To democratize or not to democratize AI? That is the question

Abstract

AbstractThis paper advances the debate surrounding whether to democratize AI and explores some of the challenges and benefits of democratization through community-based work and direct democracy. We contend that community-based strategies can incorporate local knowledge and control, thereby providing more effective AI solutions that are human-centric and less harmful. However, democratization needs to be approached with caution and care, since this process requires a deeper understanding of who participates, the decision domain, and the different realities at stake. Moreover, we highlight the importance of participation in AI development to ensure its legitimacy, considering the capacity of this technology to shape reality. We emphasize that participation should be more than just involving stakeholders or seeking input from users. Rather, participation should involve local narratives that generate knowledge and shape information landscapes, thereby producing a different, anti-Cartesian scene. We conclude by underscoring that the success of democratizing AI hinges on the careful delineation of the boundaries of participation, which should include the specific needs of the immediate context, the decision domain, and the various participants involved.