After a test that lasted approximately 5 months - from 1 August 2023, launch date of the "Beta" version, to 16 January 2024, date of release of definitive version – Elsevier can finally include “Scopus AI” among its official digital publishing products. These days, major research bodies, academic institutions and scientists around the world are busy with webinars and presentations that should illustrate how Scopus AI can help them with content selection and with what quality assurance.
The starting point is the Elsevier database, i.e. Scopus. Scopus currently hosts over 29.200 peer-reviewed journals from over 7.000 publishers, meticulously reviewed and selected by an independent expert committee aka the “Scopus Content Selection and Advisory Board” (CSAB). Elsevier is a joint-stock company which, since 1993, has been part of the Reed Elsevier group, then RELX plc, a British multinational operating in the field of scientific information and analysis. The group, over the years, has carried out various acquisitions and mergers, being able to count on an ever-increasing number of databases and has thus become the world's largest publisher of academic articles.
Scopus AI draws from the metadata and abstracts of published database documents since 2013. This selection minimizes the risks of false information generated by artificial intelligence and allows you to always refer to the sources used in a completely transparent manner. Scopus AI suggests paths for further investigation, suggesting, among the “fundamental documents“, those with the greatest impact on the selected topic. Then there is the help of "conceptual maps“, i.e. keywords coming from research abstracts connected to each other in a scheme that always allows us to trace relationships with other relevant research areas, even those which, on a superficial human reading, might seem out of place. Therefore, the functionality existsindustry experts” that is, the ability to select the over 19,6 million author profiles in Scopus to find the best researchers linked to the queries and generate a summary of their work and contributions complete with an explanation of why each individual was selected.
“It is still necessary to use a certain critical judgment when examining the results – said Bruno Augusto, doctoral student at the University of Aveiro, among the first to test the final version of Scopus AI – since the tool provides you with all the references he quotes from, you can look at them, see if they fit what you need or if there is a better outcome.” By all accounts, the biggest change is the speed with which you can obtain the reference of the article you need. Bruno confirmed this with an example from his recent work: “Last week I was looking for documents on the urban areas most vulnerable to air pollution and emissions. So I just searched for that exact term directly, instead of searching for “emissions, air quality, urban areas.” Scopus AI provided me with a couple of quick quotes. One of them wasn't what I needed, but the other was exactly what I was looking for. I checked the article to make sure it was a good reference and then I was able to include it.”
All this enthusiasm for the entry of artificial intelligence into the world of scientific publications could cause alarm among non-experts, also in light of the recent case of the article by Chinese researchers, written with a AI chatbot, which reported the beginning of the typical software response: “Of course. Here is a possible introduction for your topic.” In fact, the sentence left little doubt: the authors of the article had certainly used an artificial intelligence-based system to write the introduction to their work. However, this is not a cause for concern. First of all because, at most it is about lack of transparency by those researchers and a lightness by some reviewers. Furthermore, it should be specified that the researchers certainly did not use artificial intelligence to draw conclusions, that is, to be completely replaced in their main task, but they made use of a tool that reduces the time in carrying out some practices ancillary to the research.
What is generative artificial intelligence and how does it work?
Generative artificial intelligence is a branch of artificial intelligence that aims to create computer systems capable of autonomously generating texts, images, programming code, sounds, videos and algorithms of all kinds. These systems use machine learning and artificial neural networks to learn from training data and generate new content that can appear to be created by a human.
A particular category of generative artificial intelligence, known as "natural language processing", commonly abbreviated as LLM (from the English "Large Language Models") subjects the machines to pre-training which consists of exposure to a vast amount of texts coming from archives that can be classified as "big data", i.e. containing an enormous quantity of data. From these texts, LLMs can learn the syntactic structure of the language, the semantic field of words or general knowledge concepts. In this way, these systems are literally able to understand a question and generate answers grammatically correct e semantically adequate. An example above all is "chatGPT".
Scopus AI is one such tool, but it is not accessible to everyone, like chatGPT. Instead, it is a module that can be added to a institutional Scopus subscription, as supplement. Therefore, requests cannot be made by individual researchers and/or scholars, but it is necessary to involve your organisation, your university, the body for which you carry out research.
