In my Pervasive 2007 Doggie Bag

My unstructured take-aways from this week’s Pervasive conference in Toronto.

Doctoral Colloquium
I presented a shortened (10min+20min discussion) version of the talk I gave at CHI. I received more or less the similar feedback as in the review of the submitted paper and at CHI in San Jose. I was once again encouraged to take advantage of my engineering skills and interests for the human side of technology. The new suggestions consist in considering the relation between designers-developers-users. That is make them aware of the uncertainty. The analysis of the user behaviors could both the engineering and interface design of location-aware applications. I could help designer think about all the possible issues for the design. It could maybe take the form of a systematic approach on the issues described in the CMPPC workshop paper).

The faculty advisors were: Albrecht Schmidt (Ludwig-Maximilians-University Munich), Mike Hazas (Lancaster University), Tom Rodden (University of Nottingham), Marc Langheinrich (ETHZ), Boriana Koleva (University of Nottingham), Rene Mayrhofer (Lancaster University), and Gerd Kortuem (Lancaster University). The slide and notes of my presentation are available here.

Probably the last of my “bridging social-technical gap” serie of presentation. Time to move forward and act.

Scrutability
In his talk on PersonisAD: Distributed, Active, Scrutable Model Framework for Context-Aware Services, Mark Assad (The University of Sydney, AU) mentioned the concept of scrutability (i.e. make the what and why of context visible). For a scrutable application should be able to reveal the user what sensor detected him/her and for what reason (activity, location, movement). Scrutablity is extremely relevant to the “self-disclosing” design principal mentioned by Adam Greenfield in his keynote speech earlier that day.

scru·ta·ble (skrt-bl) adj. Capable of being understood through study and observation; comprehensible.

Authoring tools for location-based applications
Leif Opperman presented his research on facilitating the authoring of location-based applications. He has been running a first set of experiment that let artists use his authoring tools for location-based applications. He uses an ethnographic approach to study of displaying the ubiquitous infrastructure and the uncertainty of positioning and communication influence the work of the artist. Previously, he had been developing games on top of the Mobile Bristol Toolkit which has been released this week under the name of MScapers. InternetActu reported on the release. Apparently, Barcelona has been the stage of a mediscape game.

Source: Leif’s UbiComp 2006 paper Extending Authoring Tools for Location-Aware Applications with an Infrastructure Visualization Layer.

Usefulness and relevance
In his talk on the paper An Exploration into Activity-Informed Physical Advertising Using PEST, Matthias Sala showed a graph comparing the usefulness and the relevance of information (the sweet spot being when the 2 converge). Since I sometimes mix these terms when talking about the delivery of location information, it is important that I go back to the definition:
Usefulness: Having a beneficial use; Being of practical use
Relevance: Pertinence to the matter at hand

Team awareness and wearable computing
Andrew Vande Moere presented a work on wearable visualization with a study based on the hypothesis that awareness will change how people play: “Evaluating a Wearable Display Jersey for Augmenting Team Sports Awareness”. I was most interested in the user-centered approach in the design of TeamAwear: 1. evaluating ethnography, 2. participatory design I, 3. participatory design II, 4. usability evaluation.

Replay tool
Alistair Morrison presented further work on the University of Glasgow’s Replayer used to perform both qualitative and quantiative analysis for the evaluation of pervasive systems. Using Location, Bearing and Motion Data to Filter Video and System Logs.

Spatial cloaking
In his talk, John Krumm mentioned spatial cloaking as a way to increase the privacy from the analysis of location tracking. Spatial cloaking consists in revealing spatial coordinates with less accuracy. Similarly, temporal cloaking reducing the accuracy in time. Related to that, I stumbled on Anonymous Usage of Location-Based Services Through Spatial and Temporal Cloaking that differentiate location-based services aong the dimensions of frequency of access, time-accuracy and position accuracy:

In the introduction of the subject of his talk/paper “Inference Attacks on Location Tracks” John mentioned that people concern in privacy emerge in the trade-off between usefulness (how much people would pay) and risk.

