AI Lab Strategies Mirror Dance Styles
Research groups face a fundamental choice when adopting artificial intelligence: whether to standardize every workflow or let individual researchers experiment freely. Neither approach is inherently wrong, but each suits a different kind of laboratory culture, just as a waltz suits a formal ballroom while a mosh pit suits a punk concert.
The push for uniform AI policies across institutions has created an uncomfortable squeeze for many research leaders. They watch journals, universities, and government agencies implement broad rules about large language models, yet find themselves unable to translate those rules into meaningful guidance for their own teams. The gap between institutional policy and daily laboratory practice has left many principal investigators uncertain about how to even start the conversation with their students and postdocs.
Sarah Burke-Spolaor, an astronomer at West Virginia University in Morgantown, experienced this struggle directly. She and her colleagues watched AI policies spread across astronomy and government institutions while their own team struggled with how to raise the topic with students and postdocs. Simply telling researchers not to use AI no longer works, because many are already using these tools whether officially sanctioned or not.
Burke-Spolaor and her team concluded that the conversation should focus not on the technology itself but on what each research group values collectively. The question becomes one of shaping team culture rather than policing a tool. This insight led them to draft a white paper posted on the arXiv preprint server in July 2025, which has not yet undergone peer review
Four Personalities, One Shared Question
The white paper proposes that no single AI adoption strategy fits every research group. Instead, the authors define four distinct laboratory archetypes, each with its own priorities and preferred phases of the research process where AI might prove most valuable. These archetypes function like personality profiles for research teams, helping members recognize what they collectively care about before deciding how to integrate AI tools.
Michelle Ntampaka, an astronomer at the Space Telescope Science Institute in Baltimore, Maryland, and co-author of the paper, found inspiration for visualizing the archetypes in an unexpected place. The team plotted each lab’s traits on a radar diagram modeled after the difficulty chart used in the video game Dance Dance Revolution. The gaming connection proved practical, giving researchers a familiar visual framework for mapping their group’s values across multiple dimensions.
The paper includes a worksheet at its conclusion so research groups can systematically work through their own priorities. Teams plot where they fall on various axes, identifying which archetype most closely matches their collective identity and what that implies for their AI adoption strategy.
From Policy to
Practice in the Laboratory
The practical implications of this framework extend beyond mere philosophical debate about research culture. Different archetypes might reasonably use AI during entirely different phases of the scientific process. A group that prioritizes speed and iteration might deploy language models for generating initial code drafts, while a group focused on accessibility could use the same tools to help non-fluent English speakers refine their scientific writing.
The paper’s authors note that AI applications span a wide range of laboratory needs, from coding assistance to language support for international researchers. This breadth means a blanket policy, whether permissive or restrictive, inevitably fails some portion of the research community. What works for a computational astrophysics group may hinder a field-heavy ecology team, even within the same institution.
Whether the four-archetype framework can accommodate the full diversity of scientific practice across disciplines remains untested. What began as a conversation among space-science researchers about their own teams has produced a tool that other fields may need to adapt or replace with frameworks that better match their particular values and workflows.