A/80/278 24. It is a matter of urgency that clear and effective limitations be created for companies with respect to data sourcing. In the absence of effective protection, creatives have embedded in their works invisible modifications known as “filters” or “algorithmic camouflage” to prevent AI systems from learning from them. 41 But these measures cannot be the way forward. Users of AI systems must be able to understand how they are developed, including the sources of data, the methods of training, the actors involved and the intended uses. Transparency across the entire AI life cycle and standardized protocols to effectively implement an opt-in model on the Internet must be put in place by States. A collective compensation fund, financed by a tax on the revenues of large AI platforms, or harmonized standards in accordance with legislation, as developed in the European Union by the European Committee for Standardization and the European Committee for Electrotechnical Standardization, would also be solutions. 42 There is an urgent need to act to protect cultural rights from uncontrolled data grabbing. There is no time to continue with mere reflection on these issues, nor time to plan lengthy and complex reforms. 43 C. Undermining cultural diversity 25. In promoting only the creative forms that will sell, AI restricts cultural diversity and promotes homogenization. AI models remain fundamentally designed to reproduce statistically average patterns from their training data, reinforcing the most common styles. “AI systems trained on dominant cultural inputs tend to reproduce dominant aesthetics. The effect is subtle but significant: over time, global platforms begin to favour homogenized output, marginalizing regional, Indigenous, or non-Western modes of expression.” 44 This leads to the standardization of creativity. In addition, the average patterns easily promote pieces that defy the search for quality in art. The quality gradually and continuously falls as successive generations of AI are trained on the previously average content, intensifying the standardization of styles and ideas and further reducing originality and diversity. The same can be said of redirection by AI towards conformity with the mainstream, which reduces the prevalence of non-mainstream opinions. 45 In parallel, reducing the range of diversity to which people are exposed also weakens their ability to compare and appreciate a range in quality; as a consequence, their freedom to know and to choose is also significantly reduced. 26. The extremely rapid development of AI systems that generate text, images, videos and music accelerates this downward spiral of quality in creativity. AI tools adapted to specific forms of human creativity struggle to reproduce the nuances and variations that artists introduce into works. In Indian classical music, the microtonal variations known as shruti and raga cannot be reproduced by AI tools such as GaMaDHaNi and NaadSadhana, thus failing to capture the emotional essence and depth of what artists are trying to convey through the art form. The widely popular dance form from Tamil Nadu has its own set of codified gestures and rhythmic sequences that become a struggle for AI models such as Natya.AI. AI robs the dance __________________ 41 42 43 44 45 10/21 Submission by Benjamin Harbakk, p. 6. Submissions by Pau Aleikum Garcia of Domestic Data Streamers, p. 3; the Alan Turing Institute, p. 4; and the Writers’ Guild of Great Britain, p. 2. Submission by Bureau Malien du Droit d’Auteur, p. 1. Submission by Davar Ardalan and Amir Banifatemi, p. 2. Daragh Murray and others, “The chilling effects of surveillance and human rights: insights from qualitative research in Uganda and Zimbabwe”, Journal of Human Rights Practice, vol. 16, No. 1, 31 July 2023. 25-12403