CHI 2021 trip report: “trying to establishing connection…”
Eighteen years ago, I remember sitting on a couch in front of a whiteboard arguing about the boundaries of human-computer interaction (HCI) research with two of fellow Ph.D. students. We were all in our first year at Carnegie Mellon University’s Human-Computer Interaction Institute, and the question felt urgent: this is what we were getting our PhDs in. How could we not know what was and wasn’t HCI? We argued about the nature of interaction, the essential nature of humanity as tool builders, the duality and friction of human cognition and computation. I knew I was in the right place.
A few months later, I found myself at the CHI 2003 conference for the first time, suddenly having to reconcile our grand philosophical claims with the messy reality of community. At the time, it was about 1,400 people, all with different ideas about what “interaction” meant, what “computing” meant, and even what “human” meant. There were papers about user interfaces of course, but also papers about non-digital cultures of computation and new theories of design. I suddenly realized what it meant to be an interdisciplinary community: I would never know the boundaries of HCI, because they are ever shifting, redefined by people in the community. And now, this community included me, and my ideas.
Eighteen years later, the CHI conference—formally, the ACM Conference on Human Factors in Computing Systems, pronounced “kai”—is something very different to me. It’s an event where I go to see old friends, to strategize with the community about it’s growth and policy, and to meet newcomers, welcoming them, but also learning from them. That conference of ~600 is now a conference of thousands, coming from all around the world. I last attended in Glasgow in 2019, several months before SARS-CoV-2 made the leap to humans, not knowing it would be the last time I’d see my HCI friends in person. CHI 2020, which was to be held in Hawaii, was canceled.
This year, which was to be held in Japan, there was time to plan a virtual event, spread across several days, drawing attendees from across Earth. The buggy, confusing platform, called Delegate Connect, used a sterile combination of Zoom and in-page streaming. The organizers’ first blog post of the conference was about workarounds for a time zone defect in the interface that was leading people to miss their sessions. And so rather than the usual anticipation that came from flying across the planet, having a lively pre-conference jet-lagged dinner with friends, I began this weekend with a sense of virtual conference dread that has become all too familiar this year—and a dread that is all too ironic, given that it is a conference that gathers together the world’s leading experts on designing user interfaces.
I warmed up to the conference by joining for a session at the CHIMe mentoring workshop, which gathers together new researchers from across the world, including undergraduates and new graduate students. Students spent two days supporting each other, getting advice from near peers and more senior researchers. I joined for a panel with De’Aira Bryant (Amazon Web Services) and Molly Nicholas (UC Berkeley) to talk about the many pathways into research, the role of internships, and the many strategies for improving writing skills, both for publishing and fundraising. Martez Mott was a fantastic facilitator for the session, asking great questions and keeping the audience engaged.
Sunday: opening, keynote, and talks
The next day, the conference opened. The session opened with some streamed music, some sponsor photos, then glitched with a couple of prompts that looked like they were from MSDOS. The screen froze. I refreshed the page, pressed an oddly placed play button, and the same video streamed, then buffered indefinitely. The virtual conference platform said that the session started at 14:15, but the conference program said 14:30. I did some squats while I waited, and whined on SIGCHI Discord—with my Ph.D. advisor of all people!
After 15 minutes, the glitchy sponsor video loop ended and the keynote began, but the buffering was overwhelming. The same few seconds kept replaying; I couldn’t even hear a full sentence. Instead, the thousands of people attempting to watch the keynote just chatted in one big chat room. It was chaotic but fun, with a dozen different conversations happening all at once. Eventually, they repaired the stream, and the talk started.
The speaker was Chieko Asakawa, an IBM Fellow at T.J. Watson Research Center who has devoted much of her career to advancing access technologies for people who are blind. (She is blind herself). Chieko gave a history of the many gaps in digital technology accessibility, and some of her work to examine digital braille, accessible web pages, home appliances. She reflected on how smartphone accessibility has been transformative for so many of her experiences, offering a mobile sensor platform for accessing the visual world. She also pointed out how many truly basic problems like interface elements without text are still ubiquitous, requiring more than just standards and technology, but changes in designer and developer behavior and decisions.
Near the end of the (late running) keynote, I ducked over to one of the sessions starting at 4 pm. I got a cryptic error saying “The media could not be loaded, either because the server or network failed or because the format is not supported.” I wrote support, but never received a reply. Rumors spread about a 5–30 minute delay to avoid overlap with the keynote.
Suddenly, one of the sessions started streaming with no warning and no welcome from the session chair. Fortunately, it was one of the award papers by one of our iSchool students Kung Jin Lee, investigating synchronous participatory design with children online. One of the most interesting findings from this paper was how much improvisation was required in order to sustain engagement, drawing parallels to improv comedy. Kung Jin did a great job fielding questions about cultural differences, talking about how similar design might unfold in South Korea.
