It was another engaging day at UbiComp 2006, with a number of good paper presentations (of which, again, I only saw half), videos, an innovative open session and a town hall meeting.
Tim Sohn, from UCSD, started things off with a nice presentation on "Mobility Detection Using Everyday GSM Traces". Nearly every talk I've seen here raises existential questions (at least for me), and for this one, it was "what is stationary?". Tim and his 9 colleagues (tied for the largest number of authors of any paper at UbiComp this year) developed an algorithm for GSM radio fingerprinting in which Euclidean distance was used to calculate similarities in signal readings, and three clusters were identified, representing stationary (with 95.4% precision and 92.5% recall), walking (70.2% precision /80.0% recall), and driving (84.3% precision / 81.7% recall). A second experiment, which compared GSM data to pedometer data showed 90% accuracy between the two data collection methods.
Mike Chen, from Intel Research Seattle, talked about work he had done with 9 other colleagues (including Tim, and a number of other co-authors on the first paper) on "Practical Metropolitan-scale Positioning for GSM Phones". The team configured a ThinkPad with 2 GPS receivers, 3 GSM modems (AT&T, Cingular and T-Mobile), 3 GSM phones and 2 WiFi antennas, and collected 43500 km of wardriving data in the metropolitan Seattle area over 208 hours. 3 classes of positioning algorithms were tested: centroid, fingerprinting (like previous talk) and Gaussian. A cool video was created to show the results, which averaged out to 94m of accuracy downtown and 194m in residential areas. The code is available at http://pols.sourceforge.net.
John Krumm, of Microsoft Research, gave what I consider the best presentation of the day, on "Predestination: Inferring Destinations from Partial Trajectories", in which features such as terrain, prior destinations, trip times and rational actor assumption of efficient driving were fused together to estimate future destination probabilities, resulting in a median error rate of 2 km (based on GPS data) at the halfway point of trips. In addition to an interesting approach and findings, John gave us an opportunity to practice continuous distributed attention by sprinkling liberal product placements both on his slides and in "real life" (taking a sip from a bottle of Coke while a slide with a Coke log was being shown), suggesting the types of contextual advertising that may be possible with better predestination estimation.
I'm glad to see progress being made on positioning technologies and techniques, but the question I have for all of this research is how good is good enough -- how accurate do positional systems have to be in order to be useful / acceptable? I'm reminded of the expression "close isn't good enough, except with horseshoes hand grenades" ... I wonder if / when we'll be able to add more, er, applications to that mix.
The second session was devoted to the video program. Several videos were quite good, but I'm only going to mention one here: "Spalogue: Designing Men-Women Communication in a Public Bath", presented by Saiki Ito and Mariko Koizumi. This video showed a system that could detect gesture writing with a finger in a spa, translate that into characters that are then projected onto a wall in the spa; if a participant of the opposite sex in a separate spa chooses to write back, the ink appearing on both walls becomes transparent in the overlapping areas as the response is "written" over the original text. The video did a nice job of invoking a number of cultural customs, such as Japanese poetic communication style (“WAKA”), bridging the gaps between cultural epochs while bridging the gap between the sexes, playing the edges of [con]sensual revelation in an engaging way.
[For reasons I don't understand, Internet Explorer was self-destructing on this post when it included a fifth photo, so I deleted it -- the fifth photo and lots more information is available on the Spalogue web site.]
Thomas Hansen, from Aarhus University, presented "Experiences from Real-world Deployment of Context-Aware Technologies in a Hospital Environment" [slides], where 3 sections of a hospital (including is operating theatres) was instrumented with an inexpensive location tracking system based on [Nokia] Bluetooth phones and PC dongles, context aware infrastructure, a number of displays. At the beginning of each shift, staff members would pick up a phone and associate it with themselves at a kiosk; they could then be tracked throughout the three sections of the hospital and the system could incorporate real-time awareness within its scheduling support tools. The system was to be deployed for a 3 month pilot study, but the users liked it so much they asked to keep using it, and it is still running, and Thomas presented data and analysis about how use had evolved over time. Interestingly, they found that privacy not an issue, due to the perception of high benefit -- it was useful -- and low cost -- it only covered pat of the building, kept no history, and perhaps most importantly, there was a high degree of trust among the staff. More information on the project can be found here.
