Wikipedia is a free, web-based, collaborative encyclopedia project that aims to provide accurate, neutral, and verifiable information to the public.
It is written and maintained by volunteers around the world and allows anyone with internet access to create and edit articles. Wikipedia's content covers a vast range of topics, from historical events and scientific concepts to popular culture and current events.
The platform operates under certain principles, such as neutrality, notability, and verifiability, ensuring that information is well-sourced and reliable. However, since anyone can edit, content accuracy can sometimes fluctuate, but articles are regularly reviewed and updated by editors to maintain quality.
And this time, following the rise in generative AI products, which happens after the release of ChatGPT by OpenAI that kickstarted the arms race for the development of Large Language Models, a lot of people began stuffing Wikipedia with convincing AI-generated misinformation, disinformation and malinformation.
The issue is so widespread in the encyclopedia that that Wikipedia's editors have assembled a dedicated task force to cleaning it up.
The group is called the 'WikiProject AI Cleanup,' and it's described as "a collaboration to combat the increasing problem of unsourced, poorly-written AI-generated content on Wikipedia."
The objective is to safeguard one of the world’s largest information repositories from the same kind of misleading AI-generated content that has affected Google search results, books sold on Amazon, and academic journals.
"A few of us had noticed the prevalence of unnatural writing that showed clear signs of being AI-generated, and we managed to replicate similar ‘styles’ using ChatGPT," said Ilyas Lebleu, a founding member of WikiProject AI Cleanup.
"Discovering some common AI catchphrases allowed us to quickly spot some of the most egregious examples of generated articles, which we quickly wanted to formalize into an organized project to compile our findings and techniques."
The methods they use are the same as what others have used to find AI-generated content in scientific journals and Google Books.
One of the most used method, is to look for phrases commonly used by ChatGPT.
For example, there was one Wikipedia article about the Chester Mental Health Center, which in November of 2023, its revision included the phrase "As of my last knowledge update in January 2022," a sentence that is often used by AI to refer to the last time the large language model was updated.
But scouting for AI-generated content isn't that easy.
Most other instances are much more difficult to detect.
Lebleu and another WikiProject AI Cleanup founding member who goes by Queen of Hearts, said that the most "impressive" examples they found of AI-generated content on Wikipedia so far is the article about the Ottoman fortress of Amberlisihar.
The page had more than 2,000 words, and was filled with detailed information within its many paragraphs, divided into sections about its name, construction, and various sieges it faced.
Information also included restoration efforts after it “sustained significant damages as a result of bombardment by Russian forces” during World War I.”
But here, the team found one mistake that quickly made them realize that it was AI-generated.
"One small detail, the fortress never existed," Lebleu said.
Aside from a few tangential facts mentioned in the article, like that Mehmed the Conqueror, or Mehmed II, was a real person, everything else in the article is fake, including the citations.
"The entire thing was an AI-generated hoax, with well-formatted citations referencing completely nonexistent works."
WikiProject AI Cleanup's job isn't only to find and remove AI-generated text, because members must also find images AIs create.
One example, is the article about Darul Uloom Deoband, a real Islamic seminary in India.
At one point featured an image, which like many Wikipedia articles about people living centuries ago, depicts period-appropriate painting related to the subject of the article.
But upon closer-examination, the image has people with mangled hands and a foot with seven toes.
It's worth noting though, that the WikiProject AI Cleanup page that tracks AI-generated images on Wikipedia clarifies that it doesn’t remove AI-generated images from Wikipedia just because they are AI-generated, because in some cases, the AI-generated images are actually appropriate.
If the article is about or mentions an AI-generated image, it makes sense to include it.
Lebleu said that the editors have discussed using AI to detect AI, but they ultimately consider relying on human knowledge to detect them.
As a result of this approach, Wikipedia is at this time, considered one of the best major internet services that has a focus on detecting and filtering AI-generated content, simply because the site has always relied on human volunteers to review new articles and track down any claims they make to reliable sources.
Opposed to social media platforms and search engines, many of them use a combination of human team and AI, but have repeatedly reported failure in catching misleading AI-generated content.
Many of them that went unnoticed, were only removed after they were reported, or after receiving complaints from users.
"Wikipedia articles have a more specific format (not only in terms of presentation, but also of content) than Google results, and a LLM that isn't familiar with it is likely to produce something that is much more easy to spot," Lebleu explained.
"Things like verifying references also help: as Wikipedia aims to be a tertiary source (synthesizing other sources without adding original research itself), it should theoretically be possible to verify if the written content matches the sources."
"While I'd like to think Wikipedians are decent at detecting and removing AI content, there is undoubtedly a lot that slips through the cracks and we're all volunteers," said Queen of Hearts, whose real name is Charlotte. "While major companies' failure to detect and remove AI slop is concerning, I believe they could do better than us with properly allocated resources."
What makes AI-generated content so harmful is that, it can make fake information feel legit and so real.
And being on Wikipedia means that the fake information can be the source of many studies and research. And this can cause further harm that if not properly corrected, can create confusion, academic and professional impact, amplification of biases, and a lot more issues, including the rewriting of history.
In the end, it's a loss of credibility for Wikipedia.
Wikipedia's reputation as a trustworthy, community-driven source of knowledge could suffer, leading people to lose confidence in the platform.
And since Wikipedia is a highly ranked source of information in search engine results, AI-generated falsehoods could spread across the broader internet ecosystem, further amplifying the problem.
Due to how massive Wikipedia is, and the many languages it has, many of the fake articles can stay undetected for months.
Even if someone was using an LLM trained on a corpus of data relevant to the Wikipedia article, it could generate text that reads well and with correctly formatted citations of real sources, it still wouldn’t be able to correctly match a citation to a specific claim made in a specific body of work.