of another ethnicity because of a habitual focus given to particular features' variations. I supposed that there would be a similar disconnect in how the algorithm coldly measures interocular distance (or whatever).
Fig. 9 Self-Portrait ('Heather and Patrick' by Ashley Richards), 2007
Perhaps it is a human tendency to see patterns, but I do think a similarity can be discerned in the selected self-portraits (one peer-reviewer has assured me that this is entirely in my imagination). There's a definite tendency for the matches to have geeky glasses. And while the facial recognition seems to ignore gender and hair length/style, the matches do tend to have long, narrow faces and pointed chins. Do the matches collectively represent me? Here's my face compared with a number of Self-Portraits superimposed together.
Or for a less processed look, here's a video of all the matches as of 14 August 2008:
Thanks to Turbulence.org and its co-directors Helen Thorington and Jo-Anne Green for commissioning Self-Portrait, the Jerome Foundation for funding, and Neurotechnology for donating the facial recognition API.
Copyright © New Media Caucus. All rights reserved.