Part of this article was written using good old human brains!
After the teaching community, the world of research is also being shaken up by the arrival of artificial intelligence (AI). Researchers must now review scientific articles “written” by peers, but which may have been largely written by AI. Graduate students can use AI to help them write part of their master’s thesis and doctoral thesis. How to react?
We are two researchers from the National Institute of Scientific Research (INRS), one of whom has been using and developing AI algorithms as part of his research since 2004 – mainly for the processing of geoscientific data and the other introduces them gradually in order to maximize their efforts to reconcile all spheres of academic life. Above all, we want to inform our peers that these tools exist.
Few existing university policies
Few universities have published clear policies on the use of AI in research. Since the official release of ChatGPT from OpenAI in 2022, we are however dealing with a significant tidal wave for the scientific community, which is important to reflect on. That said, the more time passes, the more the gap widens between researchers who use it and those who are not yet using it.
Among the universities that have started to think about policies governing the use of generative AI is the Massachusetts Institute of Technology. One of the key recommendations made by the university is to declare when and what AI was used during ideation, research and writing.
Already thousands of subscribers to the La Conversation newsletter. And you ? Subscribe today to our newsletter to better understand the major contemporary issues.
Unlike conventional computing resources (Stackoverflow, Google, etc.), there seems to be a taboo in recognizing the use of a generative AI algorithm in references or acknowledgments of scientific articles and reports.
The efficiency gain is surprising
Just as access to personal computers was a revolution in productivity, the arrival of generative AI has increased productivity in certain areas of research.
The most well-known generative AI is ChatGPT and it has several “sub-GPTs”, including “ScholarGPT”. The latter can carry out literature reviews on given subjects. He can evaluate articles, dissertations, theses, books. He can review financing applications. It can synthesize career of life and write letters of recommendation.
The “Logo Creator” sub-GPT can quickly generate logos, while the “DALL-E” sub-GPT can generate quality diagrams and illustrations. As these different programs are all interrelated, no more copying and pasting from one to another.
The “Perplexity” AI, for its part, is capable of providing elaborate responses based on several generative AIs. When answering a question asked, Perplexity also provides the references it used to generate its answer.
Is this ethical? Is it fair?
We believe it is important to use this new technology carefully and that it still requires a human brain to verify the answers, because for now, generative AIs are oracles that are based on the probability of the next word in a sentence. They are therefore not aware of the general framework of the question. Work is being done in this direction to add context to the response as in ChatGPT 10.
Most programmers use generative AI because it is an integral part of programming environments like Google Colab, VSCode or PyCharm. It’s hard now to differentiate which line was completely written by a human, or if they used suggestions from their favorite AI.
Most scientific journals and universities have started to question themselves to develop policies to regulate the use of AI. However, to the extent that the tool is there, should we wait for policies to be put in place before using it?
How ChatGPT can optimize the work of researchers
On December 3, 2024, we asked ChatGPT this question: “Can you write us a text that explains how ChatGPT can optimize the work of academic researchers? » Here is his response:
The rise of artificial intelligence (AI) offers academic researchers unprecedented opportunities to optimize their research, analysis and scientific communication activities. ChatGPT, an advanced natural language processing model, can act as a versatile assistant for researchers, allowing them to save time, improve the quality of their output and foster innovation.
-1. Assistance in documentary research
ChatGPT can quickly provide summaries of concepts, summaries of scientific articles, or contextual information on a given topic. This makes it possible to: a) Identify current research trends; b) Quickly understand interdisciplinary fields and c) Find relevant references (even if they need to be checked through databases like PubMed or Scopus).
Example: An ecotoxicology researcher can request an overview of the latest advances in the study of endocrine disruptors and obtain a summary of the main studies.
2. Writing and revising manuscripts
ChatGPT can help with the writing of articles, reports or presentations: a) Preliminary writing: Generate a draft introduction or discussion based on the results provided; b) Linguistic improvement: Rephrase complex sentences to make them more clear and concise and c) Scientific translation: Translate articles or abstracts into other languages while maintaining scientific accuracy.
Caution: The use of ChatGPT to write parts of scientific manuscripts must comply with the ethical guidelines of academic journals.
3. Data analysis and visualization
Although ChatGPT does not replace specialized tools like R or Python for complex quantitative analyses, it can: a) Explain statistical concepts or mathematical models; b) Propose approaches to analyze a dataset and c) Assist in the interpretation of the results.
Example: If a researcher has data on stress biomarkers in fish exposed to contaminants, ChatGPT can suggest methods to identify meaningful correlations.
4. Optimization of scientific communication
Communication is a pillar of research, and ChatGPT can be used to: a) Develop popular science for non-specialist audiences; b) Prepare slides, scripts or conference posters and c) Formulate responses to reviewers’ comments when revising articles.
Example: For a presentation intended for the general public, ChatGPT can reformulate technical terms to make them accessible.
5. Idea generation and brainstorming
ChatGPT can be used as a brainstorming tool to generate innovative ideas or explore new hypotheses. For example: a) Propose alternative experimental approaches; b) Identify potential gaps in the existing literature and c) Suggest interdisciplinary collaborations.
6. Time saving and automation
By automating repetitive tasks, ChatGPT allows researchers to focus on high value-added activities: a) Standardized responses to academic emails; b) Writing cover letters for scholarships or projects and c) Rapid creation of lesson plans or teaching materials.
7. Limitations and ethical considerations
Despite its many advantages, ChatGPT has limitations: a) Verification of information: ChatGPT may generate incorrect or unsubstantiated answers. Researchers must always validate the information obtained; b) Intellectual property: AI-generated contributions should be reported to avoid any ethical issues and c) Data sensitivity: It is not recommended to use ChatGPT to process confidential or sensitive data.
Should we use it or not?
As an AI tool, ChatGPT can greatly facilitate the work of scientists by providing them with flexible and rapid support. Although it does not replace human expertise, it acts as a lever for efficiency, allowing researchers to devote more time to scientific exploration and innovation.
However, judicious, ethical and supervised use remains essential to maximize its benefits while minimizing the risks.
The advent of AI in research risks being just as important as the invention of the combustion engine! We will not stop progress, so it is better to train as we did in the past with the use of the computer or the Internet. It is essential that universities quickly develop policies for using these new tools before inequality in productivity digs even deeper gaps between all international research institutions.
While waiting for the bell to sound, we wish you a good initiation!
