@groceryheist, @yiweiwu, @kaylea, and @mako attended CSCW 2026 from Oct 10 through Oct 14. A previous blog post covers our presentations. Here we’re sharing some takeaways and highlights from the conference. I (Nate) wrote this post, so it naturally reflects my views most, but I’m also drawing on my conversations with the other CDSC attendees.
Takeaways from OSS + AI Workshop
Kaylea Champion, Benjamin Mako Hill, and Matt Gaughan helped organize a CSCW workshop on Open Source Software and artificial intelligence, and Kaylea, Mako, and Nate attended. The workshop was conceived and framed as a conversation focused on problems open source practitioners have raised related to building sustainable collaborative software projects in the open. However, I was both astonished and energized by how workshop participants’ focus seemed to shift toward more foundational and future-looking questions about the nature of collaboration and its role in software production in the first place. Kaylea shared how the OSS developers she has interviewed recently are using AI intensively and enthusiastically, and Mako seemed joyous at the fun he’s been having contributing in previously unimagined ways to projects like LaTeX. But even though AI seems to make open source developers more productive, it also raises the question: what makes code worth sharing at all? Nate shared how he’s used Claude Code to create a bespoke application for data annotation, and then three other attendees described doing the same thing. Should we share these software projects, even if they are idiosyncratic? How do ubiquitous AI tools change the role of open software and knowledge commons when the value of a project shifts from its code or features (which may be easy to replicate) to its reputation, trustworthiness, ubiquity, or role in a project ecosystem?
Mako, building on insights he also shared in a presentation at UT Austin in September about his new book Peer Production in the MIT Press Essential Knowledge Series (co-authored with Aaron Shaw and Christian Pentzold; available open access here), described a potentially deeper threat AI poses to Open Source now that Open Source developers are using AI coding tools. If I use a “frontier” agent to write a patch, and it generates buggy or otherwise inadequate code that I fix, it sends my corrections back to its company, which can use this information to train the next model version. In this way, AI agents collect knowledge from open source developers and aggregate it into new models, which their companies can then sell back to open source developers. This distributed production of knowledge, which is then aggregated and disseminated, resembles a peer production process, only the aggregate artifact is a private good (although distillation of open-weight models may transmit some of this knowledge into a commons in some sense). A CSCW poster by Caitlin Morris and Pattie Maes presented a system that I found really interesting for envisioning future modes of knowledge sharing among collaborators using AI tools.
Notes from the Lasting Impact Award
The question “what’s CSCW’s role in AI research?” was asked in one form or another in panels and conversations at CSCW 2026. Does CSCW need to jump on the AI bandwagon and study industry-facing problems? Maybe, but what makes that different from mainstream AI or NLP research? Should CSCW focus critically on social problems related to AI? Certainly, but that work also happens at FAccT among other conferences. What does CSCW have to contribute that other fields wouldn’t? Surprisingly, I found one of the most compelling answers in Nikki Kittur’s presentation at the session on the lasting impact award he received with Robert Kraut for their 2008 paper “Harnessing the wisdom of crowds in Wikipedia: quality through coordination“. Mako was also on the panel, and in his discussion he pointed out the same thing that impressed me from Nikki’s talk.
Nikki showed how the concepts of implicit and explicit coordination he and Kraut introduced in 2008 for understanding Wikipedia are useful for designing tools that use AI to support collaborative workflows in ways that just prompting a chatbot or agent can’t achieve. The more general answer, which I find inspiring, is that CSCW has built a rich base of concepts that can be productively applied to create new understandings of how AI can be designed, built, and used collaboratively. Nikki’s talk showed how engaging concepts from CSCW’s legacy can lead to novel research ideas, and I think such scholarship has a disciplinary character that can earn respect and distinction in a way that bandwagoning rarely does.
Highlights from paper sessions
Here are a few paper presentations by people outside CDSC we thought wroth highlighting. There’s one pick from each of us who attended. I (Nate) have lightly edited the others’ descriptions.
Nate’s pick: People Can Accurately Predict Behavior of Complex Algorithms That Are Available, Compact, and Aligned by Lindsay Popowski, Helena Vasconcelos, Ignacio Javier Fernandez Heinsen, Chijioke Mgbahurike, Ralf Herbrich, Jeff Hancock, and Michael S. Bernstein. https://doi.org/10.1145/3788067
This paper’s title is a great description of the findings and central contribution. The problem of designing algorithmic systems to make them less opaque and confusing is usually framed as making AI “explainable” or “interpretable”. This framing centers the model and seemingly constrains the solution space to the problem of making the inner workings of complex mathematical models comprehensible to end users.
I love how this paper reframes this problem from computational complexity to cognitive complexity in terms of a user’s ability to construct a valid intuition for the model’s behavior. Then they show that this sort of problem, which also seems closer to what many use cases actually need, is solvable by models that are good at one comprehensible and useful thing. This is a really cool example of positivist HCI research that conceives a theory by drawing from HCI and cognitive science and then tests it in a straightforward experiment.
Kaylea’s pick: The Persistence of Retracted Papers on Wikipedia by Haohan Shi, Yulin Yu, Daniel Romero, and Emőke-Ágnes Horvát https://doi.org/10.1145/3817014
Kaylea wrote: It takes a (very) long time for retracted papers to get removed from Wikipedia, and the problem is worse for highly-cited papers but better for open access papers. I suspect that journals and academics can do more to set the record straight when retractions happen.
Mako’s pick: Creating Group Rules with AI: Human-AI Collaboration in WhatsApp Moderation by
Mako wrote: A speculative design paper that worked with 20 WhatsApp admins to use an AI to generate rules. It’s an interesting experiment that I think challenges some things the CDSC work on rules has taken for granted and that uses methods extremely different from those we would use to approach the problems. The work points to a range of interesting future research directions for rule researchers.
Yiwei’s pick: The Activist’s Guide to the Decentralized Social Universe by Sybille Légitime and Harini Suresh. https://doi.org/10.1145/3817010
Yiwei wrote: I like how the authors of this paper take platform choice as a question of collective action, showing that leaving a platform only works if communities can find somewhere that fits their specific threats and resources.
