AI for the public. How public interest theory shifts the discourse on AI

Abstract

AI for social good is a thriving research topic and a frequently declared goal of AI strategies and regulation. This article investigates the requirements necessary in order for AI to actually serve a public interest, and hence be socially good. The authors propose shifting the focus of the discourse towards democratic governance processes when developing and deploying AI systems. The article draws from the rich history of public interest theory in political philosophy and law, and develops a framework for ‘public interest AI’. The framework consists of (1) public justification for the AI system, (2) an emphasis on equality, (3) deliberation/ co-design process, (4) technical safeguards, and (5) openness to validation. This framework is then applied to two case studies, namely SyRI, the Dutch welfare fraud detection project, and UNICEF’s Project Connect, that maps schools worldwide. Through the analysis of these cases, the authors conclude that public interest is a helpful and practical guide for the development and governance of AI for the people.

Reconciling Governmental Use of Online Targeting With Democracy

“…First, we suggest requirements for the design and execution of online governmental targeting campaigns that are in alignment with democratic values. We will reference the recent AI governance framework developed by Züger and Asghari [82] as our initial guideline. Second, we propose the creation of an independent body to monitor the development and implementation of these campaigns.…”

“…According to Züger and Asghari [82], there are five prerequisites for a system to align with the public interest: (1) public justification, (2) focus on equality, (3) inclusion of a deliberation/co-design process, (4) implementation of technical safeguards, and (5) commitment to openness for validation.…”

“…Züger and Asghari [82] propose that for an AI-based solution to be recognized as serving the public interest, it must possess a normative democratic justification that is widely accepted by the public. This justification should include a lucid explanation of the issue that the AI solution seeks to address and why it is superior to alternative solutions.…”

“…Züger and Asghari [82] assert that serving equality (and at the very least not hurting it) must be the most important normative goal of a solution that aims to promote public interest. Thus, they argue that any public interest AI-based solution must find a way to solve the problem of algorithmic bias and not create unwanted power imbalances in society.…”

“…In Section 5, we argue that online targeting by governments can still be reconciled with democratic principles and values. Drawing on Züger and Asghari's AI governance framework [82], we advocate for the establishment of an independent institution to oversee campaign design and deployment, serving as a check on government power. We show that this solution can, in principle, effectively address the three problems raised in Section 4.…”