Living with Algorithms: Human-AI Assemblages and Ethical Governance in West African Pastoral Conflict

Abstract This study examines how artificial intelligence (AI), assisted conflict governance interacts with local legitimacy in pastoral conflict systems across Plateau State, Nigeria, and Tillabéri Region, Niger. Using a sequential explanatory mixed methods design (n = 183 household surveys, 27 focus groups, 6 algorithmic audits, 14 months ethnography), we test a Human-AI Assemblage Framework positing that legitimacy mediates algorithmic performance. Structural equation modeling confirms 63% of governance quality's effect on effectiveness operates indirectly through ethical legitimacy (indirect effect = 0.394, p  < 0.001; direct effect = 0.229, p  < 0.001; CFI = 0.957, RMSEA = 0.041). Counterfactual bias audits reveal herder-affiliated villages face + 18 percentage points higher aid denial likelihood ( p  = 0.024). System precision declined from 0.78 to 0.63 (2021–2023) due to patterned civilian resistance. A three, safeguard intervention, participatory threshold, setting, quarterly bias audits, hybrid tribunals, reduced false positives by 50%, upheld 42% of appeals, and raised cooperation from 58% to 79% ( p  < 0.01). The formalization frontier at 55% codifiability marks where additional data capture erodes legitimacy. Algorithmic governance requires institutionalized ethical oversight, not merely technical optimization.