For information about day 1, see here
I’m not sure how many people reading this are familiar with the CHI conference so here is a quick overview. Computer Human Interaction (CHI) is the premier academic conference every year for academic Human Computer Interaction researchers and for professional usability experts, user experience people and interaction designers who are doing cutting edge stuff. Every year, something like 3000 or so people descend for 4 days of often overwhelming talks, panels, demonstrations, video presentations, networking and, ideally, plenty of drinking. It’s considered somewhat of a taste maker and arbiter within this community. No one person can give a completely impartial view of CHI, most of the time, there’s literally about 14 different things going on at once. The most you can do is go to as much of the stuff that interests you as possible and hook into interesting conversation in the crowd.
Still, CHI is somewhat remote from the workaday world of people who are building just another web application or mobile app. The stuff talked about at CHI is often highly abstract and, quite frankly useless to professional developers and designers. Despite that, I love it. It’s a bunch of overwhelming smart people with an astonishingly diverse array of interests all talking about the stuff they’re passionate about with a mix of brilliant insight and hilarious naivety.
Computer Mediated Communications
phew, there is much more social stuff this year than previous years and I’m glad CHI is finally turning into a place where the social people feel a bit more comfortable. The first talk in this session was a really excellent piece of work on deception on Instant Messaging and, specifically, what the author referred to as “Butler lies” aka, lies your butler would have told for you if you had one. They had a really nice piece of experimental methodology where they instrumented the Pidgin IM client and asked users to rate every single message they sent on a scale of 0 – 5 for deception. Overall, about 10% of all user messages involved some kind of deception and 2% were butler lies. People mainly used butler lies for providing a convenient fiction for exiting a conversation, pretending they had work to do or they were going to cook dinner.
I don’t know what real insight can be gleaned from this paper but I think it’s excellent in setting a context for talking about design that needs to accomodate deception and polite fictions. In particular, the idea of designing for narratives is one I think is particularly important and I’m glad I’m seeing more of that kind of talk within this community.
The second talk, I’m honesly baffled by and I don’t know how to interpret it, even now. So “In CMC we trust, the role of similarity” was an attempt to replicate the classic work on similarity and trustworthiness within a virtual setting. They had participants play essentially an iterated prisoner’s dilemma game with 5 minute chat sessions after every 5 rounds. They then compared similarity metrics of the chat and showed that most types of similarly were correlated with better co-operation but explicit talk about money or negative language lead to less trust.
There’s a umber of things with this studyt that make it tricky to interpret, one was the complex relationship between the game and “trust”. One of my collegues indicated to me that a lot of that could be similarly explained by trying to navigate different strategies and not about trust at all. By defining trust as as being about trustworthiness within a game,I think there was a competing desire between seeming similar to the other person and seeming like a reliable person.
Another thing I couldn’t really wrap my head around was the way they interpreted and coded the text data. It wasn’t clear to me exactly what the things they were measuring meant in terms of real world behaviours.
Finally, their data seemed to suggest that there were a bunch of confounding factors that make the results hard to interpret. The first chat session between users was only after the 5th game but it seemed to me at that time that people had already decided whether or not to trust the other person. What would have been really interesting but was not probed was what factors caused people to shift from a high trust to low trust condition and vice versa.
The final paper in the talk, I didn’t pay that much attention to. It involved some kind of interesting visualisation of showing who does how much talking in a group chat box which is an interesting idea but I felt like they didn’t develop it in any interesting way. The effect they were looking at was how quickly groups come to consensus when they have this indication and I didn’t feel like they asked a provoking enough question for me to stay engaged.
Scientometric analysis of CHI
This was a panel based on the analysis of previous CHI papers from the last 20 years. I thought the presented weaved an interesting story of various interesting things he found in the CHI data. Probably the most relevant and controversial being that a best paper nomination/award is essentially useless for predicting future impact and, with that, the implication that if the committee is useless at picking a best paper, is it really great at picking accepted papers? The panel then presented 4 different reactions to this work and I thought all of it was interesting and thoughtfully teased apart.
Social networking sites
The first talk for this was the one I’ve been most interested in for the entire conference. Moira Burke presented some work she did at an internship at facebook where they examined what motivated new users to share photos on facebook. She did both a quantitative study with the massive reams of facebook data she had access to and then sprinkled it with anecdotes from qualitative interviews. The results are what you expect from large data sets, low p-values but also reasonably low mean effect sizes. What will be really interesting to me is the next step of this research where they start moving some of this stuff into design and A/B testing.
The next work did essentially the same thing, except looking at what predicts helpfulness in reviews on amazon. Probably the most interesting thing from that is that recency has a significant effect, probably due to the design of amazon. One thing I didn’t think to ask during the talk is that that seems like a highly weird result given that every review was recent at some point. The only way I can see to explain this is explosive growth in the usage of reviews at amazon. Maybe I’m interpreting the data wrong…
The final piece of work was on tensions on facebook and negotiating different social spheres interacting on facebook. It ties in with a lot of my work on faceted identity and I think it did a good if unimaginative job at exploring some of these issues. Since I plan to be writing much more about this in a published form, I won’t bother to re-iterate my thoughts on faceted identity at this point.
The final section, like yesterday, held nothing of interest so I’ve been spending this time catching up with people and also writing this blog post. For those of you actually at CHI, the dub reception is 7:30 tonight at the Marriott Copley Place Hotel, 4th Floor Salon E. Looking forward to letting loose and partying a little.