Democratization is a Process, not a Destination: Operationalizing Ethics and Democratization in a Cyberinfrastructure for AI Project

In the space of ethical AI, ambiguities abound. Lofty goals such as ethical alignment and systematized fairness are difficult to parameterize, and therefore difficult to achieve. Democratization of AI adds more alignment and operational challenges, but those challenges have not been well articulated or documented as an operational matter. In this paper we present a case study of a large-scale AI research project and discuss processes taken to operationalize ethical and democratizing AI, and the frictions that are inherent in this context. In particular, we discuss challenges associated with a distributed workforce and a mindset focused on research and development of cyberinfrastructure (CI) for AI, both common and seemingly innocuous, but frictious to AI democratization. We find that governance is an important link that holds ethical AI and AI democratization together and moreover that the procedural aspect of governance means that democratization, like democracy, is a process. In telling the measures undertaken in the ICICLE project towards democratization of AI, we contribute a better understanding of the practical challenges of democratizing AI that are particular to infrastructure projects and suggest possible ways for advancing democratization efforts.