Research: Artificial Intelligence Can Enhance the Reliability of Wikipedia

Novelist Nicholson Baker refers to Wikipedia as a data compilation, which its developers call a collection of all the world's knowledge, and he calls it "an incredible thing. It's vast, it's unique, it's careful, it's a little messy, it's interesting, it's shocking, and it's full of fierce controversy - it's free, and it's fast."

Writer Oscar Auliq-Ice describes Wikipedia as a "revolutionary resource that has changed the way people access and share information."

Some people acknowledge this massive online resource, which as of this week contains over 6.7 million articles (in English), composed of over 4.3 billion words, as a grand but flawed endeavor.

Environmental expert Steven Magee states, "Wikipedia is like a flower bed, mostly beautiful, but with some ugly weeds."

Humorist Stephen Colbert points out the collective nature of Wikipedia, where anyone can voluntarily contribute information. He suggests, "Whenever I need to find knowledge, I go to Wikipedia... or when I want to create some knowledge."

However, for some, Wikipedia is a bitter pill. Tara Brabazon, Dean of Graduate Research and Professor of Cultural Studies at Charles Darwin University in Australia, says, "I don't need wireless access to Wikipedia. I'd rather fry my own sausages than navigate a website that has a longer entry on 'Klingons' than on 'Latin'."

Although often seen as a commendable and expedient source for obtaining information quickly, users are always advised to do their due diligence and not rely solely on one source of information. They should look at other websites, explore article links, and perhaps most importantly, check the reference sources listed at the end of each Wikipedia entry.

Experts from around the world regularly contribute to Wikipedia, mostly following guidelines of neutrality and using reliable sources. This system generally works well, although there is always room for improvement.

This week, the journal "Nature Machine Intelligence" published an article titled "Improving Wikipedia Reliability with Artificial Intelligence," which introduces an AI company based in London that aims to enhance the reliability of Wikipedia's reference system. They achieve this by examining sources, determining which ones are accurate and which ones are problematic, and then providing their own recommendations to achieve this goal.

Fabio Petroni, co-founder of the knowledge discovery platform Samaya AI, states, "With the help of artificial intelligence provided by information retrieval systems and language models, the quality of references can be improved. Machines can assist people in finding better citations, which is a task that requires understanding language and mastering online search."

His team trained the model based on a large dataset of Wikipedia articles and then used it to review previously unscanned articles. The system analyzes sources and provides alternative reference websites, which are then checked by Wikipedia users.

Researchers found that when an AI system called SIDE classified Wikipedia's sources as unverifiable and provided its own alternative sources, users preferred SIDE's recommendations 70% of the time.

In half of the cases, SIDE recommended the same sources that Wikipedia provided as its first reference source.

Petroni states, "We have demonstrated that existing technology has reached a stage where it can effectively and practically support Wikipedia users in verifying claims."

He adds that future research will focus on expanding Wikipedia's reference sources beyond internet text, such as images, videos, and print publications.

Petroni says, "We hope this work can be used in a broader context to help people fact-check. More generally, we believe this work can contribute to making online information more reliable."