The idea of more accessible AI generated videos became a reality after OpenAI’s second Sora 2 AI Generation was released on Sept. 30. OpenAI grants certain countries, including the United States, public access to Sora 2 AI.
Although the public access to Sora 2 AI Generation isn’t established everywhere yet, YouTube has already established in their “How YouTube Works” content policy that the company allows all video content, including AI generation.
According to the “How YouTube Works” content policy, videos should be “disclosed” when they’re made from Generative AI services, and it also mentions how the company’s updated privacy policy allows individuals to “request” that AI-generated videos that feature recognizable people can be removed, though only after YouTube moderation considers a variety of factors.
In addition, YouTube introduced the content ID: a system that allows YouTube to identify videos that infringe on a creator’s copyright; a new policy to alert performers, creators, musicians, athletes, and artists when their voices and/or faces are being featured on AI content on YouTube.
Meanwhile, according to Radius Insights’s poll, which was used in Youtube’s AI-Generated content policy in May 2024, 74% of creators approve of the idea of having guidelines regarding generative AI video content.
However, the “How YouTube Works” AI-generated content policy also confirmed that YouTube uses AI in its moderation system to “increase the speed and accuracy” of video moderation.
Likewise, Google Support stated that YouTube employs humans to review content that AI has flagged as a violation of YouTube’s content policy. Google also mentioned that, “When our systems have a high degree of confidence that content is violative, they may make an automated decision.”
While YouTube’s moderation system’s inclusion of both human and AI does make sense to increase the efficiency of YouTube moderation, AI is still a fairly recent innovation. Allowing it to moderate videos itself seems premature in the current stages of AI’s accessibility to public use.
In addition, the May 2024 survey administered by Radius Insights which was used in Youtube’s AI-Generated content policy predates the release of Sora 2 AI Video Generation. After considering that YouTube creators would be more aware of the impact of Generative AI videos after Sora 2’s release, it would make more sense if YouTube revisited the Radius Survey regarding AI content guidelines and updated the statistics. Given that the Radius May 2024 survey isn’t easy to search on the web specifically, it would be helpful for YouTube to include a hyperlink so people can directly receive this information.
As for the mandate that all AI-generated videos are disclosed as AI-generated, many AI-generated videos are made to look realistic. Even some videos that do not have any disclaimer of being AI-generated in the content description are clearly AI-generated. Although YouTube occasionally places its own automated label that a video is AI-generated, errors can be made, and there is a chance that YouTube will not be able to give an automated disclaimer to every AI-generated video.
Sora 2 AI’s own content policies raised questions when its generation capabilities were covered by the New York Times article, “OpenAI’s Sora Makes Disinformation Extremely Easy and Extremely Real.”
For instance, although Sora 2 AI Video Generation does not permit the generation of any physical violence, the current version of Sora 2 AI can generate videos of exploding buildings and crimes such as burglary, which can incite misinformed fear of war or criminal activity.
Meanwhile, although Sora 2 AI doesn’t recreate certain political figures and current events, when the AI video generator was prompted to generate a video of people wearing blue clothing in a political gathering, former U.S. president Barack Obama’s voice was used in the video. Therefore, despite regulations to prevent Sora 2 AI from being misused, these examples show how the Sora 2 AI Video Generation still has room for improvement in its safeguards.
Regarding the global impact, the flaws in Sora 2 AI’s design also suggest YouTube’s own mistake in allowing AI-generated content to be distributed on its platform. Although AI can arguably be used to increase the efficiency in a creator’s work, the consequences of YouTube’s commitment towards integrating Generative-AI into its service suggest that they could have prevented more trouble if they had spent more time and effort waiting than eagerly grabbing the latest technological innovation.
Sources:
Sora App and Sora 2 – Supported Countries | OpenAI Help Center
How Creators Use AI for Content Creation – How YouTube Works.
OpenAI’s Sora Makes Disinformation Extremely Easy and Extremely Real – The New York Times
