AI promises speed, scale, and novel forms of data-driven discovery. But it also changes what scientific research looks like on the ground — shifting how scientists pose questions, justify claims, evaluate results, and interpret surprises. These shifts unsettle foundational disciplinary assumptions: What counts as proof? Where does simulation end and experimentation begin? When is a result explainable, and when is it just predictive enough?
Beyond their philosophical implications, these are practical questions that emerge in the ordinary rhythms of using AI in a manner similar to a scientific instrument or a research tool. They emerge in how problems are framed, prompts are written, outputs are debugged, findings are verified or dismissed, papers are drafted, datasets are documented, and tooling and APIs are maintained. AI does not merely transform knowledge; it reshapes the infrastructure, labor, processes, and social life through which knowledge is produced and organized.
These shifts call for conceptualizing AI as a scientific instrument within a long-standing STS lineage. Classic studies of laboratory practice show that scientific facts are enacted through instruments that inscribe meaning into measurements. Histories of experimental culture emphasize that credibility of a scientific claim depends on organized procedures, trusted witnesses, and shared interpretive norms through which it is produced. Analyses of the “experimenter’s regress” illuminate how judgments of correctness and method bootstrap one another — you can’t know if the result is right unless the experiment was done correctly, and you can’t know if the experiment was done correctly, until you know that the result is right — while research on formal verification and computer-assisted proof demonstrates that even proofs require social negotiation, not just technical verification. Extending this lineage, scholarship on data-intensive life sciences traces how computation, databases, standards, and software reorganize scientific work, reframing what counts as an object of inquiry and how evidence travels across institutions.
What this “lab studies” lineage has not always foregrounded are the political-economic currents in which research practices unfold. The rise of AI — especially when tools originate from Silicon Valley firms backed by venture capital — invites scholars to reexamine how funding structures, platform logics, and proprietary infrastructures shape the conduct of science itself. In this sense, AI adds new dimensions for scientists and ethnographers of science to account for, even as it intensifies older dilemmas around trust, reproducibility, credibility, and epistemic authority. Recent critiques of model-mediated understanding further warn that machine outputs may appear coherent or confident even when they are ungrounded — raising urgent questions about how scientific judgment is being redefined.
Drawing on and extending this scholarship, the workshop examines the craft of doing science with AI. We invite empirical, methodological, conceptual, and theoretical work that surfaces how AI tools and agents are being taken up in everyday research, and how their use is reshaping scientific practice. Our aim is to move beyond simple narratives of acceleration or automation to better understand the interpretive labor, epistemic dilemmas, organizational frictions, and institutional constraints that accompany AI in the lab. We ask:
What counts as a good scientific question in the age of AI? What does it mean to trust a model output? How do scientists reason, argue, and take responsibility when inference is distributed across humans and machines? How do these practices reshape the nature of disciplinary expertise?
We’re particularly interested in projects that engage one or more of the following cross-cutting threads:
- Evidence and verification: Tracing the emergence of new interpretive protocols, examining how evidentiary thresholds shift in practice, investigating reproducibility, and surfacing the tensions between what can be predicted and what needs to be understood. How are traditional evidentiary standards evolving — or eroding — in the face of AI-mediated research? What distinguishes explanation from prediction, or simulation from experiment, in this new context?
- Craft, intuition, and tacit knowledge: Making tacit judgment visible, tracking how skills are taught, maintained, and tested, and identifying practices that sustain or dull expert intuition when AI tools are in the workflow. How do labs test and recalibrate judgment under automation, and what practices keep a feel for data and instruments alive?
- Materiality and automation: Following the material stacks of AI-mediated science: instruments and sensors, data pipelines, vendor ecosystems, robotic and cloud labs, and the maintenance and repair ecologies that keep them running. What breaks and with what consequences? Who fixes it and under what constraints? How do platform dependencies, calibration routines, and access arrangements shape which questions can be asked and which answers count as credible?
- Training, mentorship, and careers: Examining how training, authorship, and evaluation are being reorganized as AI assistance becomes ambient. How are expectations for coding, writing, attribution, and independence shifting across labs and disciplines? What now counts as an innovative proposal, dissertation, or job talk, and how can mentorship and assessment practices cultivate judgment rather than outsource it?
- Expertise and authority: Mapping how epistemic authority and decision rights shift as software, data, and infrastructure roles shape scientific discovery. Who defines problem formulations, acceptable error, and verification protocols? When does “infrastructure work” count as theory- or method-building? What boundary-work distinguishes “tooling” from “science,” and with what consequences for disciplines, expertise, careers, and voice in research directions?
- Institutions and gatekeeping: Interrogating how rules and norms are adapting (or not): funding logics, data-sharing and disclosure, IP and licensing, and peer review amid model-generated literatures. How are reviewers and editors triaging, auditing, and calibrating trust? What new disclosure or provenance practices are emerging, and how do policy changes and platform terms reshape incentives for openness, reproducibility, and credit?
We welcome a wide range of works-in-progress, including:
- Ethnographic studies of labs and research groups integrating AI into daily work;
- Conceptual essays on automated laboratories, simulation, explanation, and epistemic risk;
- Infrastructure and protocol analyses that surface assumptions embedded in AI instrumentation and vendor ecosystems;
- Legal or policy research on how AI-generated outputs interact with norms of scientific evidence;
- Experimental formats — prompt logs, AI-assisted research diaries, and data maps — that capture the friction and improvisation in AI-mediated scientific practice.
- Domain-grounded cases (for example: ecology using computer vision on animal video, radiation oncology and clinical decision support, climate and materials modeling, digital humanities projects expanding corpora via NLP).
We encourage all attendees to approach the Data & Society workshop series as an opportunity to engage across specialties, and to strengthen both relationships and research through participation. While we recognize the value for individual projects, we also see this as a valuable field-building exercise for all involved.
