that world models of reality are always incomplete and require constant refining [ 6]. This inherent challenge of attempting to model reality [ 7] demonstrates a further consequence of this approach: the loss of nuance. Complexity is flattened into models that a computer can track and understand. Lastly, these incomplete and simplified models, even if offered originally as approximation, take on a life of their own and, in a sense, become reality.

The Switcheroo in Action. Every day we encounter the implications of representation as reality. Anyone who has been denied insurance unjustly, been detained unnecessarily at cus-toms/immigration, and received unwanted direct mail has experienced a mismatch between a model and his own life. Abstract models serve a purpose. They

help us learn from singular cases and detect predictable patterns. They focus our attention on what is considered the most important information. But they also hold the seeds of reification. Once codified, whether in public policy or computational systems, models are inherently difficult to detect and modify.

Imagine, for example, a natural language processing application that parses text and measures it against a model of veracity. Managers could gauge their employees’ loyalty; an online dater could rest easier about her potential partner; and governments could add it to their intelligence arsenal. Sounds reasonable, depending on the model’s accuracy, who it is used on with what permission, and what is done with the information. And this is exactly the point. Abstract models seem benign in the

abstract, yet can be potentially troublesome in the living.

Values in Action. Let’s return to grandma. The proactive home can provide peace of mind and signal events of concern. There is something comforting about handing data over to smart machines that augment the reach of our senses and process enormous amounts of data in a manner seemingly free from human error. To be sure, the system will have false alarms and missed emergencies, for example if grandma falls in an unmonitored part of the house, thus requiring more and more sensors. The system will also introduce its own error rate due to technical limitations. But the success/fail-ure ratio of the proactive home is only one way of looking at it.

The system will also change the way grandma and her relatives live and relate to each

[ 6] Leahu, L., Sengers, P., & Mateas, M. “Interactionist AI and the Promise of UIbicomp, or, How to Put Your Box in the World Without Putting the World in Your Box.” Proceedings of Ubicomp 2008, ACM Press: 134–143.

[ 7] Rod Brooks has said, “The world is its own best representation.” See Brooks, R. A. “Intelligence without Representation.” Artificial Intelligence 47, 1-3 (1991): 139–159.

November + December 2009