Thea Pitman

In a post on his Cybernetic Forests blog, Eryk Salvaggio (2025) asks the question, ‘Are Other AIs Possible?’ This is a question that others have been addressing for a while and in broader contexts than just the relationship between AI and art. Alternative theoretical approaches to AI design from the field of critical AI studies (Raley and Rhee 2023) include decolonial AI (Mohamed, Png and Isaac 2020; Ricaurte 2022; Nas 2024) and the related concept of a federated AI Commons (Varon et al. 2024), among others.¹ Such approaches seek to address the extractivist and totalising modus operandi of hegemonic AI (Ricaurte 2022), with its unethical approach to where it gets data from, its exploitation of labour in the labelling and ‘cleaning’ of data, and its massive environmental impacts, as well as its inherent tendency to replicate or even enhance negative stereotypes and social inequalities based on the extant biases in the datasets used, and to render whole communities invisible where it lacks sufficient data. This approach clearly ends up favouring the predominantly white, Western, heteronormative, male, ableist worldviews of the majority of the creators of AI tools and the data they use.

As Ameera Kawash argues,

Decolonizing AI is a multilayered endeavor, requiring a reaction against the philosophy of ‘universal computing’ – an approach that is broad, universalistic, and often overrides the local. We must counteract this with varied and localized approaches, focusing on labor, ecological impact, bodies and embodiment, feminist frameworks of consent, and the inherent violence of the digital divide. (Untold Mag 2025, response to first interview question)

Similarly, Shakir Mohamed, Marie-Therese Png and William Isaac (2020) argue that decolonial AI should endeavour to avoid ‘techno-solutionism’ and propose a critical, self-reflexive approach where ‘AI systems can be adapted to locally specific situations in original ways’ (674), and Joana Varon, Sasha Costanza Chock, Mariana Tamari, Berhan Taye and Vanessa Koetz’s (2024) concept of an ‘AI Commons’ draws on the thoughts of a range of groups working towards alternatives to hegemonic AI that ‘are focused on critiquing, safeguarding, improving, imagining, and/or developing alternatives to the current “default settings” of AI as a tool to advance the matrix of domination (capitalism, white supremacy, patriarchy, and settler colonialism)’ (6). Most recently, Mirca Madianou, in her Technocolonialism: When Technology for Good is Harmful (2025), has built on the work of Sasha Costanza Chock, as well as Arturo Escobar’s earlier and highly influential Designs for the Pluriverse: Radical Interdependence, Autonomy, and the Making of Worlds (2018), arguing that, in a world where AI and big data are often seen as technological solutions to ‘problems’ of development, we need to challenge ideas of universal computing and design ‘for good’ with ‘design for the pluriverse’ and ‘design justice’ (194). Madianou also points out the linguistic elephant in the room: ‘it is impossible to decolonize AI as long as the algorithms and large language models are trained on huge datasets in the English language’ (194), and this has implications not just for the generation of text but also for the generation of images based on textual prompts.

It is no surprise that in terms of the design principles of what decolonial AI might look like, several authors draw explicitly on Indigenous concepts of community responsibility and collective work for the common good, such as ‘tequio’ / ‘minga’ / ‘mutirão’ (Ricaurte Quijano 2021; Ricaurte 2022, 737), and of living in harmony with others and with our environment, such as Buen Vivir (Varon et al. 2024),² and ‘Indigenous AI’ has gained currency as a term in its own right. The intersection of AI and Indigenous Peoples or even Indigenous knowledges is not, of course, identical to ‘Indigenous AI’. It is the case that international organisations, such as the UN and the World Wildlife Fund, and non-Indigenous-led Global North-based research teams working in this field tend to take a very utilitarian, often Sustainable Development Goals-focused approach, as per Madianou’s above-mentioned criticisms, seeking ways to use AI ‘for good’, to ‘help’ Indigenous Peoples with healthcare solutions, environmental monitoring and food security, as well as language revitalisation and cultural preservation, for example.³

Instead, it is Indigenous academics such as Jason Edward Lewis (Native Hawaiian and Samoan), working in an international team with other Indigenous academics, including artists such Suzanne Kite (Oglala Lakota) and computer scientists such as Michael Running Wolf (Northern Cheyenne and Lakota), who have coined the terms ‘Indigenous-centred AI design’ and ‘Indigenous AI’. In their Indigenous Protocol on Artificial Intelligence, Lewis et al. (2020) argue for Indigenous involvement at all stages of the AI ‘ecosystem’ and the careful consideration of Indigenous data sovereignty in relation to AI, as well as advocating for the development of more radical alternatives to hegemonic AI that are aligned to Indigenous epistemologies of relationality, reciprocity and pluriversity.⁴ Lewis and colleagues are not averse to practical solutions-focused AI tools development, but they need to be localised, ethical and steered by Indigenous community needs and principles, ‘rethinking and prototyping AI through Indigenous epistemologies’ (Lewis, Whaanga and Volgörmez 2024, 8).⁵ In summary, their ‘Guidelines for Indigenous-centred AI Design’ comprise,

designing AI systems in partnership with local communities; building relationality and reciprocity into the foundations; the requirements that AI systems developed by, with, or for Indigenous communities should be responsible to those communities, provide relevant support, and be accountable to those communities first and foremost, using Indigenous protocols as the basis for governance and regulation guidelines; recognizing that all technical systems are cultural and social systems and that computation is a cultural material; applying ethical design to the full stack of a technology and respecting and supporting Indigenous data sovereignty. (Lewis 2023b, 214)

Beyond this Indigenous-centred but still quite pragmatic approach, Lewis and colleagues have also gone on to theorise AI in relation to Indigenous epistemologies focusing on the concept of abundance and on what they call ‘the future imaginary’. The concept of abundance refers both to the need to design AI in the light of the plurality of Indigenous knowledge systems and to a care to ensure that the impact of AI on the environment preserves abundance rather than results in degradation and loss (Lewis 2023a, 17; Lewis, Whaanga and Volgörmez 2024). Abundant AI thus aligns very closely with discussions of decolonial AI and counters hegemonic, monolithic AI’s tendency to universalise and to promote a technosolutionism that creates more problems than it solves.⁶ In order to achieve this decolonisation of AI that Abundant AI proposes, Lewis, Whaanga and Volgörmez (2024) argue that Indigenous people need to use their imaginations, informed as they are by different epistemologies, to invoke the future scenarios for AI development that they want: ‘In order to transform AI, we must envision Indigenous futures grounded in community priorities and dreams [by] developing “future imaginaries” that point us toward alternative research and development paths for AI that promote abundance rather than scarcity, exploitation and control’ (9).

This speaks directly to the imagination of Indigenous futures that was a key dynamic in the first stage of the AIAI project. Nonetheless, we appreciate the fact that Abundant AI and ‘the future imaginary’ are both formulations that, while deriving from Indigenous worldviews, do not limit their ambitions to just being applicable to Indigenous Peoples.