AI-generated fake science videos being recommended to children on YouTube, BBC reports
BBC journalists have found that YouTube channels are using AI to create false ‘scientific’ videos, which are being recommended to children alongside legitimate educational content.
BBC journalists have found that YouTube channels use AI to create false ‘scientific’ videos, while YouTube recommends these videos to children alongside legitimate educational content.
The investigative team has identified over 50 channels in more than 20 languages that spread disinformation disguised as science, technology, engineering, and maths (STEM) content. The creators labeled their ‘bad science’ videos as ‘educational content’, making them more likely to be recommended to children. These videos used sensationalist commentary, catchy titles, and dramatic images to attract viewers and make wild claims. Many channels have over a million subscribers, and millions often view videos.
The team also revealed that most of these videos used AI to generate text and images, scrape information from a web page, and manipulate material from real science videos. This explained how these channel creators could publish multiple videos each day.
The team created children’s accounts on YouTube’s main site to test whether and how the ‘fake science’ videos would be recommended to children. After four days of watching legitimate science education videos, the accounts started to be recommended AI-generated ‘fake science’ videos. More similar videos were recommended when any ‘fake science’ videos were clicked.
The BBC contacted several companies producing this kind of misleading AI content. One replied that their content was for ‘entertainment purposes’. YouTube responded that it suggested children under 13 years old use ‘YouTube Kids’, which has a ‘higher bar’ for the quality of videos.
Why does it matter?
These misleading AI-generated videos would appeal to children’s natural curiosity. They run the risk of confusing children about what is true. More powerful AI will make such misleading content easier to produce and more difficult to detect. There is an urgent need to understand the real impact of AI-generated content on children’s intelligence and learning, with clear measures in place to address the foreseeable risks.
It is also questionable what responsibility hosting platforms such as YouTube should take, especially when they profit from the ads that appear alongside these misleading videos. The more people watch the videos, the more money the channels make from the ads. YouTube also benefits from high-performing content, as it takes around 40% of the profits from advertising on someone’s channel.