YouTube's current AI disclosure policy mandates that content creators inform viewers when they use AI to meaningfully alter or generate photorealistic content. The policy applies to AI-generated music but not to AI-generated animations of a missile in a fully animated video. Creators are free to clone their own voices for voiceovers and use AI tools for production assistance, such as creating or improving video outlines, scripts, or thumbnails. This results in scenarios where AI-generated content may not require disclosure, even if the entire video is AI-generated. For example, a 10-second video showing someone riding an AI-generated steed through a fictional swamp would not require disclosure, while an AI-crafted lute ballad about the experience would. This policy gap becomes more consequential for longer videos aiming to sway opinions on complex topics like geopolitics, where AI can generate premises, conduct research, and draft scripts without requiring disclosure. Source: arstechnica

YouTube’s concern is primarily about outright deception, requiring disclosure when a video shows a real person appear to say or do something they didn’t do or when a realistic scene that didn’t actually occur is presented. This helps prevent the cruder forms of disinformation. However, the policy does not address how AI can shape the very basis of creation, such as ideation, research, and outlining. AI-assisted research may supply answers but not domain mastery, and an AI-generated outline may lock the mind into a predetermined track before it has the chance to wander and arrive at more personal insights. Source: arstechnica

Noted science YouTuber Hank Green recently apologized to fans for his overreliance on AI, acknowledging that his use of AI as a research aid may have compromised his creative process. Green concluded that the pressure to produce content led him to rely heavily on AI for locating papers and other resources, which resulted in a lack of clarity in his own creative process. He stated that making more things does not necessarily make better things and that the dopamine from interacting with large language models is not healthy for him or the world. The result is likely to be fewer videos. Source: arstechnica