Bourdieu wrote about cultural intermediaries in 1984. He meant journal editors, peer reviewers, and supervisors. These were the human actors who shaped what counted as legitimate scholarly work. Parker and Becker (2026), in a new conceptual paper in Frontiers in Education, make a small but consequential extension. AI systems, they argue, now perform the same intermediary functions in research, just through different mechanisms. The paper builds out a full AI literacy framework for researchers from that foundation.
The framework fills a real gap. Most existing AI literacy work targets K-12 students, citizens, or general digital skills. Researchers, who actually produce scholarly knowledge, have been mostly left out.
AI as a Cultural Intermediary
The theoretical move is the part worth lingering on. Parker and Becker argue that AI systems mediate scholarly norms, priorities, and forms of expression even without intentions or deliberate judgment. The mediation happens through structural mechanisms: training data composition, pattern reproduction, interface design. The concept shifts from actor-centered to function-centered.
This shift has practical weight. A journal editor or peer reviewer can be identified and engaged with directly. An AI system’s intermediary influence is distributed across architecture, training data, and interface, and is largely opaque to the user. The intermediary effect is real. The intermediary agent is missing.
That difference is what AI literacy for researchers has to start with.
What AI Literacy for Researchers Includes
The authors build on Selber’s (2004) digital literacy work and adapt three literacies to AI-mediated research. Functional literacy covers effective tool use and knowing when not to use AI. Critical literacy is where evaluation, skepticism, and ethical judgment all live. Rhetorical literacy concerns maintaining agency, voice, and meaning-making when AI participates in scholarly communication. The three build on each other.
These three literacies are then crossed with seven stages of the research lifecycle: research question formulation, literature review, methods selection, data analysis, writing, peer review, and dissemination. The result is a 21-cell capability map that operationalizes what AI literacy looks like at each point in the research process. It’s not abstract. It names specific behaviors: verifying AI-generated citations against primary sources, checking for what AI summaries include and omit, maintaining authorial voice across AI-assisted drafts, disclosing AI involvement in peer review.
The Assessment Rubric
The accompanying rubric translates the capability map into three performance levels: emerging, developing, and proficient. Emerging researchers begin to notice AI’s limitations but apply verification selectively. Developing researchers verify and evaluate AI output, but unevenly across stages. Proficient researchers do this as a matter of routine, across the whole lifecycle.
The rubric is built for supervisors and program directors. Parker and Becker are explicit that it’s designed to support learning, not punish AI use. The authors want a shared language for talking about AI in research without resorting to all-or-nothing judgments. They also acknowledge the risk that any capability framework can degrade into a compliance checklist, and they design the rubric to resist that drift.
Where This Lands in the Wider Conversation
Recent AI literacy work has tried to map this space from different angles. Hillman, Holmes, and Duarte (2025) did a rapid review of existing AI frameworks for the Royal Society and found that most treat AI literacy as a discrete skill set isolated from other literacies.
Chiu (2025) separated AI literacy from AI competency conceptually. The most recent piece in this lineage is Chiu and Rospigliosi (2026), who compared six global frameworks (UNESCO, OECD, China, US, UK, Australia) and called for a teacher-pedagogy bridge between policy and practice.
Parker and Becker complement that work by giving it specificity for one population (researchers) and one set of practices (the research lifecycle). The capability map and rubric are concrete in a way most AI literacy frameworks aren’t.
The authors are explicit that “AI systems can assist with many research tasks. However, they cannot be accountable for accuracy, ethics, or interpretation” (p. 8). That’s the line worth carrying into any conversation about AI in research.
Limitations
The paper is conceptual, not empirically validated. The authors say so directly. The capability descriptors and rubric come from sustained practitioner engagement with hundreds of researchers, which is meaningful, but formal validation hasn’t happened yet.
Future work needs to test whether the three performance levels are reliably distinguishable across assessors and whether the descriptors hold across disciplines. Clinical research, computational sciences, and humanities scholarship may need substantial adaptation.
The other open question is what happens when AI capabilities shift. Autonomous research agents, multimodal models, and real-time research assistants are moving faster than any static framework can track. The authors acknowledge this and frame the document as a living structure.
The closing reframe is the part worth carrying forward. Parker and Becker emphasize that “research integrity depends on human judgment, not technological control” (p. 9). For doctoral supervisors, research integrity officers, and program directors, this framework gives a concrete starting point. The kind of AI literacy that protects scholarly judgment isn’t a checklist of skills. It’s a developing form of scholarly practice.
References
- Chiu, T. K. F., & Rospigliosi, P. ‘asher’. (2026). Six global frameworks for human-centred AI literacy and competency: Comparative analysis and a way forward. Interactive Learning Environments, 34(3), 1003-1005. https://doi.org/10.1080/10494820.2026.2648342
- Chiu, T. K. F. (2025). AI literacy and competency: Definitions, frameworks, development and future research directions. Interactive Learning Environments, 33(5), 3225-3229.
- Hillman, V., Holmes, W., & Duarte, T. (2025). A rapid review of AI literacy frameworks, with policy recommendations. A report prepared for the Royal Society. London: The Royal Society. https://royalsociety.org/-/media/policy/projects/ai-in-education/hillman-et-al-a-rapid-review-of-ai-literacy-frameworks.pdf
- Parker, J. L., & Becker, K. P. (2026). Defining and assessing AI literacy for researchers across the research lifecycle. Frontiers in Education, 11, 1827603. https://doi.org/10.3389/feduc.2026.1827603