Integrating AI Bots into Collaborative Requirements Decision-Making Through Temporal Graph Modeling

Abstract

Requirements engineering is a sociotechnical process in which stakeholders communicate, negotiate priorities and establish consensus under uncertainty. As AI agents increasingly participate in collaborative workflows, their integration may reshape group dynamics, induce automation bias and weaken human oversight. To address this challenge, we propose a spatiotemporal graph neural network (STGNN) framework for modeling continuous collaborative decision-making among human stakeholders and artificial agents. The proposed framework abstracts negotiation as a spatial decision environment in which users, options and AI bots interact over time, allowing the system to capture hesitation, social influence and consensus formation. We validate the approach using ambiguous multi-stakeholder proxy tasks from medical and biological classification, which emulate key properties of requirements negotiation such as uncertainty, competing alternatives and the need for collective agreement. Results show that the proposed temporal graph model improves the prediction of collective outcomes over static majority voting and a previous STGNN architecture, especially in highvariance proxy tasks when AI-agent trajectories are included, while not implying that bots directly improve human decisionmaking in requirements engineering. However, the observed influence of visually indistinguishable bots also highlights the need for transparent trust calibration and explicit human oversight. The framework therefore provides a foundation for tracing, analysing and auditing AI-augmented collaborative decisionmaking processes, supporting the integration of artificial agents as auxiliary, accountable participants rather than replacements for human professional judgment.