Relying solely on information from a Wikipedia page can be risky, which underscores the importance of cross-referencing with the original sources listed in the footnotes.
Nevertheless, even primary sources can sometimes prove unreliable. To enhance the trustworthiness of Wikipedia references, researchers have introduced an AI system geared towards scrutinizing citations within the platform that may raise doubts.
Strengths
This program, named SIDE, serves a dual purpose: it verifies the accuracy of primary sources and suggests alternative ones. However, it’s essential to note that this AI operates with the presumption that a given Wikipedia statement is accurate. Consequently, while it can assess the validity of a source, it cannot independently verify the claims presented within an entry.
In a study, it was observed that people favored the AI’s recommended citations over the original sources in 70% of cases. Notably, the researchers discovered that in almost half of the instances, SIDE suggested a source that was already the top reference on Wikipedia.
Furthermore, in 21% of cases, SIDE demonstrated its proactive approach by proposing a reference that had already been deemed appropriate by human annotators in the study.
Weaknesses
Although the AI appears to offer effective support in fact-checking Wikipedia claims, the researchers acknowledge that there may be alternative programs with the potential to surpass their current design in terms of both quality and speed.
It’s important to recognize that SIDE has its limitations, primarily focusing on web page references. In reality, Wikipedia draws on a diverse array of sources, including books, scientific articles, and multimedia content such as images and videos. Beyond technical constraints, the fundamental premise of Wikipedia allows any contributor to assign references to a topic, potentially introducing bias depending on the nature of the subject matter.
It’s a well-established fact that any software, particularly an AI system reliant on training, has the potential to reflect the biases of its creators. The data employed for training and assessing SIDE’s models may indeed be constrained in this respect.
Nevertheless, the advantages of employing AI to enhance the efficiency of fact-checking, or at the very least, to utilize it as an assisting tool, could extend to a broad range of applications in various domains.
