Al Technologies: Algorithmic Monocultures, Arbitrariness, And Global Divides

Artificial intelligence (AI) technologies are becoming increasingly complex, powerful, and opaque. The multi-million dollar cost of developing Al models has concentrated state-of-the-art Al production within entities in the Global North. The majority of Al systems are designed by a limited set of actors, leading to a uniformity that fails to reflect the pluralistic nature ofsocieties, potentially overlooking or misinterpreting the needs and values of Global South populations [10]. This centralization fosters an “algorithmic monoculture:” a common set ofAl models is deployed across multiple decision-makers. This monoculture paves the way for the emergence of “Algorithmic Leviathans” [3,13] – where Al exercises disproportionate control over access to resources and the exercise of rights and can exacerbate inequalities and introduce new forms of discrimination. This policy brief recommends an ethical governance framework for mitigating two harms of Algorithmic Leviathans: arbitrariness and bias. 1) Arbitrariness manifests when seemingly innocuous decisions during development of Al models lead to unexpected and detrimental individual and collective outcomes. Arbitrariness needs to be addressed by specific mechanisms of transparency as well as standardized procedures for evaluating and reporting potential harms, even at the early stages of Al development. 2) Bias and discrimination occur when an Al system’s performance varies across legally protected populations and groups. These discriminatory effects produce inequalities at scale in algorithmic Leviathans and monoculture contexts. Such effects are aggravated because the Global North does not consider the Global South’s legal and demographic contexts. To remediate arbitrariness and bias, we recommend systematic processes for human appeal and review. These safeguards are essential for Al to foster social development, mitigate inequality, and advance social justice.