In my UbiComp06 Doggy Bag

The notes that made into my paper and digital moleskines during last week’s UbiComp06 conference at the O.C.

My first participation to ubicomp reaffirmed my conviction that the field is still very much at its infancy and looking for itself. What more can we expect from a 20 years young research field. There are still debates on the definition and categorization of ubiquitous computing. In his conference notes Joe McCarthy (Nokia Resarch, Palo Alto) notes that the field seems to be moving beyond “technology in search of a problem” to a “technology in search of an application” within a problem domain. From the many papers on the subjects location, location and location, it seems to be agreed as a problem area. I would comment that niche ubicomp applications exist (object tracking, fleet/emergency management, health care). However, we are still considering ubicomp systems could have a more general direct profitable impact on people.

The mix of researchers, designers, hackers, and sci-fi writers, makes ubicomp probably one of the most dynamic and exciting computer science field to do research in. However, I hope the community will be able to move beyond the now yearly classic (yet imaginative) “ubicomp umbrellas” into more defined research approaches, methodologies and methods. In that sense, the quali-quanti discussion (replacing the highly expected Genevieve Bell (Intel Research)’s No more SMS from Jesus talk) was precious to me. Marc Davis (Yahoo! Research, Berkeley) mentioned that we now are able to access an incredible quantity of data (ranging from interview to logged actions) that allow us to gain information about different layers: from micro scale cognitive insights to large group processes (e.g. social groups, national issues). He coined this mixed qualitative and quantitive method as the new “computational social science” or the “new social science of the 21st century”.

Papers
The presentations related to my work covered the themes of positioning, activity inferencing (from location data), seamful design, mobility and user studies.

Activity inference was presented in several forms. Timothy Sohn (UC San Diego) demonstrated in Mobility Detection Using Everyday GSM Traces the ability to recognize mobility modes from GSM traces. The logs generated by the everyday lives of three data collectors over a period of one month, yielding an overall average accuracy of 85%, and a daily step count number that reasonably approximates the numbers determined by several commercial pedometers. He interpreted motion detection by interpreting changes in the set of nearby towers and signal strengths as indicative of motion. In order to do so, he used the Euclidean distance values between consecutive GSM measurements. In Practical Metropolitan-Scale Positioning for GSM Phones, Mike Chen (Intel Research, Seattle) showed that existing GSM devices can achieve a positioning accuracy with a median error of 94-196 meters. He observed that the positioning accuracy varies significantly across algorithms (centroid, fingerprinting and gossian processes), by a factor of almost 4. John Krumm (Microsoft Research) introduced in Predestination: Inferring Destinations from Partial Trajectories an open-wold model to evaluate driving destinations (closed world model would be based on destination people have already gone before). Data from the Natural Household Transportaton Survey were used to define a probabilistic grid of destinations and efficient driving likelyhood. This study shows that destination tend to cluster in steady state over a couple of weeks. The best performance on 3667 different driving trips gave a median error of two kilometers at the trip’s halfway point. This works did not show any direct application and Krumm presented it as an anticipation of future LBS (mobile marketing?). In a comment, Saadi Lahlou, (EDF R&D) mentioned that in this work we are dealing here with privacy on the future and not only on the past. Jon Froehlich (UC Irvine) Voting With Your Feet: An Investigative Study of the Relationship Between Place Visit Behavior and Preference presented a study on implicit indicators of preferences that showed that the travel effort as an aspect of preference on the places people go to. The study used two experience sampling method (EMS) triggers (contextual based on the GSM signal and a random time trigger). The results show that, first, sensor-triggered experience sampling is a useful methodology for collecting targeted information in situ. Moreover, there exist positive correlations between place preference and automatically detectable features like visit frequency and travel time. The most suprising piece of work on location detection was probably Shwetak Patel‘s (Georgia Tech) PowerLine Positioning. His system uses 2 modules installed on a house electrical power system. The modules send low frequency signals back and forth. Receivers can detect the strength of the signals and a fingerprinting algorithm is used to estimate a position based on those signals. The approach achieves an accuracy distribution of 92% for 3 meter regions, 67% for 1 meter regions, and 42% for 0.5 meter regions.

On the seamful design subject, it was not Gregor Broll (Ludwig-Maximilians-University Munich) but Leif Oppermann instead( Mixed Reality Lab at the University of Nottingham) who presented Tycoon (map zone with events). He showed (later also presented by Matthew Chalmers) a video, visualizing GPS shadows prediction, made by Anthony Steed (University College London) as part of the Equator project . Tycoon visualizes the invisible at authoring time. Not only does it consider the physical and content layers but also the infrastructure layer. Part of the work is based on the assumption that data will never be complete and complitely accurate.

Enrico Rukzio (from Albrecht Schmidt‘s Research Group Embedded Interaction, Ludwig-Maximilians-University Munich) presented a user study on physical mobile interaction techniques (i.e. interaction with things and smart objects). The investigated interaction techniques include touching, scanning, pointing. He used a user-centered design approach (analysis, low-fidelity prototype, high-fidelity, user study with 20 participants). As a result, users preferred touching: good, then pointing:average, and finally scanning: bad. He added that interaction depends on location, activity and motivation. Current location seems to be the most important. Enriko’s next step might be to investigate further the location aspect.

