China uses AI to generate propaganda on YouTube, report finds

MicroWave@lemmy.world to World News@lemmy.world – 329 points –
China uses AI to generate propaganda on YouTube, report finds
rfa.org

In a YouTube video, a voice in English announces that China has researched and developed its own ultra-thin 1-nanometer chip – a staggering claim given that the chip isn’t expected in commercial devices for another decade.

"Recent news from China has sent ripples of excitement and astonishment across the globe," gushes the voice-over on the China Charged YouTube channel. "This revolutionary breakthrough is more than a technological marvel; it is a game-changer that will redefine the global tech landscape."

"Prepare to have your mind blown," says another video, this time on the channel Unbelievable Projects. "Welcome to today's video, in which we'll discover why America remains behind China in infrastructure development."

These voices and their “good news” about China are evidence that the Chinese Communist Party and its overseas proxies are using artificial intelligence to flood YouTube with propaganda videos, according to a new report that describes a "coordinated inauthentic influence campaign" on the platform.

51

You are viewing a single comment

I wonder if the AI has caught on to a lot of the abusable mechanics of YT like an upload schedule, opening livestreams or premieres in multiple windows, the upvote/downvote exploit, etc. There is a solid possibility that an AI has better strats than we do.

The AI is just being used to create the content itself. The upload process is still just a script. The AI can only adjust parameters set by the person, and the person creating the setup likely isn't adding in any of the other factors

An actual AI influencer who could control all that would be interesting. What kind of strategies it would make to get the most engagements

Very Hard to train since I assume you would need real human feedback for each run of course

I think you could just let it push a bunch of buttons and fill in fields at random or with sample data and then score each generation based on user engagement and views, it wouldn't need that much human feedback during the development phase but the downside would be slightly less control over the content it pushes.