ABSTRACT
This chapter examines the question of how power can be concentrated at the intersection of philanthropy and AI. It focuses on the analogue ways in which foundations and foundation AI models create tensions with democratic values. Drawing on existing literature in political theory, it argues that large foundations can operate outside democratic institutions, with substantial societal impact and minimal accountability, leading to considerable concentration of power. These foundations tend to promote donors’ individual conceptions of social good. Lacking transparency and public accountability, they result in large concentrations of power that are unresponsive to the individuals whose lives are affected by their work. Developers of foundation AI models share similar characteristics. These foundation models encode the values and priorities of their creators, which largely shape their societal impact. Design decisions reflect how developers balance conflicting values. Given the complexity of foundation models and inadequate regulatory frameworks, there is a lack of transparency and accountability to the public whose lives are impacted by these models. Moreover, the high cost of training and deploying these models leads to there being only a few companies that can train and deploy them, leading to a disproportionate concentration of power that is in tension with democratic ideals. The final section of the chapter discusses the method of deliberative alignment, a recent initiative that AI firms have used to democratize decision-making about foundation models. This method, I argue, could counteract the anti-democratic tendencies of foundation models, by ensuring that a broad range of values and perspectives are considered in the model development and governance. This approach enhances transparency and accountability. The chapter concludes by discussing some of the key drawbacks and advantages of deliberative alignment as a method for enhancing democratic governance over foundation models.
