Thea Pitman

AIAI: Artificial Intelligence, Art and Indigeneity started life as a pilot research project designed to explore the reactions of a group of Indigenous artists, writers, thinkers, community leaders and traditional knowledge holders to the new wave of AI image-generation tools – the ways in which they may see themselves (mis)represented or invisibilised by such tools and the ways in which they may choose to critically and creatively engage with them.¹ Its genesis was in (UK) summer 2022, when academics based at the University of Leeds (Tom Jackson² and Thea Pitman) and Dublin City University (Andreas Rauh), all with prior experience of working on media arts projects with Indigenous people in Abya Yala/Latin America through collaboration with the Brazil-based Indigenous arts and communication NGO Thydêwá (led by Sebastián Gerlic together with Nhenety Kariri-Xocó, Atiã Pankararu, Maria Pankararu and Mayá Pataxó-Hãhãhãe),³ decided to apply for a small amount of funding to undertake the first iteration of the project. The same academic team would work once again in collaboration with Thydêwá and with Sandra De Berduccy, the Bolivian multimedia textile artist better known as aruma, who had also collaborated with other members of the team on previous projects.⁴

A group of seventeen Indigenous people from different ethnic groups located in what is today Argentina, Bolivia, Brazil, Chile and Peru were recruited through Thydêwá’s and aruma’s networks and invited to two online focus groups (held in either Portuguese or Spanish), which took place in May 2023. Participants included the artists Azul–Nicolás Jiménez Reyes (Mapuche), Lucian de Silenttio (Aymara), Tadeu Kaingang and Haylly Zamora Aray (Wichí-Kariñ’á) and the writers, thinkers and community leaders Nhenety Kariri-Xocó and Mariela Tulián, all of whom were integral to the project’s development over time, becoming named team members for its later phases, and have thus authored their own chapters about their experiences for the present book. Other participants in the first stage of the project who generated images that have been selected for analysis here include the artists Aymar Ccopacatty (Aymara), Loreto Millalén (Mapuche), Horacio Montes de Oca Ayala (Andean ancestry), Kadu Tapuya and Kuenan Mayu (Tikuna-Tariano), as well as the community leaders Wera Moru Tupi-Guarani and Ayra Ymboré.⁵

During the first focus group, after brief discussion of what generative AI is and its pros and cons, the tool Midjourney was demonstrated, and the process of designing and translating text prompts to be input in English was explained. Midjourney was selected because it was one of the major commercial tools that was hitting the headlines at the time, particularly for its high-quality, often surrealistic photographic outputs, as well as in relation to its handling of ethnoracial diversity. Midjourney v.5.0, which is the version that was available in late March 2023 when we started work on the AIAI project, has just switched from an ‘exaggerated illustration style’ to a more ‘generic stock photography look’, which was a big success with users, but it had also changed its algorithm to introduce ‘more ethnic diversity’ (Meyer 2025, 8). The latter change to the algorithm turned out to be less popular, resulting in a switch back to making ‘whiteness the norm again’ in version 5.1, released in early May 2023 (Meyer 2025, 8). While a prompt can easily specify the version of the model to use, inevitably most images were generated with the default version available at the time and thus did not end up exploring the more ethnically diverse version.

In between the first and second workshops, participants proposed their own prompts focusing on their visions or dreams of the future for their communities, and sets of images were generated collaboratively between participants and original team members. In the second focus group, a selection of these results was shared and discussed.⁶ Another impromptu experiment conducted with some of the participants who had missed the second session for Portuguese speakers involved creating images using only the name of a person’s Indigenous ethnicity as the prompt. This offered a more controlled experiment to explore the representation of ethnicity in Midjourney than the free-text visualisation of scenes of Indigenous futures.

Subsequently, in the latter half of 2023, Thydêwá went on to obtain funding for the AIIA: Apropriação Indígena da Inteligência Artificial [AIIA: Indigenous Appropriation of Artifical Intelligence] project to run a series of virtual encounters in order to ‘[apropiarse] de la inteligencia artificial con identidad y a favor del Buen Vivir’ [appropriate AI on behalf of (Indigenous) identities and in support of Buen Vivir] as the project presents itself on its YouTube channel in Spanish (Gerlic 2023–4a), resulting in the co-creation of an AI-illustrated e-book – Imagine: La Pequeña Francisca en Abya Yala [Imagine: Little Francisca in Abya Yala] (Red Abya Yala con Amor 2023) – based on the format of a series of stories all featuring ‘La Pequeña Francisca’ [Little Francisca] published by Mariela Tulián (Avalle Montes 2024). This project also used Midjourney, with some outputs being generated collaboratively and many more the result of individual experimentation with the tool. The vast majority of images generated using Midjourney were created on the AIAI project’s server on Discord, which has greatly facilitated the study of the processes of image generation and the nature of some of the outputs discussed in Part III of this book. Some artists, such as Lucian and Haylly, also went on to experiment with generating images with Midjourney and other tools in relation to their own artistic agendas and in ways that far exceeded any of the funded projects; others, such as Azul and Tadeu, have demonstrated a far more circumspect approach to the availability of image-generation tools.

