the nature Fabio Petroni and his colleagues at London-based AI firm Samaya have reportedly developed a system called SIDE that analyzes whether Wikipedia links truly support the claims associated with them, and where they don’t, suggests better alternatives.
It may seem ironic that AI is being used to optimize links, in cases like the notoriously fancy ChatGPT, which often makes mistakes when using links. But we must not forget that there is much more to AI language models than chatbots
– said Noah Gianciracosa, who studies artificial intelligence at Bentley University.
SIDE is trained to recognize good citations of Wikipedia articles that are primarily promoted on the site and that receive special attention from editors and moderators. With the help of the formed verification system, it is able to identify data with incorrect links located on other pages. It can also scan the web for reliable sources and classification options to replace bad links.
The result is positive but not sufficient
To test the system, Petroni and his colleagues asked SIDE to suggest links to existing (verified) Wikipedia articles that the program had not seen before. In approximately 50 percent of cases, the article already included the reference that SIDE considered best. In other cases, suggest alternative sources.
The system can be useful in finding links that are potentially incorrect. Instead, the real question is what the Wikipedia community will find most useful
– said Alexandra Orman, a researcher specializing in the field of computer communications at the University of Zurich.
The developers tested this as well, and although there were more users who would use the AI-suggested source than the one originally mentioned in the article, the majority of survey participants didn’t care at all as long as the information was reliable. Thinking about this, in Orman’s opinion, currently, users prefer to search the sources themselves and check their authenticity.
