Why and how should authors use AI responsibly? The power asymmetry

AI has become an everyday part of academic writing. At this point the meaningful question is not "should AI be used?" but "how should it be used responsibly?" and "which power asymmetries does its use create or deepen?" This post addresses both questions on the basis of the academic literature and offers a practical framework. (Note: claims below are given with in-text author-year citations; full references are in the References section at the end.)

Why should authors use AI?

The case for responsible use rests on a concrete benefit.

• Time and efficiency. Large language models can speed up tasks like literature review, data coding and editing academic text, saving researchers valuable time (van Dis et al., 2023). That means more time for the substance of research — the question, the design, the interpretation.
• Crossing the language barrier. Non-native English authors face a measurable disadvantage in scientific publishing: higher rejection and far more revision requests on the basis of language alone (Amano et al., 2023). AI-assisted language editing can lower this barrier so that the quality of the idea is not overshadowed by the quality of the language.
• Accessibility. Used well, AI eases drafting and clarifying text, widening access to the writing process itself.

In short, used correctly AI is a lever, not a shortcut — it doesn't replace human judgment but strengthens it.

The rules of responsible use

What turns the benefit into reality is a few principles that discipline the use.

1. AI cannot be an author

Leading publishing bodies agree: AI tools cannot be listed as authors, because they cannot take responsibility for the accuracy, integrity and originality of the work (ICMJE, 2023; COPE, 2023; WAME/Zielinski et al., 2023). Morally responsible publishing requires authors to be accountable for what they write, and AI tools lack that accountability (Hosseini, Rasmussen & Resnik, 2023). Accountability always belongs to the human.

2. Disclose your use transparently

AI use is not a matter of authorship but of transparency (disclosure). ICMJE (2023) recommends disclosing use in the appropriate section: use for writing/editing in the acknowledgements, and use for data collection/analysis in the methods. Clearly stating what was used, how and to what extent lets readers and editors evaluate the work correctly (Hosseini, Resnik & Holmes, 2023).

3. Verify every output

Large language models can be fluent but unreliable: they can produce non-existent citations and fabricate facts as "hallucinations" (van Dis et al., 2023). So every citation, number and factual claim must be verified against an independent source. Using an unverified AI output as-is leads — beyond plagiarism and similarity risks — directly to scientific error.

4. Respect confidentiality and journal policy

Don't feed sensitive, personal or unpublished data into open AI tools — such data may be retained or leaked. And every journal's AI policy differs; check your target journal's rules before submitting. This should be part of your pre-submission checklist.

Power asymmetry: is AI an equalizer or a deepener?

At least as important as the rules of use is this question: how does AI affect the existing balances of power in academia? The answer isn't one-directional — it's two-way.

The equalizing side. AI-assisted language editing can reduce the language disadvantage non-native English authors have carried for decades (Amano et al., 2023). In this sense the tool can serve equity by foregrounding the quality of science rather than the language.

The deepening side. The same tool also creates three new asymmetries:

1. Access gap. The most capable models are often paid; access to advanced tools, compute and institutional subscriptions is unequal. This disadvantages under-resourced institutions and regions.
2. English bias. Because models are trained predominantly on English data, they are most capable in English, which can produce uneven performance for authors writing in other languages.
3. Detector bias. AI text detectors can unfairly flag non-native English authors' original text as "AI-generated" (Liang et al., 2023) — so a technology expected to increase equity can turn into a new mechanism of penalty.

This duality shows why responsible use is not only an individual ethical matter but also a structural equity matter. The tool is not neutral; how it is distributed and audited determines who wins and who loses.

A responsible balance

These two axes — responsible use and power asymmetry — together point to three principles:

• Transparency — disclose use so that evaluation is fair.
• Equitable access — design tools and policies to reduce, rather than deepen, existing inequalities.
• Human accountability — ultimate responsibility always rests with the author; AI is a tool, not a co-author.

Peerfect and responsible AI

Peerfect uses AI within exactly this frame: the free engine that matches your paper to fitting journals in seconds and flags weak points before submission is designed to inform human decisions, not replace them. The final decision — which journal, how, and with what disclosure to submit — is always yours. If you want a closer look at the ethical dimension of AI's role in academic writing, see AI use in academic writing as well.

In short: used responsibly, AI is a powerful lever that saves researchers time and can reduce language inequality. But it does so only under four conditions — keeping authorship human, disclosing use, verifying every output, and respecting confidentiality and journal policy. The same tool can also create new power asymmetries through access and detector bias; that's why responsible use is not limited to individual ethics but is a structural equity matter.

---

References

• Amano, T., Ramírez-Castañeda, V., Berdejo-Espinola, V., Borokini, I., Chowdhury, S., Golivets, M., … Veríssimo, D. (2023). The manifold costs of being a non-native English speaker in science. PLOS Biology, 21(7), e3002184. https://doi.org/10.1371/journal.pbio.3002184
• Committee on Publication Ethics (COPE). (2023). Authorship and AI tools: COPE position statement. https://publicationethics.org/cope-position-statements/ai-author
• Hosseini, M., Rasmussen, L. M., & Resnik, D. B. (2023). Using AI to write scholarly publications. Accountability in Research. https://doi.org/10.1080/08989621.2023.2168535
• Hosseini, M., Resnik, D. B., & Holmes, K. (2023). The ethics of disclosing the use of artificial intelligence tools in writing scholarly manuscripts. Research Ethics, 19(4), 449–465. https://doi.org/10.1177/17470161231180449
• International Committee of Medical Journal Editors (ICMJE). (2023). Recommendations for the conduct, reporting, editing, and publication of scholarly work in medical journals. https://www.icmje.org/recommendations/
• Liang, W., Yuksekgonul, M., Mao, Y., Wu, E., & Zou, J. (2023). GPT detectors are biased against non-native English writers. Patterns, 4(7), 100779. https://doi.org/10.1016/j.patter.2023.100779
• van Dis, E. A. M., Bollen, J., Zuidema, W., van Rooij, R., & Bockting, C. L. (2023). ChatGPT: five priorities for research. Nature, 614(7947), 224–226. https://doi.org/10.1038/d41586-023-00288-7
• Zielinski, C., Winker, M. A., Aggarwal, R., Ferris, L. E., Heinemann, M., Lapeña, J. F., … WAME Board. (2023). Chatbots, generative AI, and scholarly manuscripts: WAME recommendations on chatbots and generative artificial intelligence in relation to scholarly publications. World Association of Medical Editors. https://wame.org/page3.php?id=110