AI4D - Digital and Biometric Identity Systems

This policy paper examines issues emerging around the deployment of Artificial Intelligence (AI) in Digital and Biometric Identities (BDI) being rolled out across Africa as a central part of digital strategies to meet the UN 2030 Sustainable Development Goals (SDGs).

SDGs Target 16.9 aims: “to provide legal identity for all, including birth registration by the year 2030″. Digital identity is also seen as key to unlocking various other development goals including universal health and education access, and financial inclusion. BDI systems present opportunities for enhancing the visibility of benefectors of social services, enhancing efficiencies in digital transacting, and a variety of other potential social and digital economy benefits.

The literature demonstrates that emergences of AI in the BDI context, however, exacerbate risks already present to both fields that arise from the centrality of personal data, mass collected and analysed, within their systems. And there are broader risks – some of which arise from the centrality of identity, some of which arise from the nature of AI, and some due to the combination of both. These include for instance, exclusion from systems due to bias or inefficiencies; lack of accountability given stakeholder relationships and the nature of the technologies themselves; heightened risks for surveillance or monitoring; improper delegation of functions; and different ramifications of technology dependencies. These challenges might be directly addressed by different forms of policy solutions, which specifically advance forms of transparency mechanism, human rights and other legal instruments, and/or design solutions.

This paper draws on two case studies – one in Ghana involving facial recognition software, and another in South Africa involving natural language processing – to add depth to these background findings on the complexities of BDI systems and AI in Africa. Within conversations on BDI, there are two key forms of digital identity: foundational digital identity is associated with foundational public sector functions, such as national and civil registration systems, whilst functional digital identity systems are those decentralised identity systems for specific sectors or use cases (Bhandari et al., 2020). Both case studies emerge as largely functional examples of digital identity projects, with only tangential relationships to foundational identity systems. This divergence, however, is an important finding for considering the potential environment in which the AI aspects of BDI will emerge…