Digital footprints and exposure awareness
In Virtual Walls: Protecting Digital Privacy in Pervasive Environments, Apu Kapadia introduced the concept of
exposure awareness with the example of an environment that give the people the information about how much information they expose. By deploying “virtual walls,” people can control the privacy of their digital footprints much in the same way they control their privacy in the physical world. Once again a design approach to reveal the systems following Adam Greenfield’s self-disclosing principal and very similar to the concept of scrutability.

Middlewares for context aware computing
I bumped into Mike Blackstock who very briefly introduced me to the current works in the field of middleware supporting context-awareness and most specifically the Java Context-Aware Framework and the PerCom workshop CoMoRea. Later, in chating with Paddy Nixon, it seems clear that:
a. interaction design may impact the implementation of the middleware systems,
b. uncertainty is rarely (if not never) taken into consideration and it is still a challenge.

Location-update strategies
In Zone-based RSS Reporting for Location Fingerprinting, Mikkel Baun Kjærgaard introduced a list of Location-update strategies I could use later (similar to Leonhardi’s A Comparison of Protocols for Updating Location Information. Strangely, the type of user-context generated update was not mentioned.

Tutorials
The day of tutorials lead the participants through all the layers that make pervasive computing a thriving research domain: sensors, location awareness, context awareness, machine learning, middleware systems, interaction, evaluation, ethnography. I recorded most of the lectures.

Tom Rodden introduced the issues in dealing with both the technical and human perspectives of pervasive computing. The HCI approach is about rethinking and reconsider the notion of context, or the human nature of the context. He showed a couple of videos of pervasive systems such as the Stanford i-Room artificially situated within a lab. He argued for more research work more situated in the real world, in every day, because the best way to test the use of such technologies is to make them leave the laboratory. The role is HCI is also to reveal the kind of interaction we can make emerge. Tom talked about “new value systems” with example of subtle interactive and ambient with a refreshing touch of english-humor. A video of the Key Table (video)was without a doubt the most surprising. This definitively goes into the trend of chili computing. The Drift Table (video) was another example

Paul Dourish and Ken Anderson went through the historical periods and transformation in the practices of ethnography to later reflect on they way we use this method today. This lecture was an extension of Paul’s Implications for Design as they described ethnography’s prime goal to write about people and not about trying to fix things (i.e. implication for design). More than understanding what people do, an ethnographic study analyzes how people think/the rationality/conceptual breakdowns. In other words, it aims at understanding people’s conceptual thinking. It is not necessarily the amount of data collected that counts for interpretation but rather it is very important to spot the “symptomes”.

I was absolutely not aware the the differences between emic and etic data to describe human behaviors.

Gregory Abowd argued that an evalutation of pervasive system needs both a technicaly savvy and evaluation savvy people. It is very rare to be good at the 2. There is a uniqueness in the evalution of ubicomp system. Because they take the world as laboratory and they need technology development in order to perfrom such evaluation. In other words, ubicomp helps ubicomp evaluation. An evaluation can be formative (part of the development process) or summative (e.g compare 2 applications). Of course, the exploratory work and the design approach of ubicomp makes some scientist think it is “soft”. However, ethnography and qualitative studies help bring rigor.

The evaluation work done by Gregory and his team made me think about what could be done in one of my real-world study. I could take their “beepers studies” as example to ask “where are you?” and questions about the awareness of the whereabouts.

Paddy Nixon refreshed my knowledge in distributed systems and middlware. He went through the approaches of event based distributed programming (publish-subsribe) and defended it and p2p for ubicomp rather than corba/jini (they make the bad assuptions of a perfect world, balancing the local vs. global). However, p2p has been about sharing and not efficiency. So far it has been extremely hard to build generic middleware systems for ubicomp

Interestedly, he mentioned that interaction design may impact the implementation of middleware systems and that uncertainty is still a challenge that nobody really started to tackle.

Anind Dey admited that the community does not try to model the entire world anymore (something that I heard at the DC as well… researchers have given up on over-infering and automate). Therefore now the goal is to understanding what people actually want and find the important pieces of context. Similarly to my talk at LIFT, he showed relevant examples of automatic doors and the difficulty notion of intentionality (i.e. The Amazing AutoDoor, Star Trek manually operated automatic door). Similarly, cleaning automatic doors has become a slow process (video).

Automation has an impact on cleaning. The door should understand that it is being cleaned?

Finally, representing uncertainty is still a challenge along with the balance between autonomy and control.