{"id":6366,"date":"2025-07-29T15:53:13","date_gmt":"2025-07-29T05:53:13","guid":{"rendered":"https:\/\/blogs.deakin.edu.au\/article\/?p=6366"},"modified":"2025-11-05T11:38:30","modified_gmt":"2025-11-05T00:38:30","slug":"blak-focus-july-2025-edition-ai-bias-and-indigenous-knowledges","status":"publish","type":"post","link":"https:\/\/blogs.deakin.edu.au\/article\/blak-focus-july-2025-edition-ai-bias-and-indigenous-knowledges\/","title":{"rendered":"Blak Focus July 2025 Edition \u2013 AI bias and Indigenous Knowledges"},"content":{"rendered":"

As AI and Generative AI become embedded in our everyday digital activities, there are ethical concerns about inherent bias. Digital technology and, by extension, machine learning and AI systems are not <\/span>neutral (Schuman, 1985; Crawford, 2021; O\u2019Neil, 2016; Noble, 2018; Hare, 2022; Wang, Chen, Huang, Redwing & Tsai, 2024; Worrell & John, 2024; <\/span>Khurana, 2025<\/span>)<\/span> as they hold the values and biases of those who build and train these systems. Indigenous knowledges, which are relationally and place-based, have historically been marginalised and mispresented. Generative AI further compounds this misrepresentation and raises questions about the \u2018sacredness of knowledge\u2019 <\/span>(Khuarna, 2025, p. 1)<\/span> and the ethical use of AI and the incorporation of Indigenous knowledges into these algorithms and systems.\u00a0\u00a0<\/span>\u00a0<\/span><\/p>\n

AI models’ design, training and implementation rely on big data and data systems that elevate Western knowledge systems and reflect power dynamics (<\/span>Peng, L., & Zhao, B., 2024).<\/span>\u00a0 In <\/span>Fairness and Bias in Artificial Intelligence: A Brief Survey<\/span><\/i>, <\/span>Ferrara (2024, p. 4)<\/span> describes bias in AI design and usage as fitting into the following categories: sampling, algorithmic, representation, confirmation, measurement, interaction and generative bias.\u00a0<\/span>\u00a0<\/span><\/p>\n

\u00a0<\/span>Bias refers to the systematic errors that occur in decision-making processes, leading to unfair outcomes. In the context of AI, bias can arise from various sources, including data collection, algorithm design, and human interpretation. Machine learning models, which are a type of AI system, can learn and replicate patterns of bias present in the data used to train them, resulting in unfair or discriminatory outcomes <\/span>(Ferrara, 2024, p. 2).<\/span>\u00a0<\/span>\u00a0<\/span><\/p>\n

Data colonialism<\/h2>\n

In <\/span>Weapons of Math Destruction,<\/span><\/i> O\u2019Neil <\/span>critiques how big data can increase inequality and threaten democracy by highlighting the problems of mathematical models, complex tapestries, and probabilities that comprise algorithms (<\/span>2016, pp. 19-20).<\/span> Large language models and AI systems are generally trained on datasets of internet-scale, \u2013 such as LAION-5B, Common Crawl or YouTube-8M, which include audio, video, image, text and multimodal datasets <\/span>(Vijay, 2024).<\/span> The training process on these datasets often results in large-scale and algorithmic bias. O\u2019Neil explains that these models are frequently.<\/span>\u00a0<\/span><\/p>\n

Lack specific data or information, so they tend to \u2018substitute stand-in data or proxies\u2019 and form discriminatory correlations <\/span>(20216, p. 21).<\/span> This has significant implications for Indigenous knowledge representation, especially since large data sets often reflect dominant cultural narratives and underrepresent Indigenous voices <\/span>(Wang et al., 2024).<\/span>\u00a0<\/span>\u00a0<\/span><\/p><\/blockquote>\n

Algorithms of <\/span><\/i>Oppression <\/span><\/i>(2018)<\/span> Nobel discusses how tools are used in decision-making processes, including big data and algorithms that reproduce societal bias, structural racism, and do little to promote equality. In discussing <\/span>On Our Back, <\/span><\/i>a black feminist publication, Noble describes the challenges also faced by information workers:\u00a0<\/span>\u00a0<\/span><\/p>\n

\u2026from the digitization of indigenous knowledge from all corners of the earth that are not intended for mass public consumption, to individual representations that move beyond the control of the subject. We cannot ignore the long- term consequences of what it means to have everything subject to public scrutiny, out of context, out of control <\/span>(2018, p. 132).<\/span>\u00a0<\/span><\/p><\/blockquote>\n

Landini (2025)<\/span> argues that AI systems draw Indigenous data from sources that distort narratives and don\u2019t preserve underlying relational and ethical dimensions. In doing so, AI systems fail to understand the connection between Indigenous knowledges, culture, intangible cultural heritage and intellectual property rights, which leaves Indigenous knowledges open to misuse and misappropriation (<\/span>Landini, 2025, pp. 505-509).<\/span>\u00a0<\/span>\u00a0<\/span><\/p>\n

Mitigating bias in AI<\/h2>\n

Addressing bias in AI is a pressing concern for researchers, developers, and users. <\/span>Ferrara, <\/span>in addressing sources of bias, emphasises technical mitigation strategies, including:<\/span>\u00a0<\/span><\/p>\n