THE generative artificial intelligence aspires to disrupt the traditional Internet. And he wants to do it quickly; All you have to do is do a quick Google search to find out. If earlier you had to go to web pages to solve questions, now in many … Sometimes the search engine itself solves it for you with an answer collected at the beginning of the page and which was created through the use of artificial intelligence trained with data from all corners of the network. And if you want to go broke, ChatGPT is there too. EITHER Gemini. either Grok. This rise in the power of robots capable of responding to everything is already affecting many pages whose sustainability depends on the visits they receive. AND Wikipediathe largest compendium of digital human knowledge ever assembled, with over 60 million articles and available in over 300 languages, is no exception.
The space, one of the most visited on the Internet in its almost 25-year history, has functioned for years as a kind of online container in which practically every imaginable subject fits: from the Spanish mastiff to the law of universal gravitation or baroque painting. But that doesn’t mean it’s invulnerable to the changes taking place on the Internet. Recently, the foundation Wikimediaa non-profit organization that supports and supports Wikipedia, shared a statement reporting that the digital encyclopedia lost 8% of human traffic in just one year. The main cause? AI.
“Wikipedia always came up number one or two in Google results when you searched for something. Now what Google does is it shows you the results of its own AI and a lot of the traffic is happy with what is said there. It doesn’t go any further or scroll any further. If the AI already tells you what you were looking for, why are you going to visit another site? “, explains Miguel Ángel García, lawyer, Wikipedian and former member of the board of directors of Wikimedia Spain, in a conversation with ABC. “A 10% drop in traffic may not seem too bad. But we must take into account that Wikipedia survives thanks to the donations it receives from the community;
Grokipedia
Wikipedia’s problems with AI don’t stop with declining visits. Last November, Wikimedia shared a post in which it directly asked artificial intelligence development companies to collaborate financially on the project. The foundation also recalls that “AI relies on knowledge resources documented by humans, such as Wikipedia”, and that the site is one of the most exploited by development companies to obtain information to train their machines. Because none of the chatbots currently present on the Internet would be able to respond to the queries posed to them if they had not previously been trained with data created by humans. For machine training, the digital encyclopedia constitutes a veritable gold mine.
Wikipedia contributors must already ensure that AI-generated content is not shared on the encyclopedia.
“In many cases, these apps are regurgitating Wikipedia, and I don’t think it’s profitable for them to let the site go, because they’ll have to make deals with others to get all this human-created information with which to train their machines,” says García.
Perhaps the best example of artificial intelligence’s reliance on Wikipedia can be found in Elon Musk’s Grokipedia, a space launched on October 28 that aspires to function as a sort of alternative to Grok’s AI-generated content. The problem is that of the million articles it contains, many, perhaps even the majority, have been literally copied from Wikipedia.
“We share our knowledge in a free and transparent way with the community, this tool (the Grokipedia) seems to respond more to personal objectives”, explains to this newspaper Matilde Cuadro, partner of Wikimedia Spain, referring to the favorable prejudices of the new encyclopedia based on AI towards Musk and the postulates he defends. Something that several researchers have already warned against. Cuadrado also points out that “almost all major AI models use Wikipedia because it has generated high-quality content, and removing everything an AI knows from Wikipedia is not easy.”
Copied articles
The encyclopedia is also very concerned that its articles end up being filled with artificial intelligence-generated content. Currently, the platform allows Wikipedians to use this technology to speed up their work, although all published articles must be created by humans. “New editors are already appearing and offering apparently very elaborate texts which we immediately realize were produced with AI. This worries many members of the community, because according to Wikipedia principles, all uploaded texts must be reliable and verifiable through the cited sources,” explains García.
The expert points out that, precisely, the contents created through the use of machines like ChatGPT stand out for the fact that they are not verifiable: “Often they don’t tell you where they got the information they share, and other times they invent it directly.” Fortunately, he notes that most of the time, it’s easy to spot when someone is trying to pass off an AI-created article as their own; It’s about how to “detect the ChatGPT stink”: “Anyone who has used a chatbot knows that machines tend to follow a more or less logical structure in their responses. First they present the problem, then they develop it and at the end they always give you some sort of conclusion, like “this is a fundamental problem for understanding the 20th century”, which are the typical sentences that AI leaves everywhere.
García also points out that chatbots, when trying to create entries with the same Wikipedia format, make great efforts to adapt to what is common in the encyclopedia: “For example, the normal thing with a person who has recently become famous is that their Wikipedia entry is in principle shorter. When they try to pass you one that has many sections and lots of text, chances are it was made with AI, because that’s something we’ve already detected and machines usually do.