Value-based Requirements Engineering for Human-Centred AI: Foundations, Challenges, and a Research Agenda

Abstract

Artificial intelligence (AI) systems are increasingly central to decisions that impact individuals and society, showing the need for these systems to align with stakeholder values. However, traditional engineering approaches often relegate values to implicit considerations or early design phases, which diminishes their influence on system behaviour and operational decision-making. This paper investigates how Value-based Requirements Engineering (VBRE) can facilitate the development of human-centred AI systems that uphold stakeholder values, especially within dynamic and collaborative human-AI contexts. We propose that VBRE serves as a foundational framework for integrating stakeholder values as first-class artefacts throughout the entire AI system lifecycle. Our analysis consolidates existing research and identifies three core challenges: (1) the multilevel operationalisation of values from abstract principles to measurable system behaviours, (2) closing the gap between design-time value specification and runtime system behaviour, and (3) enabling shared value-driven decision-making between humans and AI at runtime. This paper analyses these challenges and proposes a research agenda for advancing VBRE to support the development of value-aligned, transparent, and accountable human-centred AI systems.