RETHINKING AI FOR GOOD: CRITIQUE, REFRAMING AND ALTERNATIVES

Resumen

AI for Social Good (AI4SG) initiatives have emerged in various sectors. However, AI's non-neutral nature challenges claims that the “good” can simply be inferred by association with broad goals such as the Sustainable Development Goals (SDGs). The lack of a clear definition of "the good” or what it entails in practice risks making AI4SG an empty signifier. This ambiguity allows unchecked interventions, undermining societal efforts to align future AI developments with public good. In this article, we adopt a socially situated public good framework from the social studies of quantum technologies proposed by Roberson et al (2021) and use insights from critical AI scholarship to tailor this framework to AI4SG initiatives. Analysing AI4SG initiatives, and building upon existing critical literature, we scrutinize these initiatives with regards to the framings of the research problems, the wider social and institutional context in which AI initiatives are imagined to be applied and used, as well as the wider network of scientists, stakeholders and publics involved in their co-production. We argue that much of the AI4SG literature abstracts AI from social and contextual realities, making it difficult to clarify the ways in which they might in fact have an impact in the world. Outlining our first iteration, we argue that co-creating this framework demands iterative refinement and ongoing dialogue with diverse stakeholders, especially in the Global South.