Collective action on artificial intelligence: A primer and review

Progress on artificial intelligence (AI) requires collective action: the actions of two or more individuals or agents that in some way combine to achieve a result. Collective action is needed to increase the capabilities of AI systems and to make their impacts safer and more beneficial for the world. In recent years, a sizable but disparate literature has taken interest in AI collective action, though this literature is generally poorly grounded in the broader social science study of collective action. This paper presents a primer on fundamental concepts of collective action as they pertain to AI and a review of the AI collective action literature. The paper emphasizes (a) different types of collective action situations, such as when acting in the collective interest is or is not in individuals’ self-interest, (b) AI race scenarios, including near-term corporate and military competition and long-term races to develop advanced AI, and (c) solutions to collective action problems, including government regulations, private markets, and community self-organizing. The paper serves to bring an interdisciplinary readership up to speed on the important topic of AI collective action. •Collective action situations occur when it takes the efforts of two or more agents to achieve a goal.•Collective action is needed to improve the outcomes of AI development.•The AI collective action literature focuses on avoiding catastrophic outcomes from advanced AI.•The study of AI collective action should draw more on the broader literature on collective action.