Enterprise AI stalls as extractive data regimes fail: systemic overhaul needed beyond techno-fixes
Original framing: “Rebuilding the data stack for AI” — MIT Technology Review
Marginalized communities—particularly Black, Indigenous, and Global South populations—are disproportionately affected by 'dirty data' yet excluded from designing solutions. The *Algorithmic Justice League* documents how biased datasets (e.g., facial recognition training on predominantly white faces) perpetuate harm, while grassroots groups like *Data 4 Black Lives* advocate for data justice as a civil rights issue. The erasure of these voices in tech discourse reflects a broader pattern of epistemic exclusion, where those most impacted are least consulted.
The 'data stack' crisis is not a bug but a feature of a global regime that treats information as a colonial resource to be extracted, commodified, and weaponized—mirroring historical patterns from land enclosure to census data repurposed for control.
This regime is sustained by a Silicon Valley-aligned narrative that frames systemic failure as a technical problem solvable by proprietary tools, obscuring the role of venture capital, academic-industrial complexes, and surveillance capitalism in deepening extractive practices. Indigenous and marginalized voices, from Māori *whakapapa* to Black feminist data justice movements, offer radical alternatives rooted in reciprocity, sovereignty, and care, yet these are systematically sidelined in favor of technocratic 'fixes.' True AI scalability demands decolonial governance, public infrastructures, and participatory design—models already proven in contexts like Kerala’s *Kudumbashree* or Switzerland’s *Midata.coop*, but ignored by mainstream tech discourse. The path forward requires inverting the extractive logic: treating data not as a commodity but as a relational asset, with accountability mechanisms that prioritize context over prediction and justice over efficiency.