A second phase of the AIAI project took place in March 2024. In an attempt to move on from the use of mainstream commercial models and their deficiencies, this focused on the design of a bespoke AI image-generation tool which we chose to call IndigenIA. The tool is based on a downloaded version of the open-source and customisable model Stable Diffusion, which is located on an individual laptop accessed remotely by project members thus offering greater data security and more modest energy consumption overall. (The team chose to call the laptop itself Dragon, on account of the MSI laptop’s branding.) The prototype allows individual project members to train their own LoRA (Low-Rank Adaptation) models to work in combination with the main Stable Diffusion model and thus potentially create images that respond more closely to the Indigenous realities and worldviews they want to represent. This stage of the project involved collaboration with the UK-based artist-technologist collective Immersive Networks (Dave Lynch, Sam Hallas and Christophe de Bézenac) and consisted of a five-day in-person workshop in Leeds to discuss ethical, environmental, epistemological and practical issues relating to the proposal to develop the prototype. The discussion of the prototype also benefitted from the development of a ‘Manifesto for Digital Buen Vivir’ in the course of a sister project called INDIGENIA that involved many of the same collaborators (Potiguara et al. 2024), and Indigenous researcher Alex Potiguara has also gone on to conduct further postdoctoral research on the subject of ‘Digital Buen Vivir’ through the Digital Good Network at the University of Sheffield.⁷ While the prototype was ready for use by May 2024 and was trialled by aruma and Mariela in particular, secure remote-access issues continued to dog the project throughout the latter half of 2024.

Opportunities to exhibit materials from the project were many. Thea organised a pop-up exhibition at Convention House in Leeds on 9 March 2024 to follow on from the design workshop. This mainly focused on the artists’ work beyond their participation in the AIAI project, although some of the images generated during the earlier stages of the AIAI project were included, together with commentary on individual artists’ objectives and analysis of the results. She also organised for the project to be presented at the Be Curious family engagement day at the University of Leeds in May 2024. Subsequently, aruma took the lead in facilitating the inclusion of the project in the ‘Ficciones Generativas’ [Generative Fictions] group exhibition at the Centro Cultural Gabriela Mistral in Santiago de Chile, 11 June to 13 July 2024, as well as representing the project in an episode of the Universidad de Chile TV series Utop-IAs focusing on ‘Sesgos y sus consecuencias’ [biases and their consequences] (Pino Anguita 2024), and Andreas led on the inclusion of materials from the first stage of the project in an exhibition held at the National Botanic Gardens of Ireland in Dublin, 10–17 November 2024.

Since January 2025, we have continued to work on training team members to use the IndigenIA prototype. This latter phase has involved Immersive Networks working closely together with Andreas, aruma, Haylly, Lucian, Mariela and Thea, with some organisational input from Sebastián. It would be fair to say that this has been extremely challenging because of technical glitches (for example, with remote access), the technical complexity of the task itself, the rather unintuitive user interface, the need to work remotely and the constant requirement for translation of interactions and localisation of interfaces across Spanish, Portuguese and English.

Meanwhile, Sebastián has gone on to direct a series of seven very short films and a longer documentary (Gerlic 2025b and 2025a, respectively) as part of his Inteligência Ancestral [Ancestral Intelligence] project, which explores questions of spirituality, nature, art, technology and ‘ancestral belonging’ and refers back to some of the images from the first AIAI project, as well as uses the generative AI tool Runway for animation. Mariela has used a wide range of AI image-generation tools to illustrate a series of children’s books that she has published and which are widely used in the Argentine Bilingual Intercultural Education system, and Nhenety has also turned to using AI image-generation tools (DALL-E 3) via ChatGPT to illustrate his blog posts. We will return to all these aspects of the project in the following sections of the book.