The ethics of using AI in research and scientific writing

by Elizabeth Oommen George
The ethics of using AI in research and scientific writing

Many researchers feel that there is a gap between their writing skills and the requirements of high-quality manuscripts for journal submission. This leads to a lack of confidence and even anxiety. Even when they get around to writing, they may not be able to do so at the speed needed to keep up a regular publication pipeline, which is important for a successful research career. Against this backdrop, the use of AI tools to improve scientific writing sounds exciting.

AI has the power to support human effort at every front for daily activities, and the use of AI in research is no exception. AI in publishing, particularly AI-driven language solutions, is making rapid advances to reduce time, effort, and costs. The applications of AI in research and publishing include language translation, summarization, language checks, plagiarism and image quality checks, and content formatting. Unlike a colleague or editor helping you with improving your manuscript draft, an AI tool is at your service any time of the day (or night). AI assistance can make you faster and more efficient in meeting your targets.

The boundless potential of AI in publishing

Easier navigation of the literature: AI-trained algorithms can follow your reading patterns and interests and help direct you to relevant content, saving time during your literature search.

Summarizing: Let’s face it; it is impossible to read all the content relevant to a topic. Article summarizer tools can offer some relief from information overload by generating the gist of a paper quickly. Such tools can efficiently streamline literature reviewing. Some AI tools can even create abstracts and plain language summaries from scholarly research content.

AI-powered academic writing and editing: Numerous language-support services are available to researchers today to up their writing game. AI-based rapid manuscript checks before submission can save authors a lot of time and effort. AI writing assistance software allows researchers to write manuscripts free of common language and grammatical mistakes. Besides efficiently correcting such errors, AI tools catering to scholarly content also ensure the correct use of discipline-specific terminology.

Tracking and preserving authenticity: AI can be used for conducting similarity checks in a manuscript against a large body of published literature. Such checks can help an author avoid accidental and intentional plagiarism. AI tools can also be used for checking the authenticity of images to be used in research publications.

Ethical considerations of AI support for academic writing

The types of tools mentioned above offer valuable support during one’s research and writing journey. However, these technologies cannot replace human expertise entirely. One’s personal judgment is paramount at every step. The use of AI to support and improve writing raises various considerations of ethics in scientific writing.

Potential risk of bias and errors

An AI tool is only as good as the data it is trained on. AI writing tools are vulnerable to bias and therefore cannot be used to complete a writing task without human vetting. Human bias can be introduced if certain questions or keywords were unintentionally (or intentionally) excluded. This could lead to skewed results, misrepresentation, etc. In scientific writing, in particular, there are discipline-specific nuances that might not be learnt if the training data did not include such content. Developers should be aware of risks of bias and include wide and representative training sets. In addition, they should disclose the training sets used.

Further, AI writing tools might not always be 100% accurate. This is because they do not “understand” the context. Authors still need to proofread their work carefully to avoid glaring errors missed or introduced by the program.

Environmental cost of training AI models

Training language-learning models is a highly power-intensive activity. The crunching of massive amounts of data calls for significant energy consumption for cooling data centers. The advancement of machine-learning models is increasing their carbon footprint,1 making AI usage and the concomitant carbon emissions an environmental concern. Thus, those generating such technology will need to find energy-efficient options and measures to offset negative environmental effects. Disclosing how the developers are doing so will allow users to make informed choices of “greener” AI tools they’d rather use.

Potential “learning losses”

A recent study demonstrated that increased use of navigation apps (“digital” directions) worsened users’ internal navigation abilities.2 The analogy with our dependence on AI is striking. It is a cautionary tale for researchers and students to not give up continued learning of writing skills just because AI writing support is available. Overreliance on such tools should not come at the cost of losing the ability to write in your own voice and ignoring grammar or language errors. You should never stop making the effort to learn correct spellings or pay attention to minor but important details in your writing!

Ease of rewriting

“Rewriter” tools help with preserving the original meaning of a text. Such tools aim to help articulate and structure text differently or simplify or shorten text. Many rewriter tools help with paraphrasing text and generating synonyms. However, such tools should not be misused for the sole purpose of plagiarizing or repurposing published text and passing it off as one’s own! Overusing rewriter tools and synonym generators to escape plagiarism detection is downright unethical. Researchers should be mindful of using AI writing tools focused on improving their writing and minimizing errors rather than employing them to rehash pre-existing text.

It might be a good practice to disclose publicly when authors have created an article with the assistance of AI, especially when the tool’s creative input is high, for example, if a text is “cowritten” with AI,3 as is more common in fiction writing than scientific writing.

Conclusion

The increasing adoption of AI in publishing raises issues around ethics in scientific writing. The role of human control and oversight in AI systems must be underscored. What separates AI from humans is “agency,” and we should strive to exercise agency in every activity we seek AI support from.

It is important to view AI as a means to help us do things better. Think of how we moved from scratching quills on scrolls to writing on paper to typing on a typewriter. Then came the computer and word processing software, armed with spellcheck and a myriad of other functions. AI writing assistance is simply a natural progression! Embracing the ethical and balanced use of AI is the way forward rather than resisting it.

References

1. Gibney, E. How to shrink AI’s ballooning carbon footprint. Nature 607, 648 (2022). doi: https://doi.org/10.1038/d41586-022-01983-7

2. Dahmani, L., Bohbot, V.D. Habitual use of GPS negatively impacts spatial memory during self-guided navigation. Scientific Reports 10, 6310 (2020). https://doi.org/10.1038/s41598-020-62877-0

3. Penn, J.F. Co-writing with artificial intelligence with Yudhanjaya Wijeratne. The Creative Penn. (2021) https://www.thecreativepenn.com/2021/01/18/co-writing-with-artificial-intelligence-yudhanjaya/

You may also like