AI evidence extraction

Where Will AI Take Scholarly Communication? Voices From the Research Frontline

David Nicholas, Blanca Rodríguez Bravo, Abdullah Abrizah, Jorge Revez, Eti Herman, David Clark et al. · Learned Publishing · 2025

AI-generated evidence extraction, verified across multiple analytical personas. Not a substitute for the peer-reviewed original.

9/10

Relevance

2/4

Quality (LMQS)

Evidence

Citations

2.12

FWCI

Top 10%

Impact

This is an AI analysis. Read the peer-reviewed original at the publisher: https://doi.org/10.1002/leap.2008

Methodology & findings

Study design

Semi-structured, free-flowing interviews of 60-90 minutes duration with 91 early career researchers (ECRs) from 7 countries across multiple disciplines.

Sample

N = 91, 9 groups

Primary method

Thematic analysis of qualitative interview data. Keyword analysis for categorizing responses. Frequency counts and percentages reported for coded responses (Y/N/Don't know). No inferential statistical tests reported. Data organized in contingency tables cross-tabulating responses by country, discipline, and age band.

Main result

Three-quarters of the 85 ECRs who commented envisaged a transformational change to scholarly communications. "A large majority of ECRs envisaged a transformational change of scholarly communications coming" with "more than two-thirds of all ECRs believing" that "AI would turn out to be a transformative force." The findings revealed "20 different visualisations proffered, with an improved, more efficient, and open scholarly system topping the list." Additionally, "81% of all ECRs thought" that "journals will have a central role to play a decade down the line," and "nearly two-thirds thought that AI would create inequalities."

Reports effect sizes and confidence intervals.

Research paradigm

Interpretivist/qualitative

Author conclusions

The authors conclude: "Where will AI take scholarly communication? Well, the answer is clear: quite a long way and to all the nooks and crannies of the scholarly communications system. And, the specific details of which can be found in their voluminous and informed comments that pack this paper. The 'voices' have it." They further state: "What we found out is that three-quarters of the ECRs who responded envisaged the transformational change of scholarly communications" and emphasize that "ECRs are in the perfect position to act as scholarly communications soothsayers, so what they have told us about AI in more than 600 h of interviewing bears listening to and acting upon."

Risk of bias

Selection bias: convenience sample of ECRs, not representative; Attrition bias: some ECRs from Harbingers-2 cohort retained (26/91); others recruited to replace departures; Geographic bias: 35% from Poland, 24% from China; limited UK (3%) and US (4%) representation; Demographic drift: ECRs in cohort notably older than typical (median 34-37), due to retention from earlier phases and career progression; Language/translation bias: interviews conducted in national languages and translated to English; Interviewer bias: national interviewers interviewed people in their own universities; professional knowledge of interviewers supplemented data; Response bias: 5 ECRs did not answer the AI transformational force question; 12 did not answer; 15 said they did not know about transformed systems; Convenience sampling - self-selected ECRs who agreed to participate in long interviews; Country imbalance - 35% of sample from Poland, 24% from China, creating geographic clustering; Attrition bias - 26 of 91 participants were retained from previous Harbingers-2 study, while others left research or declined; Selection bias - ECRs self-selected and willing to commit 60-90 minutes to interviews; Age range variation - some ECRs retained from previous phases are "a year or two older," with oldest reaching age 51 instead of typical ECR ceiling of 40; Disciplinary imbalance - humanities and arts represent 25% of sample, social sciences underrepresented; Interviewer bias - national interviewers conducted interviews, potentially introducing cultural interpretation differences; Response bias - Chinese ECRs notably less engaged in answering open-ended questions about transformation; Temporal bias - preliminary data collection without full sample power; Selection bias: Convenience sample of ECRs, with 26 from previous Harbingers-2 study creating cohort aging bias; Self-selection bias: New ECRs recruited based on availability and willingness to participate; Interviewer bias: National interviewers conducted interviews in their own universities, potentially influencing responses; Attrition bias: ECRs who left research, no longer qualified as ECRs, or declined due to work commitments were replaced; Geographic bias: Imbalanced country representation (Poland 35%, China 24%, Spain/Malaysia/Portugal 11-13% each); Disciplinary bias: Arts and humanities ECRs over-represented (33/91, 36%) compared to actual research population; Language bias: Interview transcripts translated to English, potentially losing nuance; Temporal bias: Relatively old cohort for ECRs due to retention from previous study phases

Limitations

The authors state: "This was a preliminary study, part of the long-running, longitudinal Harbingers project, attempting to inform and plan for a major study, which would have a larger and more representative cohort of ECRs
Our findings should be treated with caution, more as informed observations, filling a knowledge vacuum." Additionally, "It was a convenience sample for a pathfinder project, so we cannot claim it is representative." The study also notes: "nearly half of these came from one country-China (9/22)" and that "there did not see much passion for change or answering the question in this country's case," suggesting potential confounding factors by country.

Open questions raised

The authors identify a critical knowledge gap: "At this early stage of the AI-driven developments, our understandings of researchers' views and practices are patchy: not only are empirical studies on the topic few and far between, but even those already available are often small scale and limited in scope." They also note the need for larger and more representative cohorts: "this was a preliminary study...attempting to inform and plan for a major study, which would have a larger and more representative cohort of ECRs."
The authors identify that "at this early stage of the AI-driven developments, our understandings of researchers' views and practices are patchy: not only are empirical studies on the topic few and far between, but even those already available are often small scale and limited in scope, since they focus on one country and/or one discipline and/or one specific aspect of the topic." They propose a follow-up study: "This was a preliminary study, part of the long-running, longitudinal Harbingers project, attempting to inform and plan for a major study, which would have a larger and more representative cohort of ECRs." Additionally, they note "Further work needs to be done here" regarding Polish ECRs' library perspectives.
The authors note "we know very little about them and AI (But lots about everyone else!)" regarding ECR perspectives. They identify that "at this early stage of the AI-driven developments, our understandings of researchers' views and practices are patchy: not only are empirical studies on the topic few and far between, but even those already available are often small scale and limited in scope." The authors propose this as "a preliminary study, part of the long-running, longitudinal Harbingers project, attempting to inform and plan for a major study, which would have a larger and more representative cohort of ECRs."

Extracted from: pdfAgreement 56%

Explore related topics

Responsible AI Use Human-AI Collaboration Qualitative