Triangulate Generative AI Tools in Academic Writing
António Pedro Costa, University of Aveiro (Portugal)
Researcher at the Research Centre on Didactics and Technology in the Education of Trainers (CIDTFF), Department of Education and Psychology, University of Aveiro, and collaborator at the Laboratory of Artificial Intelligence and Computer Science (LIACC), Faculty of Engineering, University of Porto.
Adriana Santos, Federal Institute of Bahia (Brazil)
Professor at the Federal Institute of Bahia (IFBA), Brazil. PhD candidate IFBA/UFBA/UNEB with internship at the University of Aveiro (Portugal).
Using Artificial Intelligence (AI) tools, particularly Generative AI (GenAI), as support for writing scientific articles can be a valuable approach, especially during the literature review phase. This process can be optimized by using triangulation. In this context, triangulation refers to using multiple GenAI tools to perform cross-validation or to complement the results obtained, thereby increasing the reliability and quality of the generated content
The discussion on using GenAI tools and how they should be explored in research has intensified over the past months. We talk about AI responsibility, and on March 13, 2024, the European Union approved a regulation (known as the AI Law) which establishes the first comprehensive rules for the development, use, and marketing of Artificial Intelligence
We intend to share here how various GenAI applications can be applied to writing scientific publications. Regardless of the usage, the proposal does not dispense with the knowledge and understanding of the researcher/user about the content being researched. In the example we explore, using the theme of this current text, we used the following applications.
- Consensus.app is a tool (still in Beta version) that simplifies access to scientific articles from a question or research inquiry.
- Quilbot.com is a text rewriting application that uses artificial intelligence to paraphrase or summarize content. It can be used to simplify language, avoid plagiarism, or generate alternative versions of a text.
- Elicit.com uses language models to extract data and summarize scientific articles. The tool generates a summary with references to various authors. It is also possible to access the complete article and obtain more information from each text (e.g., primary results, limitations).
- Lex.page is an AI-powered text processor similar to a traditional Google Doc, allowing users to use AI assistance whenever necessary. It uses the same model that powers ChatGPT.
- Grammarly.com is a grammatical and style correction tool that helps enhance the quality of writing in English. It detects grammar, punctuation, spelling, and style errors and offers suggestions for improvements to make the text more transparent, concise, and professional.
- Perplexity.ai allows researchers to generate, through a starting research question, other possible sub-questions that help them develop their research topics. It also provides for a question brainstorming (brainstorming).
Previous knowledge about the topic will allow the author to write a detailed instruction (prompt) so that the result is as intended. In the case of the elicit.com platform, the author only needs to define a research question (formulating good questions is a cognitively demanding task, as it involves understanding a problem, organizing ideas, synthesizing, etc.). Being a very recent area, LLM (Large Language Models), AI systems trained with large amounts of textual data, sometimes provide imprecise responses. Perplexity.ai can help with this formulation. Inputting a question into this tool generates, as a result, a small text referencing the original and, in the end, proposes three new possible questions. Most texts generated by perplexity.ai point to blog posts written by authors who understand the topic but are generally not reviewed. Therefore, the results can assist in confirming or reformulating the initial study question. From there, we can search for authors with scientific journal articles. For this, we turn to elicit.com.
Next, we present a proposed summary and authors when we searched “Triangulate Generative AI tools in writing scientific articles” – elicit.com (consulted on March 15, 2024):
A range of studies have explored the use of generative AI tools in scientific writing. Griffiths (2004) and Buchkremer (2019) both highlight the potential for these tools to identify and synthesize scientific topics, with Buchkremer (2019) proposing a "double funnel" approach to enhance the literature review process. However, Buriak (2023) and Leung (2023) caution that these tools can produce shallow, error-prone content and emphasize the need for human oversight. Burger (2023) and Grimaldi (2023) provide practical guidelines for using AI in research, with Burger (2023) focusing on systematic literature reviews and Grimaldi (2023) discussing the integration of AI-generated text in scientific articles. Liu (2012) and Guo (2020) present specific applications of AI in academic writing support and healthcare research, respectively.
Elicit.com allows consulting with more authors than those mentioned above and provides some summaries about the article (main findings, limitations, among others). It also provides access to the reference and, most of the time, to the DOI (Digital Object Identifier). Note: After several months of use by all the authors suggested by elicit.com, we have not yet detected any case where the text was nonexistent. Important: read the parts of the publications presented by elicit.com to validate what the tool presents.
