Wikipedia: Test of AI-generated article summaries stopped for the time being

Those responsible for Wikipedia have canceled a project in which AI-generated summaries were to be prepared for Wikipedia articles. This can be seen from an entry in the online encyclopaedia's section where technical aspects are discussed. At the beginning of last week, a representative of the Wikimedia Foundation announced that Wikipedia employees would be asked what they thought about it and whether they wanted to work on it. On the encyclopaedia's mobile site, a small proportion of visitors were to be given the opportunity to view the AI summaries. After much criticism, the U-turn followed on Wednesday.

"Yuck"

As those responsible for the project explain on a dedicated page, the aim is to make the encyclopaedia more accessible. Previous projects with this goal have only had limited success – including the section with texts in plain English. They also make it necessary for authors to rewrite finished articles in simple words, which is tiring. Meanwhile, an initial – experiment limited to just a few people – showed that AI-generated summaries were met with a great deal of trust and that the overall reaction was much more positive than expected. All AI summaries are to be checked by humans.

In contrast, the word "yuck", which can be translated as "yuck", appears several times under the presentation of the project on June 2. There are various references to the risk such AI-generated content poses to Wikipedia's good reputation, for example. In particular, the error-proneness of AI content is repeatedly pointed out. Others criticize the handling of the project and the lack of information in advance. The Wikimedia Foundation, which operates Wikipedia, has now come up with some insightful words: the project will not be scrapped, but the reactions will be taken seriously and the topic will be addressed again soon, according to the entry. Until then, the project will not be continued.

This does not mean that the project is over; the Wikipedia operator reserves the right to take it up again. Meanwhile, AI summaries generated for the limited experiment can be viewed on GitHub. Just two months ago, the Wikimedia Foundation made public how resource-intensive the training of AI technology for its encyclopaedia is. The scrapers used to scrape Wikipedia content for this purpose place an entirely different burden on the site than humans do. At the time, the organization criticized that this led to "growing risks and costs" without any added value. It later introduced AI tools to support moderators.

Read also

AI scrapers strain Wikipedia: 50 percent more bandwidth for multimedia requests

(mho)