History of AI on YouTube
Since 2016, YouTube has used the power of AI to assist in its primary service: creating an algorithm that decides which videos to promote to viewers (1). This is done by assessing a viewer’s satisfaction with the video played based on factors such as watch time, engagement, and their personal viewership profile. The algorithm also helps affect YouTube search results, based on the metadata (title, description, keywords, etc.) and reaction (views, likes, comments, watch times, etc.) of videos on the platform. YouTube’s basic business model is dependent on retaining viewers for as long as possible, as this makes them more desirable to advertisers (2). This means that AI is intricately connected to YouTube, helping produce an addictively personalized experience for viewers that uses their activity to recommend videos most likely to get them to stay watching longer.
Development of AI Tools
With rapid developments coming in the field of AI, YouTube plans to implement AI even further into the fabric of the service. Recently, they announced Dream Track as a connected service with YouTube shorts, its vertical video answer to TikTok (3). It allows creators to enter a prompt that creates unique soundtracks using the voices of recording artists such as Charlie Puth and T-Pain. Along with finding interested artists, they had to negotiate with major labels such as Universal Music Group, Sony Music Entertainment, and Warner Music Group to finalize voice rights in a deal the labels view as precedent-setting for future AI deals (4). One of the key issues was the resources used by the AI model to learn, while another was the options for artists to either opt in or out. Another fundamental question is how payment should be approached; should artists receive compensation for their music being utilized to train the AI model or for the music generated by the tool?
Despite the interesting creative potential, this innovation comes with the concern that YouTube will turn to its Content ID tool to flag content with voices that sound like artists signed to these labels despite the risk of false positives (6). With an incentive to keep the labels happy and a content ID system that is highly flawed, enforcement mechanisms will likely frustrate innocent creators such as impressionists or cover artists (7).
In the upcoming months, they will also continue development on another tool called Dream Screen, an innovative feature enabling users to generate AI-produced images or video backgrounds for YouTube Shorts by typing a prompt (8). Other assistance tools promise to provide AI-powered insights, helping creators brainstorm or draft outlines for their videos based on their individualized style and which videos have been most successful (9). While these do not implicate IP rights in the same way Dream Track does, it reveals YouTube’s view that increased AI use is in the interest of their creators and by extension, will lead to content more attractive and addictive to viewers.
Flagging AI use
A few weeks ago, YouTube posted a blog titled “Our approach to responsible AI innovation”. The main takeaway was adding a requirement for creators to flag their videos if they include any AI-generated content (10). The justification given is that “AI’s powerful new forms of storytelling can also be used to generate content that has the potential to mislead viewers—particularly if they’re unaware that the video has been altered or is synthetically created”. Those affected by these videos will now have the option to request their removal on the grounds that it is AI-generated (11). These can also include other creators who, for example, may have their voices mimicked by AI software without permission.
Strangely, discussions surrounding this policy suggest that using the voices for parody or satire does not gain additional protection and is still removable by request (11). In addition, it is unclear how YouTube seeks to enforce these policies on those who consistently fail to flag their own content for AI use. YouTube does not currently have a tool capable of detecting external AI use, meaning this protective system is currently operating on a form of “honour system”, and part of the danger the policy is hoping to protect against is how difficult it can be to tell when content is generated this way.
In the post, YouTube listed a few situations that required additional protection where the content discusses sensitive topics, such as elections, ongoing conflicts, and public health crises, or public officials (10). Specifically, unregulated AI use can spread misinformation through deepfake images or perpetuate biases on a larger scale (12). Ironically, it is unclear if YouTube itself has been responsible in its use of AI models; creators are pointing out the hypocrisy that YouTube, under Google, has scraped the internet and used other people’s content to train all of their AI tools, one of the fundamental arguments plaguing AI today (11). This includes another tool Google has used that auto-generates video summaries to better meet the interests of viewers, along with other search-generation tools (13).
The effect on creativity?
Through the various arms of its AI enterprise, YouTube may soon offer tools that come up with video ideas, generate key visual and auditory elements, summarize the contents of the video, and advertise directly to key demographics. An unintended consequence may be the skill development of its creators, who are less involved in the creative process and use these tools as a crutch rather than an accessory (14). Taken to the extreme, this can also lead to less variation and individual vision on the platform in favour of highly curated, impersonal content produced with one goal in mind: succeed in the YouTube algorithm. At what point would the “skill and judgment” be so completely removed from a creator’s video that they would no longer be deserving of copyright protection?
Clearly, AI holds the potential to significantly increase the possibilities of creation on YouTube; however, if approached recklessly or without proper protections in place, it may actually erode the creativity itself.
________________________________________________________________________________________
Resources:
(3) https://blog.youtube/inside-youtube/ai-and-music-experiment/
(5) https://www.theverge.com/2023/8/22/23841822/google-youtube-ai-copyright-umg-scraping-universal
(6) https://www.billboard.com/pro/youtube-developing-ai-tool-creators-sound-like-musicians/
(7) https://www.hypebot.com/hypebot/2022/02/youtubes-content-id-system-doesnt-work.html
(8) https://www.theverge.com/2023/9/21/23883749/youtube-generative-ai-dream-screen-creators
(9) https://blog.youtube/news-and-events/made-on-youtube-2023/
(10) https://blog.youtube/inside-youtube/our-approach-to-responsible-ai-innovation/
(11) https://kotaku.com/youtube-ai-disclosure-1851020956
(12) https://www.cnn.com/2023/06/16/tech/ai-apocalypse-warnings/index.html
(13) https://www.theverge.com/2023/8/1/23815321/youtube-ai-video-summaries
(14) https://www.polygon.com/23933664/youtube-ai-plagiarism-creator-tools
Further Reading:
https://hbr.org/2023/03/how-will-generative-ai-disrupt-video-platforms
*** NOTE: this is my post for the purposes of the 20% course requirement ***
Communications Law
Youtube has definitely come a long way from where it started! The idea of having AI generated songs is both interesting and sad to me. On the one hand, I’ve seen similar technology used to make videos on tiktok of AI “covers” of songs, eg. Harry Styles singing Style by Taylor Swift. If you’re a fan of both or either, it might be something you’d want to hear. But if you have to AI generate songs… like there’s already so much music out there, including music in the public domain if that’s why you’re leaning towards using AI. I would prefer a system that can find you an existing piece that was created by a human being, like a search engine for music, than whatever this Dream Track concept is. I definitely agree with your last statement… AI has a huge and scary potential to completely zap us of creativity and I think that will be a detriment to art of all kinds.