Month: September 2025

On a Friday afternoon, we met to discuss how Laud Humphrey’s Tearoom Trade: Impersonal Sex in Public Spaces came to be treated as a primary example of unethical field research methods, and–given his focus on private behavior in public spaces–how groupchat research raises new considerations for the old strategies of ethical disclosure and consent. Two hours after our meeting ended,  Gothamist published an article titled “NYC lawmakers accuse Amtrak of ‘Stonewall era’ tactics in Penn Station cruising crackdown”. Stephen Nessen reports that:

“The crackdown involved undercover officers posted in the bathrooms at urinals or in stalls, looking for men meeting up for anonymous sex. A cruising app called “Sniffies” featured a group dedicated to the bathroom. In recent weeks it featured numerous men warning others to avoid the bathroom because of the police presence.”

The tearoom has a chatroom, and it’s not researchers but police in disguise. If only they were subject to the rules of informed consent and disclosure!

Insights and notes from our conversation:

We started by discussing Tearoom Trade. Laud Humphreys’ research is still treated by mandatory IRB training videos as the primary example of unethical fieldwork. But as Michael Lenza argued, the institutional villanization of Humphreys had more to do with homophobia than the three pillars of Belmontian research ethics (respect for persons, justice, and beneficence).  Humphrey’s complex life journey from married Episcopalian priest with two children in the 1950s, to sociology grad student and civil rights activist in the 1960s, to full professor of sociology and co-founder of the Sociologists Gay Caucus in the 1970s, to psychotherapist and villain of sociology in the 1980s gave him special insight into the powers and dangers of “the breastplate of righteousness“, as he put it (and as Peter Nardi has reviewed). 

Then we shifted to discussing chatroom research. If the IRB suggests that researchers conducting participant observation in groupchats should announce themselves (disclosure) and request consent from participants, it’s been clear for a long time that chatrooms and groupchats don’t love to be researched.  These strangely private public spaces, and strangely public private spaces, present problems not unlike traditional institutional ethnography: how long do you need to lurk to figure out what’s going on enough to ask for consent the right way?  When (and how) do you talk to the moderators, write an introduction?

Historians usually work with dead people; working with living people is harder, especially when working across lines of community, culture, religion, ideology, or professional boundaries. Consent procedures can seem very strange to those who are being asked to give consent, requiring an explanation of US research ethics, legality, and law, that positions both the researcher and the research.

There’s an IRB negotiation with the researcher regarding “potential harms”. But whether online or in person, what counts as dangerous information or “potential harms” can change with the laws, or the enforcement. Circumstances can change on the ground so quickly, changing what’s safe to post online, or what’s safe to report. Several participants reported on strategies for managing these complexities: make sure that you only ever use handles, even when signing consent forms; ask interviewees to choose their pseudonyms; conceal the handles of more private (or less famous) posters with pseudonyms. 

In a groupchat where people are bringing very personal and private information, participants are highly aware of warning each other “you don’t know who is in the group.” In a group like this participants might not leave any identifiable traces (even as they reveal personal concerns); only the researcher or journalist is identifiable. Researchers participating with their own vulnerabilities also might need to be careful about what and how they post. 

In a groupchat where people are coming and going quickly, the moderators might be the only ones who can give consent. It might take offline relationships–including kinship ties and other networks–to get to know the group admin, in order to be invited into the group. 

 “Passing” or coming to be invisible–whether as a fly on the wall or a part of the activity–has often been regarded as a sign of good ethnography; and that’s easier if you’ve always already been in the places where you’re doing research. But when you pass, you might need to remind your interlocutors you’re doing research, remind them not to let their boundaries slip. 

OIKOS members Lily Chumley, Arelí Rocha, and Erica Robles Anderson join Takuya Maeda and Luke Stark to discuss Alternative Frameworks for Computing in the Age of AI at the annual SIGCIS conference (Special Interest Group on Computing, Information, and Systems).

Brain-computer metaphors have long dominated accounts of computing. No matter how often the limits of the analogy are pointed out, they persist as explanations of the state-of-the-art and as narratives for investment. At the core of this fantasy lies an idealized model of communication as dialogue between two reasoning partners. Yet today, computing is anything but a bounded system of rational exchange.

The panel highlighted approaches to computing-human interaction that foreground collectivity, context, environment, genre, and text. Together, these papers suggest that it is time to turn the page. Computing has taken many forms. If we are to guide what comes next, frameworks centered on individual intellect will not adequately describe the present or provision more inclusive futures for social action.

In “AI Has Entered the Chat–Is Anyone Else There?” Rocha drew from her recent Signs and Society article, “We Share an Unbreakable Bond: Sociality and Language Ideologies in Human Relationships with Artificial Intelligence,” to explore sociality and language ideologies in relationships with AI. As chat-based interfaces proliferate, familiar anxieties about alienation resurface. Yet for users of Replika, a chatbot built for hyper-personalized, often intimate experiences, chat is multivocal all the way through. Conversations with Replikas are collectively interpreted and debated in digital spaces, raising questions about humanness, intimacy, and the realness of love and heartbreak. Rocha’s talk traced how computational sociality emerges through vulnerability, multivocality, and shared linguistic practices that exceed dyadic frames of human–machine dialogue.

In Generative City, Robles Anderson argued that Cold War–era dreams of thinking machines still shape today’s AI discourse, even though we already inhabit dense ecologies of platforms, protocols, and services. Rather than dwelling on apocalyptic anxieties, she proposed the city as a model of participatory artifice. Urbanization—our species’ oldest techne—offers an abstraction for imagining multi-scalar, collective experience. It shifts the paradigm from dialogue as the gold standard of human–computer interaction toward demotic speech and civic multiplicity. By reframing computing through urban metaphors, Robles Anderson developed a critique that applies as much to smart machines as to master planners, opening space for more generative, civic-oriented concepts of computation.

In Open States of Command: Power, Deference and Demeanor in Interactional Text Production Chumley examines the dynamics of power, deference, and disobedience in interactions with writing assistants. Situating the prompt within a long history of interactional writing—dictation, transcription, and textual commands—she highlighted how the “open state of command” recurs across contexts. From secretaries taking dictation to wives that type and chatbots that generate text, refusals, failures, and other breakdowns of imperative, she explains how LLM metadiscourses reveal the social orders modeled in AI systems as well as continuities in how power and deference shape text production despite shifts in mediating technologies.

Finally, in Human Enough: How the Eliza Effect Became a Design Strategy, Maeda and Stark argued that the “Eliza effect”—users’ tendency to attribute human qualities to computing systems— has evolved from a psychological quirk into a deliberate design philosophy. Tracing how “human enough” intelligence functions as a strategy of social engineering, they showed how design features—from 1990s animation techniques to reinforcement learning with human feedback—align systems with human emotional cues. The Eliza effect succeeds not through capability but by exploiting users’ instincts for social interaction, encouraging misplaced trust and obscuring questions of accountability and labor. Their analysis reveals how anthropomorphic design, far from neutral, serves economic and political interests by defining what counts as human—and who can be displaced.