I tried to go to another session, but the discoverability of the button to join a room was so abysmal, I gave up. I decided to just watch some of the presentation videos instead of trying to navigate to the live streams. I learned about lots of exciting work! For example:
- “Everyone wants to do the model work, not the data work”: Data Cascades in High-Stakes AI studied data engineers high stakes contexts and contributed a notion of “data cascades”, in which data gathered earlier in a process resulted in later problems. For example, there were often brittle data sets that didn’t account for the complexity of a deployment environment.
- Human Perceptions on Moral Responsibility of AI: A Case Study in AI-Assisted Bail Decision-Making investigated people’s perception of AI morality, exploring different notions of morality in the context of COMPAS bail decisions. They generally found that people expect explanations, but also that people had higher expectations of morality for human decision makers than AI decision makers.
- Engaging Teachers to Co-Design Integrated AI Curriculum for K-12 Classrooms taught teachers about AI, then generated integration ideas, producing four curricular ideas, including social studies, image recognition for vocabulary, and other topics. Teachers found the ethics topics resonated most deeply.
Overall, I was excited by the content of this first short day, but hugely disappointed by the virtual conference platform. There was no social presence, no visibility of other attendees (aside from chat), overwhelming technical issues, and countless usability problems on the platform, including the low discoverability of how to even join a session. Most of the limited connection I was getting was from the SIGCHI Discord. It was not a promising start.
Monday: technology and marginalization
Monday morning, after some email, I woke up ready to dive in to more talks. I checked my schedule, which I’d constructed the week prior, and it turned out that all of the talks I wanted to see were actually repeats from Sunday evening. Oops.
Instead, I jumped over to the award talks session, and caught Scott Hudson’s Lifetime Research Award talk. Rather than giving a retrospective, he spoke about the future, and specifically about the future of computing power. He observed that exponential growth in computing power is far from over, and that most researchers really don’t comprehend the significance of that on invention. For example, if computing power is doubling every two years, what are the implications of that on a 3 year NSF grant, or a 5 year research plan? If we think about 10–15 year time horizons for inventions, that implies that our research needs to be anticipating computers that are 30–50x faster than what they are now. He observed that while the rest of computer science might just view that as purely going to computing things faster, the major role for HCI to play is to imagine how to use all of that excess speed to in service of people. I’d heard Scott make this observation before way back when I was a Ph.D. student at CMU more than 15 years ago, and so it’s interesting to think about how these predictions have played out: my devices are exponentially faster, and many of those excess cycles have served user experience. In my life, the’ve come in the form of voice assistants, heart rate tracking, and the prominence of streamed high resolution video.
After teaching for two hours, I had some lunch, caught up with a friend I usually see at CHI on Discord, and then did some exercise.
After an afternoon meeting, Ruha Benjamin gave a keynote about race and technology, weaving together observations about the intersections between race, capitalism, surveillance, and technology. Her talk centered on the theme that in technology, “the fantasy of some is the nightmare of others”, which creates a duality of utopia and dystopia, both of which has a technological determinism that underlies it. She then shifted to the second part of her talk, which engaged how racism distorts how we see and how we are seen. She talked about how racial hierarchies are encoded in technology and how people use technologies. She ended with three main takeaways:
- “Racism and other systems of domination are productive, not in the normative sense, but in the literal capacity to produce things of value to some, at the expense of others.”
- “Social norms, structures, all exist prior to any technology; they are coproduced.”
- “Imagination is a contested field of action, not an ephemeral afterthought that we have the luxury to fantasize… most people are forced to live inside someone else’s imagination…misery for some, monopoly for others.”
The audience was incredibly impressed, and particularly appreciated the call to teach critical histories of our cultures around technology.
After a few hours of other work, I joined the Queer in HCI SIG that I (only barely) helped co-organize; the heavy lifting was done by Caitie Lustig and others. I joined last year as an organizer, just after I’d come out, and had a wonderful time meeting other Queer HCI researchers. I learned a lot from last year’s discussions, and this year went even deeper. We talked about the challenges of doing Queer HCI research when the community is so small, reviewers lack expertise, and the community increasingly demands positionality statements. We talked about strategies for local impact on inclusion issues and the need for mentorship. We also gave some time to discussing best practices around allyship and going beyond allyship. A big theme after the third year of the SIG was developing more infrastructure and guidance for the broader community, to help scale some of our ideas to the entire HCI community.
