Explicability of humanitarian AI: a matter of principles

Abstract In the debate on how to improve efficiencies in humanitarian sector and better meet people’s needs, argument for use of artificial intelligence (AI) automated decision-making (ADMs) systems has gained significant traction ignited controversy its ethical human rights-related implications. Setting aside implications introducing unmanned warfare, we focus instead impact adoption AI-based ADMs response. order maintain status protection conferred by mandate, aid organizations are called abide a broad set rules condensed principles notably humanity, neutrality, impartiality, independence. But do these operate when is automated? This article opens with an overview AI sector, special attention concept algorithmic opacity. It then explores transformative potential complex power dynamics between humanitarians, principled assistance, affected communities during acute crises. Our research confirms that existing flaws accountability epistemic processes can be also found mathematical statistical formulas algorithms used automation, intelligence, predictive analytics, other efficiency-gaining-related processes. doing so, our analysis highlights harm people resulting from opacity, either through removal or obfuscation causal connection triggering events services so-called black box effect (algorithms often described as boxes, their complexity technical opacity hide obfuscate inner workings (Diakopoulos, Tow Center Digital Journ, 2017). Recognizing need ethics dimension AI, action, endorse “explicability” developed within framework machine learning human-computer interaction, together proxy metrics. Finally, stress developing auditable standards, well transparent guidelines frameworks rein risks what been defined experimentation (Sandvik, Jacobsen, McDonald, Int. Rev. Red Cross 99(904), 319–344, concludes mechanisms respond needs populations situation vulnerability should essential feature default, preserve respect no principle even digital aid. conclusion, while confirm concerns related advocate roadmap towards introduce tentative basis future research.