A/80/278 concentrating large volumes of AI-produced works, this leads to the invisibility of works representing these cultures, perpetuating pre-existing inequalities. Underrepresentation can also lead to stereotypical representations, offering folklorized or stigmatizing views of these groups, reinforcing prejudices, disrespecting their cultural identities and harming their dignity. 57 All these negative elements have been gradually exposed in the literature on AI in the past few years, to such a degree that scholars wonder whether AI is becoming the new colonizer of Indigenous Peoples. 58 36. Decontextualization is evident in the fashion industry. The use by AI of Indigenous garments and motifs is reduced to mere aesthetics, which strips those garments and motifs of the cultural memory that grounds their meaning. Prints “inspired” by Maori or Nigerian heritage motifs, taken without any acknowledgment of where those motifs came from or of who designed them, and without consultation with the source community, violate cultural rights and potentially propagate harm. AI systems have disassociated designs from their social and historical lineages, converting them into commodified fragments in a broader process of algorithmic consumption. 59 AI tools could be used with the active participation and consent of the communities, which would ensure the coherent use of the designs in line with their real meaning and significance. 37. Initiatives to increase the availability of data related to marginalized groups in digital environments without the consent of the source community carry risks. They can result in a loss of control over narratives and cultural representations, 60 as well as cultural appropriation, whereby “ancestral knowledge, sacred art forms and traditional expressions become training data.” 61 38. In some emerging practices, the groups concerned are involved at every stage of projects that affect them. For example, the Creative Labour and Critical Futures research cluster works with minorities. 62 The Mila – Quebec Artificial Intelligence Institute, led by Michael Running Wolf, uses AI to document and revive Indigenous languages in cooperation with local communities to the extent that they agree their data may be used. 39. AI-generated content may also reinforce stereotypical representations of women. These biases stem from training data sets that underrepresent women ’s voices, experiences and contributions, or that overrepresent them in traditional or objectifying roles. “The underrepresentation of women in AI development and leadership roles can further lead to the creation of socio -technical systems which fail to consider the diverse needs and perspectives of all genders, once again perpetuating stereotypes and gender disparities.” 63 As a result, AI systems can contribute to the rendering invisible of the diversity of women’s identities and roles in societies. This impedes the exercise of women’s cultural rights. 40. Generative AI tools pose specific challenges for women and girls, notably by reinforcing gender-based discrimination and enabling new forms of harm to their dignity and integrity. For instance, “text-to-image models can easily generate images __________________ 57 58 59 60 61 62 63 25-12403 Submission of the Human Rights Ombudsman of Guatemala. Jason Edward Lewis, ed., Indigenous Protocol and Artificial Intelligence, position paper (Honolulu, 2020). Submission by Indira Boutier, p. 7. Submission by Centro de Investigación y Docencia Económicas and Artículo 19 Oficina para México y Centroamérica, p. 3. Submission by Brazil, p. 2. Submission by Creative Labour and Critical Futures, p. 5. UNESCO and International Research Centre on Artificial Intelligence, “Challenging systematic prejudices: an investigation into gender bias in large language models ” (Paris, 2024), p. 5. 13/21