AI systems encode power: How algorithmic maps exclude marginalised knowledge and reshape societal reality

Indigenous critiques of AI centre on its reliance on commodified data, which violates principles of data sovereignty and collective ownership. Systems like Australia’s *Indigenous Data Sovereignty Network* argue that AI reproduces colonial extraction by treating traditional knowledge as raw material for corporate profit. The 'map-territory' metaphor fails when applied to indigenous epistemologies, where knowledge is inseparable from land, kin, and spiritual relationships. Without grounding in indigenous protocols, AI systems risk perpetuating epistemic violence under the guise of 'neutral' technology.

The AI 'map-territory' problem echoes centuries of cartographic violence, where maps were tools of empire to justify conquest and resource extraction. From Ptolemy’s *Geographia* to Mercator’s projections, Western cartography systematically erased indigenous geographies and knowledge systems. The 19th-century *craniometry* and *phrenology* movements similarly used 'scientific' maps of human difference to justify racial hierarchies, foreshadowing today’s biased AI models. This historical continuity reveals AI as the latest iteration of a long-standing epistemic project to control and simplify complex realities.

Non-Western traditions offer alternative frameworks for understanding AI’s limitations. In *Ubuntu* philosophy, knowledge is co-created through dialogue and context, not algorithmic abstraction. Andean *ayni* (reciprocity) critiques AI’s extractive data practices, while Buddhist *dependent origination* highlights the interconnectedness AI systems often ignore. These traditions expose AI’s 'map-territory' problem as a failure of relational thinking, where systems prioritise prediction over meaning. Cross-cultural dialogue could reorient AI toward collective flourishing rather than corporate control.

Scientifically, AI’s 'map-territory' issue stems from the *representation problem*: models compress infinite complexity into finite datasets, introducing bias and incompleteness. Research shows that training data often reflects historical power imbalances, with Global South contexts underrepresented or misrepresented. The *black box* nature of deep learning exacerbates this, as users cannot interrogate how outputs are generated. Without rigorous auditing and participatory design, AI systems risk becoming tools of *epistemic injustice*, where certain knowledge claims are privileged over others.

Artists and spiritual traditions have long critiqued reductive representations of reality, from Borges’ *On Exactitude in Science* to Indigenous sand paintings. These works highlight how 'maps' (whether cartographic or algorithmic) are always partial, serving specific agendas. AI’s reliance on statistical patterns mirrors Enlightenment-era reductionism, which spiritual traditions like Advaita Vedanta or Taoism reject in favour of holistic understanding. Creative resistance—such as algorithmic art that exposes bias—can challenge AI’s claim to objectivity, revealing its outputs as culturally situated narratives rather than neutral truths.

Future scenarios for AI must centre decolonial design and epistemic pluralism to avoid reinforcing extractive logics. One pathway is *participatory AI*, where marginalised communities co-design systems to reflect their values. Another is *algorithmic commons*, where open-source models are governed by communities rather than corporations. Scenario planning should also explore *post-extractive AI*, where systems are trained on data that communities control and benefit from. Without these shifts, AI will continue to produce 'maps' that serve power, not people.

Marginalised communities—particularly indigenous peoples, Global South populations, and gig workers—are disproportionately affected by AI’s 'map-territory' problem. Their knowledge is mined for training data but excluded from system design, as seen in cases like facial recognition bias against darker-skinned people. Grassroots movements like *Data for Black Lives* and *Algorithmic Justice League* demand accountability, but their voices are often sidelined in mainstream tech discourse. Centring these perspectives reveals AI as a tool of structural violence, where 'maps' justify exclusion and exploitation.