Jon Froehlich, of the University of Washington, presented "Voting With Your Feet: An Investigative Study of the Relationship Between Place Visit Behavior and Preference", in which GSM tracking data was compared to explicit ratings of places to assess whether place visit behavior represents an implicit expression of interest or preference. Jon and his colleagues conducted a study where 16 participants agreed to have their GSM phones tracked over 4 weeks, during the course of which they provided 11 in situ survey responses per day ("where are you, what are you doing?") and more extended web survey responses 3-4 times per week. The hypotheses were that frequency of visits indicates preference, and that amount of effort (e.g., travel time) indicates preference. Interestingly, while I share these hypotheses, the data offered only weak support for the first, and no support for the second. Jon suggested a number of potentially confounding factors, and some future work trajectories that may help shed more light on implicit preference detection.
Yvonne Rogers, soon-to-be formerly-associated-with Indiana University, gave the most provocative presentation of the day, on "Moving on From Weiser's Vision of Calm Computing: Engaging UbiComp Experiences", in which she revisited the original vision of Mark Weiser for calm computing, reviewed some of the ways ubicomp has attempted to achieve that vision, and raised serious questions about the capability -- and desirability -- of computers to act on our behalf (behalves?). I was reminded of the distinction between "strong AI", which seeks (sought?) to imbue computers with intelligence so that they could replace humans, vs. "weak AI" which seeks to enable computers to augment humans. Adam Greenfield was invoked, yet again, in observing that much ubicomp can be characterized as "daring to intervene, clumsily, in situations that already work reasonably well. Yvonne issued a call for a Kuhnian-level shift from calm technology to engaging technology, requiring a broader scope and new agendas, themes, questions, frameworks … and adjectives. Ubicomp should be exciting, provocative, stimulating, visible, engaging, playful and even uncomfortable, enabling people to be active creative and reflective I their work, learning and living. Amen.
Barry Brown introduced the Open Session, which was an experiment in which authors submitted their work to a public forum, and the reviews were also publicly posted, and reviewers publicly identified. 10 papers were submitted, and 4 were accepted, based on 47 reviews, and the presentations succeeded in stretching the boundaries of "acceptability" at the conference (which is a good thing, in my opinion).
John Light, of Intel Research, talked about Intimate Computing, raising questions about the relative importance of emotional vs. functional aspects of [perceptions of] technologies, and applying The Media Equation to ubicomp ("the ubicomp equation"?). Louise Barkhuus, of University of Glasgow, talked about "Shakra: Sharing and Motivating Awareness of Everyday Activity": ways and means of combining social practices with technology for enabling groups to support -- and/or compete with -- each other in getting their daily doses of moderate physical activity. Haiyan Zhang, of the Interactive Design Institute Ivrea, presented Control Freaks (more here), clamps and stuffed animals with accelerometers and wireless controllers, that can be attached to everyday objects, e.g., study tables or subway car bobsleds, to create games in new places, another example of the growing pervasiveness and permeability of games worlds. I felt Haiyan's work, in particular, exemplified the potential for engagement Yvonne talked about, and highlighted the value for this kind of forum within the conference.
There was a Town Hall meeting immediately after the Open Session, in which a variety of topics were discussed. The most interesting aspect (to me) was Gregory Abowd's proposal that we consider using an open review process for the entire conference for UbiComp 2007. Albrecht Schmidt [later] raised some thorny issues about intellectual property and confidentiality of submissions that may pose challenges in making this work. However, I think that this is an exciting idea, and with the proper constraints (e.g., some kind of oversight by a trusted committee), could make for a grand experiment in engaging the community in creating a more engaging vision and practice of ubiquitous computing.
[Update, 2006-11-28: added new links to papers now available on the web]