A"ira: Rethinking AI Research Assistants for Interdisciplinary Science
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Computer Science. The day in computer science: software, systems and theory on arXiv.
Tom: I'm Tom.
Jane: And I'm Jane, here with Lu, Meng and our in-house large language model, Lalam.
Tom: We're looking at the paper: "A"ira: Rethinking AI Research Assistants for Interdisciplinary Science".
Jane: Here's the gist of the paper. Scientific discovery increasingly depends on interdisciplinary teams whose members contribute distinct expertise, conceptual frameworks, vocabularies, assumptions, and standards of evidence While these capabilities improve personal productivity, they provide little support for the collaborative reasoning required to integrate knowledge across disciplines We argue that AI research assistants should evolve from tools that optimize individual workflows to systems designed for interdisciplinary teams We introduce aïra, an AI research assistant built around this idea.
Tom: The Need for Collaborative Reasoning Complex problems in health, climate, engineering, and public policy require researchers with different expertise to work together.
Jane: Interdisciplinary collaboration introduces challenges that extend beyond access to information because different fields often operate with different vocabularies, assumptions, methods, and conceptual frameworks.
Tom: Productive collaboration depends not only on understanding the scientific problem itself but also on understanding how collaborators approach that problem from different disciplinary perspectives.
Title and authors: Tom: The title itself, "Aira: Rethinking AI Research Assistants for Interdisciplinary Science," tells you they aren't just tweaking old tools; they're proposing a fundamental shift in how we think about what an AI assistant should actually do.
Jane: It suggests that the goal shouldn't be optimizing one person’s workflow, like faster literature reviews, but supporting the messy process of getting different experts to build something together.
Lu: The authors list a pretty diverse group, which makes sense because they're tackling this problem from multiple angles—from engineering to philosophy and even sociology, which is key for understanding those different vocabularies.
Meng: I’m curious how much of this design actually takes time away from the actual research and just spent on setting up these complex structures for collaboration.
Lalam: It seems like the paper argues that if an AI assistant can't handle those differences in assumptions and standards, it’s fundamentally failing its purpose in a team setting.
The paper's summary: Tom: So, what Aira is doing is taking recorded discussions from interdisciplinary meetings and turning them into something structured. Instead of just spitting out a conventional meeting summary, it analyzes the conversation through the lens of collaboration itself.
Jane: It’s not about summarizing what was said; it’s about figuring out whose perspective came from where, and what assumptions each person is making when they talk about a problem.
Lu: The system extracts all those different disciplinary perspectives—the vocabularies, the standards of evidence, even the priorities—and organizes them into a shared representation of the discussion.
Meng: So instead of one summary that everyone reads and agrees on, Aira is trying to show us where everyone is agreeing and where they fundamentally disagree based on their background.
Lalam: It’s moving from a single output to organizing all those different viewpoints so the team can see the landscape of the discussion clearly.
The paper's improvements: Tom: The authors lay out five specific design principles for this system, and they focus on making sure the AI doesn't just give us one final answer but supports ongoing teamwork.
Jane: They push for the AI to explicitly represent those disciplinary perspectives so you can see exactly how each field contributes to the larger research effort instead of just seeing a mixed narrative.
Lu: Another key idea is translating terminology across disciplines, not just defining terms, but identifying the relationships between terms where one word means something different in different fields or uses completely separate vocabularies for similar concepts.
Meng: That sounds like it would be super useful if it could automatically flag when someone is using a term that might be misinterpreted by another expert in the room.
Lalam: And they also want the AI to highlight implicit assumptions, so researchers can see those unstated premises—like assumptions about methodology or data quality—so they can check if their conclusions are based on shared starting points.
Conclusion: Tom: So Aira is essentially arguing that the future of research assistants needs to be about structuring discussions into actual future research directions instead of just summarizing the past conversation.
Jane: They want the system to recognize those unresolved questions as important outcomes of science, meaning it shouldn't rush you toward a premature consensus.
Lu: And they focus on organizing these discussions into actionable research artifacts so the team has something concrete to work on next, rather than just a pile of notes.
Meng: It’s smart that they are moving toward future work directions; that’s what we need when we have a big project running, not just historical reports.
Lalam: Ultimately, the paper calls Aira an AI Interdisciplinary Research Assistant because it moves the focus from individual productivity to genuinely helping teams develop a shared understanding of complex problems.
Diya Mirji, Tiffany Degbotse, Sharmil Nanjappa, Jiayi Zhou, Vishnu Manoj, Vihaan Nama, Boyuan Chen, Lee Tiedrich Information Sciences School University of Maryland Baltimore, MD Department of Sociology Duke University Durham NC Department of Philosophy and Kenan Institute for Ethics Duke University Durham NC
Pratt School of Engineering Duke University Durham NC Nicholas School of the Environment Duke University Durham NC Information Sciences School University of Maryland Baltimore MD Department of Sociology Duke University Durham NC Department of Philosophy and Kenan Institute for Ethics Duke University Durham NC
cs.HC, cs.MA
Submitted: 2026-07-14
Updated: 2026-11-06
Comments: Accepted to AIAS 2026
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Importance score: 83/100
The gist: Scientific discovery increasingly depends on interdisciplinary teams whose members contribute distinct expertise, conceptual frameworks, vocabularies, assumptions…
Key concepts
- Disciplinary Perspectives
- AI systems must show how different fields approach a problem rather than treating the discussion as one single story. This helps teams understand how each discipline contributes to the overall research effort.
