That might sound at first like a geek elitist position, implying that everyone should be a programmer and that those who can’t program are [lazy/stupid/inferior].  I can’t speak for the authors of that wiki article, but my point here is not to suggest that everyone learn to program, but rather that perhaps everyone should learn about programming. Think of software literacy as an extension of media literacy. People are (hopefully) taught to detect bias in newspapers and television – even if they don’t know how to produce a newspaper or television program themselves. Now that software is a mass medium - one that influences people’s lives at both consumer and institutional levels - might not it be useful if people learned to detect software’s biases?

How is software subjective?

Example 1: Google, whose search results significantly influence the information people access, touts the objectivity of their PageRank technology:
http://www.google.com/technology/index.html

PageRank relies on the uniquely democratic nature of the web by using its vast link structure as an indicator of an individual page's value. In essence, Google interprets a link from page A to page B as a vote, by page A, for page B. But, Google looks at more than the sheer volume of votes, or links a page receives; it also analyzes the page that casts the vote. Votes cast by pages that are themselves "important" weigh more heavily and help to make other pages "important."

I’d argue that the algorithm described isn’t “democratic” but is actually rather similar to becoming popular in high school. If the popular kids like you then you can easily become popular. But what if you’re not part of the in-crowd? What if you’re a dissenter – or just not trendy? According to the algorithm described above, it’s difficult to get noticed.  Google apparently refines the PageRank algorithm on a regular basis, and they keep its exact workings a secret. (If they didn’t, it’s likely we’d all see even more ads than we do for products that begin with a “V” and end with an “a.”) But at least we can begin to critically question how PageRank influences the information we read. And even though Google assures us that “Google's complex, automated methods make human tampering with our results extremely difficult,” we can keep in mind that humans determined the automated methods in the first place.

Example 2:  The United States Internal Revenue Service was recently criticized for freezing the tax refunds of many poor taxpayers by targeting their returns as likely to be fraudulent – even though most were not:
http://www.nytimes.com/2006/01/10/business/10cnd-tax.html?ei=5090&en=63292ea6712adf26&ex=1294549200&adxnnl=1&partner=rssuserland

A computer program selected the returns as part of the questionable refund program run by the criminal investigation division of the Internal Revenue Service.

The article doesn’t tell us any more than that about the computer program, but obviously someone programmed it with rules for finding a “questionable” return. Clearly, those rules were subjective, and they seem suspiciously like they may have been politically motivated. The fact that the deed itself was done by computer doesn’t make the decision “blind” or “objective.”  In a software-literate culture, the journalist who wrote the article might be expected to press for details on how the program worked, or at least discuss his inability to obtain this information from his sources.

I’m not myself today…

If we say someone “matches” a terrorist (or anything else) - what does it really mean? Some characteristics of that person’s face have been determined to be significant – they match the terrorist face more closely than others in the database. This raises the question – what are these “significant” characteristics?

For example, on  http://www.cs.princeton.edu/~cdecoro/eigenfaces/ - images document the results of an attempt to use computer vision algorithms to match photographs of individuals with those in a database. In the second photo from the top of the web page, we see that the algorithm detected the correct person from the database in a large percentage of cases. However, the few incorrect cases are interesting. The software attempted to detect similarity between photographs and faces - and it did so - according to some characteristics. Not the ones that would have given the “right” answer and identified the same person. But, the wrong answers may not be what we expected. Instead of confusing people of the same race, for example, the software will sometimes confuse two people with a smug expression on their face. Maybe in some ways smug people have more in common than people of the same race. Maybe, on days when you’re not yourself, you’re really more like someone else. In any case, attitude profiling may turn out to be a greater risk of technology than racial profiling.

But profiling concerns aren't limited to race. If the computer vision bogeyman were used to identify “undesirables,” what would those undesirables look like? Presumably, everyone could envision their own profile of an "undesirable." And in fact, such profiling could be programmed into a computer vision system. But – the profiles would need to be quantified for the computer. It turns out, computers are subject to the same type of stereotyping as humans are – only more so. For example, say you’re on the lookout for troublemaking emo kids. You could tell a human, “Watch out for emo kids,” and this would be asking the human to stereotype. But you’d have to tell the software, “Detect people wearing all black, with pale skin and very black hair.” This is more extreme stereotyping than the human would do (at least consciously.) But of course, humans chose those characteristics.

So – one of SVEN’s aims is to reflect on the human subjectivity inherent in technology. Because this subjectivity must be reduced to objective rules, such implementations obviously have limitations in mimicking the way humans would perform the intended task. However, these implementations and their results can, through their limitations and exaggerations, reveal less obvious things about how their human creators "see" things - and about humans in general. Technological development expects machines to think like humans and humans to think like machines - under this stress both give something about themselves away.

Technology and the way it’s used aren’t the same thing.

This might seem an obvious point, but the opportunities it presents for tactical software might easily be overlooked. Take for example, computer vision surveillance technology. It conjures up depressing connotations, and our gut reaction is to respond to it with resistance. That’s because we’re used to it being used to find when someone looks, in someone else’s judgment, well…  bad….  But  that’s not necessarily the case.  Why limit ourselves to defensive positions against “scary” technologies?  Why not take some offensive ones? If computer vision can determine when we look bad, we can develop some computer vision technology that can figure out when we look good. And who looks better than… rock stars?

Other video surveillance-related tactical media projects:
http://www.notbored.org/the-scp.html
http://www.tacticalmagic.org/CTM/project pages/TICU.htm

Other projects dealing with software subjectivity:
http://reamweaver.com
http://sinister-network.com

Other computer vision research that looks at human desirability:
http://www.eecs.berkeley.edu/~ryanw/research/tr_hot.html

Another text about software subjectivity:
http://art.runme.org/1107805961-9560-0/gonzalez.pdf

A text from 1987 about the then-new Macintosh as a black box:
http://www.williambowles.info/sa/maccrit.html

More software art and theory than you can shake a stick at:
http://runme.org