AI, my new research assistant?

At the end of March, I had the opportunity to speak at a French roundtable event "Documation - iExpo" about the challenges and opportunities of AI in Open Science matters. Thomas Parisot, Deputy Managing Director of Cairn joined me as a speaker and we discussed specific use cases of AI in Open Science for 30 minutes. Since this topic is closely related to your/our activities (Polaris OS and Sirius), I had summarized what we discussed and what I learned (views are my own).

The article is also available in French.

Being a vast subject, we decided to focus on specific AI challenges:


1.     Writing help


As a researcher working in an increasingly internationalized field, one of my main challenges is to increase the impact of my articles. To achieve this, it is essential that my articles are published in English using the best possible terminology. AI-powered tools like Writefull or Paperpal can assist researchers by suggesting impactful terms and improving the quality of their writing. “… non-native English speakers will have an easier time overcoming the language barrier. They will be able to produce high-quality research papers without worrying about grammar or syntax issues. Moreover, AI-assisted writing can help researchers save time, allowing them to focus on refining their ideas, framing their arguments better, and conducting more in-depth analyses.” as said Saikiran Chandha in an excellent article published on The Scholarly Kitchen blog.


Another challenge I face is identifying relevant citations for my articles. With the vast number of articles being published each year (around 2.5 million), it can be difficult to keep up with the latest research in a specific field. However, a solution as Scite recently launched an AI-powered feature that helps researchers to identify the most appropriate citations based on the text they are writing.


2.     Discovery and recommendation


In our work at MyScienceWork, we frequently use scholarly databases and platforms to search for articles, datasets, and other relevant content. However, the sheer number of articles available can make it difficult to find relevant results within our limited time. Therefore, platforms that make articles available (whether they are from publishers or open repositories) must offer readers results that are tailored to their research needs.

Several years ago, semantic search (using full-text analysis) showed promise for searching large collections of documents. However, while this type of search greatly expanded the field of discovery, it did not always produce relevant results. Now, the challenge is to use AI to develop search engines and recommendation tools that can provide more relevant results. Several approaches exist to hemp the users to identify the most relevant articles:

1.     Using data from other readers on the same platform: if a user has viewed the same article as me and then viewed other related articles, the platform can recommend those articles to me.

2.     Relying on similar metadata based on keywords, authors, co-authors…

3.     Using Natural Language Processing techniques to identify similarities and connections between publications.


Another challenge that researchers face is keeping up with the latest advances in their disciplinary field, as the number of publications continues to increase each year. To address this challenge, new AI-powered tools such as Opscidia, Scholarcy, and SciencePOD can produce automatic summaries of multiple articles in seconds. These tools can be incredibly useful for quickly obtaining a state-of-the-art understanding in a specific field.


3.     Assessment and workflow 


Having more articles also means having more journals, and with open access, it becomes increasingly difficult for researchers to identify the best option for their articles, considering the APC policy of the journals and the challenge of avoiding publishing in predatory journals. Open Access Journal Finder is a great tool for researchers to find the right journal and obtain information about APCs.


In the evaluation process of articles, another challenge arises. With around 2.5 million scholarly articles published every year, there is a lack of reviewers to review them all.

By using AI, the automatic review of certain criteria can help to reduce the number of publications that will be sent to peer reviewer. AI helps to automate certain tasks, such as what StatReviewer proposes and states: “StatReviewer is an automated review of statistical and reporting integrity for scientific manuscripts. Manuscripts are scanned, and a report is generated. The report will either resemble an actual peer review or checklist, depending on the guidelines specified by the journal."


4.     Automated translation 


Open science movement has brought an old challenge back to the fore: the richness of research, and the ability of researchers to publish in different languages. The hegemony of English is indisputable, and we understand the reasons for it - to be accessible to the greatest number, to have a greater impact, and so on. Nevertheless, publishing in other languages brings richness to research work, which is essential to maintain and encourage. How can these two subjects be reconciled? Machine translation can play an important role with a major challenge: translation algorithms today are mainly trained on generic data, and the translation of technical terms and/or concepts that are specific to certain languages often escapes them.

Although academic translation tools such as AJE Academics are emerging, and publishers are proposing them to their authors (see SpringerNature Press Release), human control is still necessary to confirm the translations performed by the machine.


5.     Fraud detections


No one could have missed the prowess of CHAT GPT and the questions it raises, including transparency and fact-checking. The world of research is facing similar challenges, mainly with the Open Science movement that encompasses the aspect of scientific integrity. Recent news has given us examples of challenges in detecting fraud (such as manipulation of images), predatory journal systems, and problems at the evaluation level (see Hindawi News). The rules of transparency that currently accompany open science policies have highlighted this as a significant issue, particularly at a time when trust in science has been shaken, making it an essential pillar for the progress of our societies and democracies. How can AI help bring more transparency by helping publishers identify fraud in the publishing process?

AI products already help researchers and publishers to:

1.     Identify image manipulations (see Proofig)

2.     Detect whether texts have been written by a human or by an AI. This is mostly done by existing plagiarism detection tools such as Compilatio


6.     Analytical purpose


At MyScienceWork, research analytical purposes are one of the objectives of most of our projects (Polaris OS and/or Sirius) because research information is key to understanding the world we live in. Decision-makers need to feed their thinking with factual key indicators. Understanding their institution’s research activities, their impact, what other research institutions are working on, the global trend of research activities around the world, etc. are essential to be effective and speed up innovation.

With Open Science, research stakeholders realized that opening access to research outputs and data can help a lot to achieve this. Tools based on AI such as Sirius can greatly help to produce analytical data indicators (trends, expertise, competencies, new research fields, market analysis, etc.).


As a conclusion, AI is a great opportunity to assist researchers and research publishing stakeholders overcome the challenges they face. AI seems to be a useful research assistant. To be followed…