- Terminology Translation
- The AI identifies relationships between terms, recognizing that the same word can have different meanings across fields. This bridges communication gaps by establishing a shared language for collaboration.
- Implicit Assumptions Highlighting
- AI makes hidden assumptions about methods, data quality, or causal links visible. This allows collaborators to spot potential disagreements based on differing underlying premises.
- Actionable Research Artifacts
- Instead of just summarizing past talks, the AI organizes discussions into concrete outputs like future research directions. This helps teams structure ongoing projects and plan next steps.
Terminology
Summary
Scientific discovery increasingly depends on interdisciplinary teams whose members contribute distinct expertise, conceptual frameworks, vocabularies, assumptions, and standards of evidence While these capabilities improve personal productivity, they provide little support for the collaborative reasoning required to integrate knowledge across disciplines We argue that AI research assistants should evolve from tools that optimize individual workflows to systems designed for interdisciplinary teams We introduce aïra, an AI research assistant built around this idea. p. 0
The Need for Collaborative Reasoning
Complex problems in health, climate, engineering, and public policy require researchers with different expertise to work together. Interdisciplinary collaboration introduces challenges that extend beyond access to information because different fields often operate with different vocabularies, assumptions, methods, and conceptual frameworks. p. 0 Productive collaboration depends not only on understanding the scientific problem itself but also on understanding how collaborators approach that problem from different disciplinary perspectives. p. 0 Interdisciplinary teams need support for collaborative reasoning, which includes defining problems together, reconciling competing perspectives, establishing shared terminology, identifying areas of agreement and disagreement, and translating discussion into actionable research plans. p. 1
Design Principles for AI-Assisted Collaboration
Based on the challenges encountered in interdisciplinary collaboration, a set of design principles guided the development of aïra. These principles include:
-
AI systems should explicitly represent disciplinary perspectives rather than treating a discussion as a single, unified narrative. p. 1
-
AI systems should translate terminology across disciplines to reduce communication barriers and establish a shared language for collaboration. p. 1
-
AI systems should highlight implicit assumptions so collaborators can identify potential sources of disagreement and evaluate whether conclusions are based on shared premises. p. 1
-
AI systems should distinguish between areas of agreement and disagreement rather than producing a single synthesized summary. p. 1
-
AI systems should preserve unresolved questions instead of forcing premature consensus, recognizing them as important outcomes of the scientific process. p. 1
Finally, AI systems should actively support future collaboration by organizing discussions into actionable research artifacts. p. 1
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Represent disciplinary perspectives explicitly rather than treating a discussion as a single, unified narrative by having AI systems
explicitly represent disciplinary perspectives rather than treating a discussion as a single, unified narrative.
This allows teams tounderstand how each discipline contributes to the broader research effort.
-
Translate terminology across disciplines by enabling the AI system to identify
relationships
between terms wherethe same term carries different meanings in different fields, or because disciplines use separate vocabularies to describe similar concepts.
This action is intended toreduce communication barriers and establish a shared language for collaboration.
-
Highlight implicit assumptions by making them explicit so collaborators can
identify potential sources of disagreement and evaluate whether conclusions are based on shared premises.
This addresses the need to makeunstated assumptions about methodology, data quality, causal relationships, or domain knowledge
visible. -
Distinguish between areas of agreement and disagreement instead of producing a single synthesized summary by having the AI system
explicitly representing these points of alignment and divergence.
This helps guide future discussion and decision-making by showingwhere researchers converge and where legitimate differences remain.
-
Preserve unresolved questions instead of forcing premature consensus by ensuring the AI system
recognizes them as important outcomes of the scientific process
rather than treating them as incomplete discussions. This supports the recognition that interdisciplinary meetings often conclude withopen questions that require additional expertise, data collection, or experimentation.
-
Actively support future collaboration by organizing discussions into
actionable research artifacts
and structuring them intofuture research directions.
This moves beyond summarizing past events to helping teams organize their work for ongoing projects. -
Develop a glossary of terms with discipline-specific definitions by having the final agent form a
glossary of terms and their definitions according to the discipline that used the term.
This directly supports establishing a shared language by mapping specific vocabulary back to its disciplinary context.
Related papers
- EmoPhone: A Multi-Wave Dataset for In-the-Wild Mobile and Wearable Affect Sensing
- Tasteply: Externalizing Stable and Situational Preference as Manipulable Spatial Objects
- Shielded Trajectory Sampling for Personalized Relational Safety in Conversational AI
- From Tasks to Turns: Measuring Human Labor Contributions in Human-Agent Joint Production
- Reconstruction and Reflection of Positive Experiences through Resurfacing Laughter-indexed Everyday Moments
- LLM-Assisted Reranking to Operationalize Nuanced Objectives in Recommender Systems