AI and academic writing

It is important to understand that AI tools cannot replicate the unique expertise, mindset, and authorial voice of a researcher and that these aspects can be negatively affected by AI tools. What's more, an experienced and knowledgeable human editor can understand, consider, and improve a text in many ways that AI tools cannot.

Understanding the capabilities of AI writing and language review tools allows you to effectively integrate them into your writing process while being aware of their limitations and when human expertise is necessary.

Benefits of AI writing tools

AI tools offer features that support academic writing, for example:

• Discovery: Can help generate potential research questions
• Content generation: Can produce text based on instructions
• Rewording/summarising: Can rewrite or make texts more concise while maintaining the meaning (sometimes)
• Proofreading: Can identify many grammatical, punctuation, spelling, prepositional, and capitalisation errors
• Linguistic suggestions and basic editing: Can (to some extent) make suggestions on style, syntax, vocabulary, formatting, and clarity
• Manuscript analysis: Can provide support for creating tables of contents or reverse tables of contents, drafting abstracts, and brainstorming content or titles
• Citation generation/management: Can produce and edit references according to different style guides
• Plagiarism checks: Can identify plagiarism (to some extent)
• Speed and cost: Fast and often cheap or free
• Word: Some tools integrate with Word

Limitations of AI writing tools

For academic purposes, AI-generated text (and AI-edited text) may be wholly or partially factually inaccurate or may imply connections between ideas that are not correct, or vice versa. Important information may be omitted (or incorrectly added) when texts are generated, summarised, or reworded. It is crucial, yet sometimes challenging, for authors to carefully consider their final text after using such tools to ensure that it is still accurate and complete. Furthermore, AI tools cannot mimic a researcher's unique voice, have limited effectiveness in refining writing style, and cannot break linguistic conventions for the purpose of adding stylistic elements.

Limitations of AI language review tools

AI tools often have difficulty dealing with long documents, which many academic texts are. And for texts that are particularly complex or need extensive editing, AI tools generally perform much worse than a human editor in identifying errors and suggesting useful changes. Only a human editor can ask questions to the author to gain deeper insights that can improve the text or be familiar with typical errors that obscure meaning and still be able to interpret the intent correctly. AI cannot yet identify nuances in arguments or suggest precise alternatives for technical jargon.

Therefore, it is crucial to hire a professional human editor for final versions of texts. A good editor will refine clarity, flow, and readability, especially in specialised fields. An editor can ensure that writing conventions specific to a journal or subject area are followed as well as preserve the author's voice and maintain the consistency and readability of the text as a whole.

Choose the right AI tool

AI tools are suitable for various writing and language review purposes:

Free for UB employees:

• Microsoft Editor: Good for proofreading in terms of spelling, grammar, clarity to some extent, passive form, and clichés.
• Copilot: Good for discovery/research, prewriting, and drafting with interactive brainstorming and content generation. Particularly useful for rewording/summarising, tables of contents, and generating abstracts and titles.

May require subscription for full features:

• Curie: Suitable for proofreading advanced academic and scientific manuscripts, especially for authors writing in English who are not native speakers of English. Integrates with Word.
• Scite: Good for literature reviews, formulating research questions, critical analysis, and assessing sources.
• Grammarly: Both general and academic AI-powered writing support and basic proofreading. Integrates with Word.

Information security

Using AI tools means sending text to external devices for processing and storage, which raises privacy concerns. Avoid AI tools for texts containing personal or sensitive information. Researchers also need to be aware of privacy and copyright issues, as AI tools can use uploaded texts for learning purposes, potentially affecting academic privacy. As a reviewer, note that you may not upload the text you received to any service that can read it. This is both problematic for information security and outright banned by publishers.

Rules of funders and publishers

 Funders and publishers have guidelines and rules for the use of AI in academic writing that must be adhered to. It might be worth looking at these before you get too far along in your writing process and have had the opportunity to test different AI tools. Many require authors to report on their use of AI language tools and specify where in the text they were used (for example, the methodology section). Keep in mind that AI can never be a co-author.

Funders have their own rules regarding applications and peer review, for example, the Swedish Research Council (VR) currently allows the use of AI in applications, but at the same time it is absolutely prohibited to use AI when peer reviewing applications. Read the Swedish Research Council's guidelines on the use of AI tools.

Guidelines on the use of AI tools (VR)

Advice from the EU

The EU requires you to account for everything you do with AI and to think about information security.

Read the EU leaflet Living guidelines on the responsible use of generative AI in research (pdf)