"I don't know what I don't know": Exploring the Benefits and Challenges of AI in Interdisciplinary Research

Indu Panigrahi, Tal AugustarXiv 2026 · arXiv

TL;DR

While AI tools are rapidly becoming research staples, they present a unique double-edged sword for interdisciplinary work: they excel at "translating" jargon and navigating unfamiliar literatures, but often create a dangerous "illusion of understanding." This study argues that instead of just automating tasks, future AI for research should be designed to scaffold deep learning and support human-to-human collaboration across disciplinary boundaries.

What Problem Does This Paper Address?

Interdisciplinary research (IDR) is essential for solving complex societal problems, but it’s notoriously difficult. Researchers venturing outside their home turf face "unknown unknowns"—they often don't even know what questions to ask, let alone how to interpret the answers. They struggle with discipline-specific jargon, conflicting methodological standards, and a lack of the "intuition" that comes from years of immersion in a field.

While general AI tools (like LLMs) are being adopted for literature search and writing, we don't yet understand how they impact the specific, high-friction workflows of researchers bridging different fields. Does AI truly lower the barrier to entry, or does it just make "epistemic trespassing"—making claims in fields where one lacks expertise—easier and more invisible?

Methods

The authors employed a mixed-methods approach to map the landscape of AI in IDR:

  1. Literature Synthesis: They first categorized IDR challenges into three dimensions: Navigating Information (finding/understanding), Assessing Information (valuing/contextualizing), and Connecting Across Boundaries (collaborating/reconciling expectations).
  2. Survey ($N=62$): They surveyed researchers across 102 disciplines to quantify the frequency and difficulty of these challenges and the perceived utility of AI in addressing them.
  3. Interviews ($N=23$): They conducted 60-minute semi-structured interviews with a subset of survey respondents to gain qualitative depth on how and why they use (or avoid) AI.
  4. Thematic Analysis: The qualitative data was analyzed using reflexive thematic analysis to identify core benefits and risks.

Datasets

This study did not introduce a new computational benchmark or train a new model. Instead, the "dataset" consists of the professional backgrounds and experiences of the participants:

  • Survey Respondents: 62 valid responses from researchers primarily in university settings (53.2% PhD students).
  • Disciplinary Breadth: Participants spanned 102 disciplines across STEM, humanities, and social sciences, with Computer Science and HCI being the most represented.
Figure 1: An overview of respondents’ background: the left panel shows the distribution of their perceptions of, usage of, and familiarity with AI. Perception ranged from 1 = Negative, 2 = Somewhat negative, 3 = Neutral, 4 = Somewhat positive, and 5 = Positive. Usage ranged from 1 = Never, 2 = Rarely, 3 = Occasionally, 4 = Regularly, and 5 = Frequently. Familiarity ranged from 1 = Not familiar, 2 = Slightly familiar, 3 = Moder- ately familiar, 4 = Very familiar, and 5 = Extremely familiar. The middle panel lists the distributions of self-reported in- stitutions and occupations that the respondents work in; most were from universities, with a majority being doctoral students. The right panel lists the top fifteen disciplines by mention; participants could list up to five disciplines.

Key Insights and Ideas

The paper uncovers several "aha moments" regarding how AI interacts with the "unknown unknowns" of interdisciplinary work:

  • The "Lingo" Bridge: One of AI's most valued roles is acting as a flexible "translator." Researchers use it to describe an idea in plain language and ask for the specific keywords used in a target discipline, effectively "unlocking" a new literature they didn't know how to search for.
  • The Illusion of Understanding: This is the core risk. AI provides summaries that feel credible, but interdisciplinary researchers often lack the "taste" or domain-specific intuition to catch hallucinations or subtle misinterpretations. As one participant noted, "You don't know what you don't know, and so you can't really assess the degree to which the summary is credible."
  • AI Lacks "Research Taste": Participants emphasized that AI is "credulous"—it accepts claims at face value. It cannot tell you which papers are considered "canonical," which methods are currently trendy but controversial, or which findings violate the "unspoken rules" of a community.
  • The Risk of "Cheapening" Expertise: Some researchers worried that if AI makes crossing boundaries too easy, it might diminish respect for the time and effort required to truly master a discipline, potentially leading to shallower, less rigorous interdisciplinary work.

Key Results

The study synthesized the challenges and AI's role into a clear framework:

Table 2: Summary of challenges in interdisciplinary research along with the usage, benefits, and challenges of AI for interdisci- plinary researchers.

Quantitatively, the survey (Figure 2) showed that:

  • Most Frequent Challenge: Deciding if information is valuable ($Mdn = 3.0$).
  • Most Difficult Challenge: Considering a work's context in its field ($Mdn = 3.0$).
  • Highest AI Utility: Addressing unfamiliar terminology ($Mdn = 4.0$).
Figure 2: The frequency of, difficulty of, and utility of AI for interdisciplinary experiences across the 62 survey respon- dents. Each row (e.g., “Unfamiliar terminology”) represents an experience question from our survey (Table 1). The rating scales were as follows: Frequency (1 = Never, 2 = Rarely, 3 = Occasionally, 4 = Regularly, and 5 = Frequently); How Chal- lenging (1 = I do not experience this, 2 = Not challenging, 3 = Slightly challenging, 4 = Moderately challenging, 5 = Very challenging, 6 = Extremely challenging); Utility of AI (1 = I do not use AI for this, 2 = Not useful, 3 = Slightly useful, 4 = Moderately useful, 5 = Very useful, 6 = Extremely useful). The corresponding questions are provided in Appendix B.2.

Qualitatively, researchers reported using AI as a search engine with flexible queries, a discipline translator, and an in-discipline reviewer to provide preliminary feedback on how their work might be perceived by another community.

Limitations and Future Work

The authors acknowledge several limitations:

  • Participant Bias: The pool leaned heavily toward STEM and Computer Science, likely due to the authors' own networks, which may skew the results toward those already inclined to use AI.
  • Individual vs. Team: The study focused on individual researchers' perceptions rather than observing actual interdisciplinary team dynamics.
  • Established Workflows: Most participants had established research habits before the generative AI boom. Future work should look at "AI-native" researchers (like current undergraduates) to see if their challenges differ.

The paper concludes by advocating for AI systems that support collaboration rather than automate it, suggesting the development of "computational boundary objects"—tools that help different disciplines find common ground and surface misaligned expectations.

Summary updated

Abstract

Interdisciplinary research is widely lauded as beneficial for society but presents unique challenges to the researchers who bridge disciplinary boundaries. Despite the rapid increase in the availability and usage of AI systems for research tasks, there is little work studying how AI systems have impacted workflows that span unfamiliar disciplines. Through a survey ($N=62$) and follow-up interviews ($N=23$) with interdisciplinary researchers, we describe the unique challenges and opportunities for AI in interdisciplinary research workflows. Participants viewed AI both as a solution and exacerbation of the general challenge of "unknown unknowns": AI systems could facilitate a broad awareness of a new research area, yet also provide an illusion of understanding. Crucially, participants lacked the expertise and intuition to catch these illusions. Our findings motivate a focus on designing AI systems to support rather than automate collaboration across disciplines, scaffold learning opportunities, and include avenues for triangulation between resources.
BibTeX
@misc{panigrahi2026i,
  author = {Indu Panigrahi and Tal August},
  title = {{"I don't know what I don't know": Exploring the Benefits and Challenges of AI in Interdisciplinary Research}},
  year = {2026},
  eprint = {2611.15121},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2611.15121},
}
Figure