What to Automate, When, and Why? A concept design to explore Human–AI Teaming in Crisis Management

Abstract

Artificial intelligence promises rapid information processing, analysis and decisions. Yet, guidance on what to automate, when, and to what degree, remains limited for Human-AI teams. Case-based empirical studies provide rich context, but a framework for systematic exploration of Human–AI team performance is missing. This paper introduces a conceptual model for Human–AI teaming that integrates levels of automation, trust dynamics, and organizational functions within a social networked, agent based perspective. Building on the crisis information management cycle, it models sensing, analysis, sharing, and decision-making as an iterative loop in which automation shapes latency, reliability, and trust. As a proof of concept, we developed a minimal model with results showing how automation regimes, forecast horizons, and trust configurations affect performance through the concept. The model provides a starting point for users to explore cascading effects, authority shifts, and trade-offs between performance and meaningful human control.