Eamonn O’Neill and Vassilis Kostakos (University of Bath, UK) presented a study (paper) on the relation between the architectural space and people mobility. An investigation of ubicomp on the people. They used BT scanning at severy key points Bath. They described a combined scanning for discoverable Bluetooth devices with gate counts and static snapshots. Apparently, 7.5% of pedestrian had discoverabl BT devices. I particularly enjoyed their analysis and visualization of the data and visualizations. This is part of the Cityware project.

Panel
W. Keith Edwards (Georgia Tech) organized a panel on “Interaction and Infrastructure: Crossing the Divide in Ubicomp Research”:

Mark Ackermann (University of Michigan) mentionned the difference between technically working and socialy workable. A subject he covers in his paper The Intellectual Challenge of CSCW: The Gap Between Social Requirements and Technical Feasibility. How and when sharing information changes behavior. That is people have very nuanced behaviour concerning how and with whom they wish to share information. We should consider social activity as fluid and nuanced (Erving Goffman), socio-historical social constructions that guide action (Harold Garfinkel) invisible technologies (Neil Postman), and bricolage (Edwin Hutchins). Moreover, wee should make the following social assumptions: roles are informal and fluide, exception is the rule, people are used to resolve breackdowns. Mark also mentioned darwinian co-evolution. That is, people not only adapt to their systems, they adapt their systems to their needs. The final message was a request to break open the infrastructure and interaction and consider what we cannot do (gap between what we do and what we know).

AJ Brush (Microsoft Research) ranted that ubicomp should be designed for the uninterested (!= dumb) user and the facilitate the management by professionals.

Anthony La Marca (Intel Research, Seattle) refered to the software making essay by Richard Gabriel on “worse is better“. In our context, the worst is when there is not interaction at all, it means failure. Simplicity of both interface and implementation is more important than correctness, consistency and completness. Currently we have barley functionning version and we should resist trying for the perfect solution (let’s throw more gas on the fire). Something to which Barry Brown (University of Glasgow) added that we in (ubicomp) research are obsesed with complexity. Matthew Chalmers (University of Glasgow) agreed that if you look fro details, you’ll find them. Decision on what complexity you want to ignore. Mark Ackermann

Matthew Chalmers (University of Glasgow) made a talk on seamful design (i.e. when the infrastructure becomes visible). The base of this approach is that systems will never fit to the context of use. Persistently used in a visible way (availability, accuracy, software functionality).
In the Q&A, Joseph ‘Jofish’ Kaye (Microsoft Research) mentioned not mention was made to ubicomp outside of the developed countries, and that no scenarios mentioned 3rd world’s needs. It is true that infrastructures dealing with intermittent connectivity such as DakNet are rarely taken into consideration.

Keynotes
Sci-Fi writer Bruce Sterling presented his Spime Meme map made of atom and bits recyling cycles/loops. His talk did not differ much from the one at LIFT earlier this year at the exception of a rant on panpsychism (objects are alive and smart). Technology is not magic and we should not be lying to people. A viewpoint I share (“AI is a God in a box”). I could not write down/remember everything, but a podcast is available from the LUCI blog.

Designer/research/teacher Brenda Laurel thoughts on Designed Animism were harder to grasp. Partially due to some references I was not familiar with and also due to her rhetoric not very directed to a globish audience. Nevertheless I enjoyed her humanist vision. Ubicomp should make us more closer and aware of nature. She concluded that “If I had more sensors, my body could be the world”.

I find attractive pervasive technologies that provide self/mutual/social/nature awareness (keywords: mirroring). Unlike prozac-like environment (keywords: seamfullness, calm) that are about “monitoring” or “motivating behavior change” or “positive reinforcement”.

Relation to my thesis: My poster triggered interesting discussions. Uncertainty is an old insovable? problem. Andrew L Kun (University of New Hampshire) on the design and deployment of location-aware systems, Mike Chen (Intel Research) on mobility inference with GSM networks. Jae-Woo Chung (MIT Media Lab) on GPS and LBS in Korea and others on uncertainty is an old insovable? problem and on my contribution comparing to Can You See Me Now.

Except the Tycoon presentation, some references made by Enirco Rukzi and Matthew Chalmers talk I have not noticed more work on the missmatch between the virtual and phisical (except Brol). Works on mobility inferences mention results in % of precision with a certain range of accuracy. I question “What do with the 10%… when do they happen and how do they impact the experience of a location-aware system”. In Joe McCarthy (Intel Research) analyses it as follow:

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

Finally, mobility/activity inference at large-scale is something that I have been looking at the last 12 months from a Transportation Research perspective and in a project at Simpliquity. From what I have seen from Intel Research, Seattle and Micfosoft Research’s works, the collection, analysis and visualization of location information and uncertainty generated by people’s mobility is an engaging/underexploited domain of ubicomp.