Thesis 41

Before they are knit together, the systems that comprise everyware may appear to be relatively conventional, with well-understood interfaces and affordances. When interconnected, they will assuredly interact in emergent and unpredictable ways.

Our information technology is difficult to consider holistically, because it is modular: Though its constituents work together, they are designed at different times, in different places, by different parties, to different ends. Furthermore, these constituents are agnostic as to the circumstances of their use: A database doesn't "care" whether it's used to enhance a rental car company's relationships with its best customers, manage a public housing authority's inventory of cleaning supplies, or keep tabs on the members of a particular ethnic group scattered through a larger population.

This modularity has historically been the very strength of information technology, at least in the years since the Internet raised notions of standardization and interoperability to wide currency in the industry. It lends our informatic systems an enormous degree of flexibility, adaptability, even vigor, and it will and should continue to do so.

But in our experiences of everyware, we'll also find that modularity will occasionally prove to be the cause of concern. Even where we recognize that certain high-level effects of ubiquitous systems are less comfortable for users than we might like, it will generally not be possible to address these effects at the level of component artifacts. Apart from those of literal imperceptibility, almost all of the issues we're interested inthe surfacing of previously latent information, the persistence lent ephemera by their encoding in mnemotechnical systems, certainly seamlessnessdo not arise as a result of properties residing in components themselves. They are, rather, an emergent property of the components' interaction, of their deployment in the world in specific patterns and combinations.

This is a thoroughly conventional microcontroller. This is an off-the-shelf microphone. This is a stock door-actuator mechanism. We think we know them thoroughly, understand their properties and characteristics in detail, and under most circumstances, that is a reasonably safe assumption. But put them together properly, embed them in the world in a particular relationship with each other, and we have a system that closes a door between two spaces any time the sound level in one of them breaches a certain threshold. And though the exact details will vary depending on where and how it is deployed, this system will have behaviors and consequences that absolutely could not have been predicted by someone considering the components beforehandno matter how expert or insightful that person might be.

Maybe this strikes you as trivial. Consider, then, a somewhat more baroque but no less credible scenario: Contributions to political campaigns, at least in the United States, are already a matter of public record, stored in databases that are easily accessible on the Internet.^[*] And we've known from as far back as Smart Floor that unconscious aspects of a person's gaitthings like weight distribution and tread periodcan not only be sensed, but can also serve as a surprisingly accurate identifier, at least when a great many data points are gathered. (As the Georgia Tech team responsible for Smart Floor rather disingenuously asked, "Why not make the floor 'smart,' and use it to identify and track people?")

^[*] The source I happen to be thinking of is opensecrets.org, but there are many others. See spywareguide.com for some interesting mash-ups and combinations.

On the surface, these seem like two thoroughly unrelated factoids about the world we live in, so much so that the third sentence of the paragraph above almost reads like a non sequitur. But connect the two discrete databases, design software that draws inferences from the appearance of certain patterns of factas our relational technology certainly allows us to doand we have a situation where you can be identified by name and likely political sympathy as you walk through a space provisioned with the necessary sensors.

Did anyone intend this? Of course notat least, we can assume that the original designers of each separate system did not. But when things like sensors and databases are networked and interoperable, agnostic and freely available, it is a straightforward matter to combine them to produce effects unforeseen by their creators.

The Smart Floor example is, of course, deliberately provocative; there's nothing in the literature to suggest that Georgia Tech's gait-recognition system was ever scaled up to work in high-traffic public spaces, let alone that any spaces beyond their own lab were ever instrumented in this way. Nevertheless, there is nothing in the scenario that could not in principle be done tomorrow.

We should never make the mistake of believing, as designers, users or policymakers, that we understand exactly what we're dealing with in an abstract discussion of everyware. How can we fully understand, let alone propose to regulate, a technology whose important consequences may only arise combinatorially as a result of its specific placement in the world?

I believe that in the fullness of time, the emergent behavior of ubiquitous systems will present us and our societies with the deepest of challenges. Such behavior will raise the specter of autonomous artifacts, even call into question the proper degree of loyalty an object should have toward its owners or users. As real as these issues are, though, we're not quite there yet; perhaps it would be wiser to deal with the foreseeable implications of near-term systems before addressing the problems that await us a few steps further out.