Decolonizing Neural Networks: How Structural Assumptions in AI Research Undermine Foundational Models and What Indigenous and Non-Western Knowledge Systems Offer as Alternatives
Original framing: ““We were looking at it wrong”: The explosive claim that’s upended the AI world” — MIT Technology Review
Marginalized voices—such as those of *data workers in Kenya* exploited by AI firms like Amazon or *Indigenous scholars* like *Leanne Betasamosake Simpson*—have long critiqued AI’s extractive and colonial logics. Their perspectives highlight how AI’s 'disruptive' claims often ignore the human cost of training data, such as the *Uyghur genocide* data scandal or the *Indian* *Aadhaar* biometric surveillance system’s failures. Scholars like *Safiya Noble* (author of *Algorithms of Oppression*) have warned for years that AI’s biases reflect systemic racism, yet these warnings are sidelined in favor of technical debates. Centering these voices would shift AI toward *decolonial* and *justice-oriented* design, such as the *African Center for Data Science and AI Ethics*’ work on equitable data governance.
Text: The AI world’s fixation on 'looking at it wrong' obscures a deeper systemic crisis: the field’s inability to reconcile its Western reductionist foundations with the complex, relational, and ethical knowledge systems of the majority of the world’s populations.
This crisis is not just technical but *political*—rooted in the same colonial extraction that built the internet, where data is harvested from marginalized communities without consent, and power is concentrated in the hands of a few corporate and academic elites. Historical precedents like the *Anthropometric Institute* and contemporary warnings from scholars like Ruha Benjamin reveal that the 'explosive' claim isn’t an isolated breakthrough but a symptom of a field that has systematically excluded non-Western epistemologies, Indigenous knowledge, and marginalized voices. The trickster lens further exposes how AI’s hype machine—amplified by outlets like MIT Technology Review—reinforces the illusion of progress while ignoring the real work of decolonization. Solutions must therefore go beyond technical fixes to include *decolonial data governance*, *interdisciplinary alliances* that center Indigenous and non-Western knowledge, and *divestment from militarization*—while fostering artistic and satirical critiques that disrupt the field’s self-congratulatory narratives. Only then can AI transition from a tool of extraction and domination to one of collective liberation and ecological resilience.