AI steals from other websites
Tech portal fails with AI experiment
01/27/2023, 7:24 p.m
At the latest with ChatGPT, artificial intelligence is no longer an abstract concept. A US website for technology topics is developing its own AI for writing texts. However, the attempt backfires.
The US portal “CNET” reports regularly on technology topics, but has been the subject of reporting for a few days. “CNET” began experimenting with artificial intelligence (AI) at the end of 2022. The news site had some of its articles written by a program it developed itself, rather than by humans. But for the time being the experiment has failed, such as the US broadcaster CNN reported.
Last week “CNET” put its AI project on hold, this Wednesday it was declared Editor-in-chief Connie Guglielmo then publicly. Because the attempt backfired on several levels. For one thing, the news site hadn’t publicly announced the project until US magazine Futurism reported on it. On the other hand, “CNET” shouldn’t be particularly transparent have been: The author’s line said that the texts were written by a “CNET Money” employee. Next to it was the name of a person who edited the text. “Futurism” complained that it was not immediately obvious that the author was an AI.
The lack of transparency was not the only problem. One of the texts also had errors in content, it was reported US magazine. In an explanatory piece on compound interest, the AI is said to have miscalculated. However, the list of accusations from “Futurism” does not end there: A recent investigation has shown that the “CNET” AI for a “Frankenstein principle” worked. The program is said to have copied its texts with sentences from other media.
A “handful” of errors
At an internal meeting, editor-in-chief Guglielmo finally commented on the allegation of a lack of transparency. “We didn’t do it in secret,” she reportedly said, according to The Verge. “We made it quiet.” After that, she also made a public statement. The AI program wrote 77 articles, which is one percent of all texts published at the time. Editors would have selected the topics beforehand, structured them and edited the finished articles again. After the erroneous “compound interest” text, the editors also scrutinized all the other pieces. He noticed a “handful” of other errors.
Guglielmo also explained that “CNET” developed the artificial intelligence itself. And although the project is currently on pause, work is continuing on it. At the same time, she emphasized that her application has nothing to do with the ChatGPT chat application, which is currently popular. The program recently caused a stir because it can write texts, answer questions and program itself. However, ChatGPT still regularly makes mistakes.
Nevertheless, AI is already widespread in journalism. The American news agency AP has been using AI since 2014. In the beginning, texts about group results were automated. This has also been practiced in Germany for some time. Some ntv.de articles are set to music using AI. Since 2019, a program has been writing thousands of match reports every weekend for “Fussball.de”, the German Football Association’s amateur football website. In doing so, it relies on tables and data provided by referees, for example. It should be noted that the texts were not written by humans. They are full of football phrases like “Team x had a razor-thin lead at the break”, but they are not incomprehensible either.
This is the current status that AI is currently providing. She can only spit out what she was previously fed. Therefore, the experiment by “CNET” says little about the future of artificial intelligence, but rather an example of how it should not be done. In an interview with ntv.de, the AI expert Albrecht Schmidt compared artificial intelligence with the emergence of the Internet. Even then, web browsers didn’t work right away and legally it was a “wild west” – in which everyone tried everything.