Over the past few weeks, many Wikipedians and chapters have been asking themselves how AI tools will affect Wikipedia. One Wiki author succumbed the temptation to accelerate content creation and regrated it immediately. Read for yourself what happened, and Wikimedia CH strongly advises against attempting to do the same:

‘Course of events: A Wikipedian was unable to attend our writing workshop in Bern and expressed the wish that the article on Swiss author Vreni Pfister should be expanded. Naturally, I am happy to fulfil such requests.

In the course of completing the article, I came across a book by the author, which is probably very popular, except in Bern, where it was not in the collection. There was also little or nothing to be found in terms of reviews. So I gave in to the temptation to have ChatGPT write me a short summary. Confident that the information was reliable, I reworded it slightly and integrated it into the article.

But now the book had piqued my curiosity. I ordered it as a little holiday read for the children at our guesthouse, Clos Magnolia. It arrived promptly: 68 pages, easy to read, suitable for children and young people. And here’s the kicker! The content has absolutely nothing to do with what ChatGPT had put together!

Great, as a seasoned wiki volunteer, I thought to myself, this is how you fall for the AI tool. Thank goodness I had the nose to buy the book in the flesh and read it.

I corrected the article immediately, of course. The old and new versions can be read below. In principle, you can smile about my experience, but you can also be frightened. I can only warn against taking the results of AI tools at face value and call for information to be used directly from the source in the good Wikipedia tradition.’

We would like to thank this user for sharing this somewhat embarrassing experience here to sensitise others. Wikimedia CH reminds to some important points in terms of using AI generated content in Wikipedia:

  • AI-generated content is generally not accepted on Wikipedia because we can’t verify the sources or determine if the content is copyrighted. Copyright laws typically protect works created by humans, not machines, but there are concerns with derivative works. For example, if someone uploads an AI-modified version of the Mona Lisa to Wikimedia Commons, it would be immediately deleted, and the user might be blocked. Additionally, some countries, like the UK and Hong Kong, classify all AI-generated content as copyrighted. If someone uses AI to summarize their own original content, that’s different.
  • It is also fine to validate an initial research step with AI. However, this must be checked for accuracy before publication. Anyone who disregards this rule runs the risk of having their content deleted by the community or having their user account blocked.
  • What our user learnt: Even with a simple summary, you can never rely on AI – AI-generated texts are generally not allowed on Wikipedia for the reasons mentioned above.

Wikimedia CH asks all voluntary Wikipedia contributors to respect these instructions.