Use of AI in peer review: a systematic review using PRISMA 2020, ROBINS-I and CASPe.

Abstract

Artificial intelligence, initially proposed as an artificial neuron through a mathematical model, dates back to 1943 by American scientists Warren McCulloch and Walter Pitts and marks the beginning of a major breakthrough for science and technology. The objective was to investigate and synthesize previous studies on the benefits and drawbacks of using AI in academic review without reducing scientific quality and ethics. The study was conducted using a systematic literature review (SLR) with a qualitative design, structured under the PRISMA 2020 statement, using RYYAN AI for screening in the Scopus (Elsevier) and Scielo.org databases, ROBINS-I to assess the risk of bias, and CASPe to assess methodological quality. A total of 833 primary studies meeting the eligibility criteria were identified from 2023 to 2025. Inclusion, exclusion, and screening criteria were applied according to the study objective, resulting in nine useful studies. Subsequently, in the risk of bias assessment, a result of 05 useful studies was obtained. In the final phase, a methodological quality assessment was applied, obtaining a total of 04 studies that adequately guide the research objective to be analyzed and synthesized. The results of the study indicate that the use of AI in general has a positive impact on the review of scientific papers, which is consistent with previous studies. However, some studies were contradictory because the incorporation of AI raises ethical questions that are not widely agreed upon and, in some cases, even controversial. Unexpected findings were that only 12% of scientific journals in Latin America have defined policies on the use of AI and that the ACTIVE LEARNING (AI) program can reduce the time it takes for an expert reviewer to review articles by 60%. As a key contribution of this study to the year 2025, it is proposed to regulate the use of AI to apply standardized processes and methods, always under human supervision, as well as to develop measurement tools that quantify the level of quality in AI review processes and restrict its use for making final decisions.