Public Reason for AI-Alignment: A nuanced account

Parallel session

Parallel session 1

Presentation slot

Track

Track 6: Ethics at Scale: Systems, Infrastructures, and Societal Impact

Presenter

Michael is a research fellow at the Institute for Technology Assessment and Systems Analysis (ITAS), Karlsruhe Institute of Technology (KIT), where he is a member of the Philosophy of Engineering, Technology Assessment, and Science research group. He is also a research associate at the African Institute for Epistemology and Philosophy of Science (ACEPS), University of Johannesburg. Within the philosophy of technology, his work focuses on the political philosophy of mobility and of AI. He also works on the method of reflective equilibrium and on the intersection of epistemology and political philosophy. He currently co-chairs the Society for Philosophy and Technology's Special Interest Group on the Political Philosophy of Engineering and Technology. He earned his PhD at KIT in 2020.

Type of presentation

Single paper

Chair

Ethan Landes

Room

Momentum 2

Abstract

Aligning AI with moral values is widely recognized as important task. Besides the problem of implementation, it remains an open question which moral values current or future AI systems should align with. The prevailing pluralism in liberal societies complicates this issue further. As a solution, it has been proposed to use the concept of public reason of Rawls’s political philosophy to justify the selection of values based on an overlapping consensus all reasonable worldviews within a liberal democracy share (Binns 2018; Gabriel 2022; Risse 2023). We agree with the basic assumption that the idea of public reason is fitting for justifying the specifications for AI alignment, however, there are further complications that necessitate a more nuanced account. First, liberal democracies increasingly face political polarization and the remaining consensus between different worldviews might become too insubstantial. Second, AI research and development as well as diffusion is happening in the global realm. Despite local regulation, like the EU’s AI Act having a global reach, AI alignment therefore should ideally be facilitated at a global scale. On the global scale, again, it is questionable how substantial any overlapping consensus would be and if there is any suitable public sphere for deliberation. Assuming that a global agreement on AI alignment is desirable and taking into account the polarization of liberal democracies, a convergence account of public reason (e.g. Gaus 1996) might appear superior to the consensus account (Rawls 2005, 2001). We explore whether this appearance is justified or deceptive. Furthermore, we investigate whether the recourse to public reasons is restricted only to human rights or constitutional essentials (narrow scope) or whether nearly all rules concerning AI alignment may only draw upon public reasons (wide scope). Lastly, we discuss which institutions might be necessary for adequate global public reasoning on AI alignment.

Co-author: Viktor M. Schubert.