Subsequently, through consensus.app, we improve the previous text by strengthening it with complementary readings:
A range of studies have explored the use of generative AI (GenAI) tools in scientific writing. Griffiths (2004) and Buchkremer (2019) both highlight the potential for these tools to identify and synthesize scientific topics, with Buchkremer (2019) proposing a "double funnel" approach to enhance the literature review process. However, Buriak (2023) and Leung (2023) caution that these tools can produce shallow, error-prone content and emphasize the need for human oversight. Burger (2023) and Grimaldi (2023) provide practical guidelines for using AI in research, with Burger (2023) focusing on systematic literature reviews and Grimaldi (2023) discussing the integration of AI-generated text in scientific articles. Liu (2012) and Guo (2020) present specific applications of AI in academic writing support and healthcare research, respectively. GenAI can aid in research idea generation, academic writing, and English writing learning, but critical thinking is crucial for maintaining rigorous scholarly standards (Hsiao-Ping Hsu et al., 2023).
Although it is still in beta, consensus.app allows searching for articles that may be indexed in different databases. This does not exclude the necessity of, depending on the type of review we are conducting, directly consulting some databases (for example, Scopus or Web of Science).
Subsequently, we can use quilbot.com to improve our text (writing syntheses, plagiarism check, among others) and grammarly.com for grammatical correction if the content is written in English.
Various studies have investigated the application of generative artificial intelligence (GenAI) methods in scientific writing. Griffiths (2004) and Buchkremer (2019) emphasize the capacity of these tools to detect and combine scientific subjects, with Buchkremer (2019) suggesting a "double funnel" method to improve the process of reviewing the literature. Nevertheless, Buriak (2023) and Leung (2023) warn that these tools have the potential to generate superficial and unreliable material, highlighting the importance of human supervision. Burger (2023) and Grimaldi (2023) offer practical recommendations for utilizing AI in research. Burger (2023) specifically addresses the application of AI in systematic literature reviews, while Grimaldi (2023) explores the incorporation of AI-generated text in scientific articles. Liu (2012) and Guo (2020) demonstrate distinct implementations of artificial intelligence in academic writing assistance and healthcare research. GenAI can assist in generating research ideas, academic writing, and learning English writing. However, it is essential to have critical thinking skills to uphold rigorous scholarly norms (Hsiao-Ping Hsu et al., 2023).
Lex.page can be used in parallel with the other tools throughout the writing. It allows for an exploration of the presented draft and suggests potential improvements to the text. The author must know how to write a good instruction (prompt) at this stage. In most cases, the instruction needs to be refined so that the obtained result meets what the author needs. Using lex.page, perhaps the beginning of this text would be better as we present it below. What do you think?
The use of generative artificial intelligence (AI) tools in writing scientific articles presents itself as a promising strategy, especially in phases such as bibliographic review, data analysis, and drafting preliminary project outlines. In the scientific sphere, triangulation refers to applying multiple AI models or tools to cross-validate results or complement information among themselves, which enriches and elevates the reliability and rigour of the “generated” content. There are intense debates about how Generative AI tools should be employed in research, and discussions often discuss the responsibility of AI. Recently, the European regulatory framework concerning the use of AI was disclosed. This text intends to share my experience in writing scientific works and explore the potential contribution of Generative AI tools to the writing process, focusing mainly on scientifically inclined texts. I recognize that this proposal may originate controversy; however, I intend to present the findings I have been gathering during my practice in employing and training algorithms in reviewing articles, where I identified indications of the use of these tools, although without being able to prove it unequivocally. Is it really so imperative to corroborate the use of AI tools in evaluating scientific articles, or should we channel efforts into raising awareness for their conscious exploration and thereby improve the value of what we produce?
Care with the use of AI persists, especially concerning ethical issues of plagiarism, intellectual property, and data privacy. However, by exploring, reflecting, and learning about the available AI tools, we can transform them into actual “thinking partners”, optimizing our time in pursuing new knowledge and enhancing the quality of research. In other words, using AI strategically, one can acquire information more efficiently, providing the researcher valuable time for reflection and improving their thinking skills. Moreover, it is essential to highlight the relevance of taking notes while performing scientific writing. We recommend that when using GenAI as aid, the researcher provides the tool with a complete summary, including context, identified study gaps, objectives, methodology, results, and conclusions. This ensures that the AI understands the scope of the study and can contribute relevant and coherent insights.
The triangulation of Generative AI tools is not only possible but also practical. This text presents a possible proposal that, we believe, should be adjusted within a few months, both by the continuous development of the tools presented here and by the emergence of new ones.
Related Publications
