Is Wikipedia in danger because of AI?

Widely used to train artificial intelligence models, the very existence of Wikipedia could now be threatened by them.

If we are far from the consensus in the community of Wikipedia editors as to the attitude to adopt towards the development of artificial intelligence, it obviously does not leave them indifferent. A conference on the issue, held last March, however, brought out a position that most can agree with: they want knowledge to continue to be created by humans.

Wikipedia, a treasure trove for LLMs

To train an AI language model, Wikipedia is simply the best possible playground. The online encyclopedia simply ticks all the boxes. It has more than 61 million articles, is available in 334 languages, presents sourced and verified information, updated almost in real time. In addition, the articles are clearly formatted and chaptered, and above all, the whole is freely accessible and free of charge. To top it all off, there is also no limit on the number of posts you can view there.

It is therefore not surprising that the chatbots that have emerged in recent months have made generous use of this formidable database, to the point that it is not an exaggeration to say that without Wikipedia, generative AI would not exist, at least not in its current form.

But now, nothing seems to prevent a future where one of the students takes the place of the master.

Will the encyclopedia be overtaken by the AI?

The threat seems serious enough that last March, a videoconference brought together many editors of the encyclopedia to discuss the risks posed by the development of generative AI. If some of them were rather confident about the possibilities offered by AI to develop the audience, but also to enrich the articles of Wikipedia, they were clearly not unanimous.

Without even distorting Wikipedia, AI already has serious advantages on the site. An editor interviewed by the New York Times thus explains that a future in which an artificial intelligence synthesizes, sources and compiles on a single site all the articles posted on the Internet almost in real time does not seem very far away. Then it probably wouldn’t take long to overtake Wikipedia, at least in terms of the amount of content. It must be said that the AI ​​​​does not have these very human defects that are the time for reflection and argumentation. For him, such a site would not even need to be up to Wikipedia, only passable.

Moreover, without going that far, chatbots already have a significant advantage over the encyclopedia: fluidity. It is indeed infinitely easier to ask a question to ChatGPT than to read a whole Wikipedia page in search of a simple information. But for the editors of the encyclopedia, being overwhelmed by artificial intelligence is potentially catastrophic. While chatbots’ responses are generally correct, they frequently continue to hallucinate and are very poor at citing their sources, sometimes going so far as to hallucinate their sources, two deadly sins in Wikipedia’s founding principles.

To be honest, it is already unlikely that Wikipedia, the last vestige of a utopian vision of the Internet which refuses the capitalist vision of all other sites with a comparable audience, does not display advertisements, does not collect information personal and values ​​the truth, has already survived this long.

Source : New York Times