I had a busy Tuesday, and didn’t see much on the schedule that I wanted to attend, so I mostly took a day off from CHI. I resumed at 8 pm by joining a student volunteer “Pick-A-Brain” session, which invited senior members of the HCI community to talk to student volunteers. Students asked about the challenges of being HCI researchers in conservative CS departments that didn’t fully embrace HCI, the role reflection in teaching, and strategies for organizational change.
I had blocked Wednesday morning to attend some sessions, but realized there was very little value in actually attending, aside from a few minutes of Q&A. Instead, I found the YouTube videos of the 8 papers I wanted to see and watched them all on 2x speed. Here’s what I learned:
- Exploring Approaches to Data Literacy Through a Critical Race Theory Perspective examined workshops that have taught youth about data literacy finding that many contexts are uncritical and need to be open to open to critique and reconfiguration from a critical race theory lens.
- Expanding Explainability: Towards Social Transparency in AI systems added a new dimension to thinking about explainability: the human, social, and organizational context of explanations and how explanations are used.
- Seeing Beyond Expert Blind Spots: Online Learning Design for Scale and Quality found that instructor perspectives on open-ended questions being harder than multiple-choice did not align with student performance: multiple-choice questions were no easier, at least in HCI.
- Improving Instruction of Programming Patterns with Faded Parsons Problems focused on programming patterns (e.g., common solution templates such as iterating through a list of events), finding that a system that offered Parsons problems improved problem solving success relative to open-ended code writing exercises and were preferred by students.
- StoryCoder: Teaching Computational Thinking Concepts Through Storytelling in a Voice-Guided App for Children considered 5–8 year olds playing storytelling games that aligned reading literacy with programming concepts by offering a guided construction of procedural stories. Children could navigate the StoryCoder app, learn about computing concepts, and did better on computing concept recognition.
- On Designing Programming Error Messages for Novices: Readability and its Constituent Factors found that message length, jargon use, sentence structure, and terminology heavily shaped student perceptions of error message readability.
- The Role of Working Memory in Program Tracing proposed that a core cognitive challenge of program tracing is limits on working memory, supporting this claim across four experiments, but finding that there were two distinct reading strategies that mediated working memory impacts. It suggested ideas for reducing working memory load through refactoring and tool support in ways that support both strategies.
- Interpretable Program Synthesis explored three different representations for making a regular expression synthesis system’s behavior transparent, enabling users to guide synthesis and its program search more directly.
Thursday: More content
In the morning, I attended the Engineering Interactive Applications session, in which our paper on Falx, tool that uses program synthesis to generate data visualizations by example (paper, blog, video). The session was full of exciting work, including:
- Interpretable Program Synthesis tackled the problem of black box debugging with program synthesis techniques; it explicitly focused on regular expression synthesis by visualizing the search space and search, showing the programs that have been considered, and allowing the user to provide feedback on the programs the synthesizer trying. It helped with harder tasks, but this was mediated by how much domain knowledge users had.
- Falx: Synthesis-Powered Visualization Authoring was a project led by graduating Ph.D student Chenglong Wang; I helped him design his evaluation and write his CHI paper. The key insight was finding domain representations for examples (grammar of graphics) to synthesize programs that transform data into forms sufficient for visualization. Our study showed that using synthesis was generally easier for experienced ggplot2 users to use than using ggplot2, especially for more complex visualizations, largely by avoiding the need for programming. Chenglong did a fantastic job bridging the underlying technical contributions to the value to user experience.
- danceON: Culturally Responsive Creative Computing was a fun live coding platform for declaratively specifying animations that combine machine vision techniques like pose detection with graphics, to enable interactive computing-enhanced choreography. Some of the disruptive creative blocks were the usual syntax errors, but also 2D to 3D coordinate system transformation logic.
Overall, I probably spent about 10 hours at the conference this year, far less than the usual six 12-hour days I usually commit. Alas, the density of valuable interaction was a lower. I benefitted greatly from the short 5-minute videos; it was a lot of fun to zip through them instead of abstracts, and get a sense of the people int the community. But session Q&A offered little value: I couldn’t watch the videos at the speed I wanted and discussion didn’t really work given the limited chat and interaction on the platform. There were really very few meaningful ways to connect with people in the community aside from a firehose of Discord messages. The result was that I only got a tiny sliver of the reason I go to CHI: to connect.
It makes me wonder what’s driving decisions to make virtual conferences so content focused. Is it inertia? Is the bias baked into the platforms? Is it a sense of fairness to authors? Or perhaps other researchers really just don’t want connection, and just want content?
We made pretty different choices at ICER 2020, trying to create an abundance of social breaks, group discussions about paper presentations, and lots of social visibility. We’ll be doing even more of this for ICER 2021, building in an abundance of networking and conversation. After all, what is a research community without the community?