Poster #1 - Task Characteristics and Delegation Expectations as Predictors of AI Appropriateness: Evidence from US Emergency Managers

• In Event: Saturday Poster Luncheon

Abstract

Rapid advancements in Artificial Intelligence (AI) have accelerated governments' efforts to adopt AI to enhance organizational decision-making and service delivery. Emergency management (EM) presents a distinctive context for AI adoption: tasks vary substantially in their suitability for AI integration, ranging from data-intensive, routine administrative work to disaster response and recovery requiring experiential, interpersonal, and ethical judgment (Misra et al., under review). Despite this variation, managers' perspectives on AI's appropriate use remain understudied, as does the relationship between how they conceptualize AI's decision-making role and which tasks they consider suitable for AI.   

This study asks: How do emergency managers perceive human-AI collaboration in their work across tasks varying in cognitive complexity and decision-making demands? Leveraging a mixed-methods approach, we surveyed 171 US-based managers. We developed a list of typical EM tasks and asked participants to rate the perceived appropriateness of AI integration for each task. Participants also rated the appropriateness of AI use at four levels of delegation (Dietzmann & Duan, 2022), following Simon's (1977) decision-making principles: (1) intelligence, (2) design, (3) choice, and (4) implementation. We also collected data on AI literacy, attitudes towards AI, demographics, and AI use.   

Our analysis followed a four-part process. First, we calculated the mean perceived AI appropriateness score for each task. Second, we conducted an exploratory factor analysis to determine if tasks cluster in theoretically interesting ways based on managers' perceptions of task-level AI appropriateness. Third, we conducted a thematic analysis of open-ended responses to understand the reasoning behind their ratings (Misra et al., under review). Fourth, we conducted hierarchical regression analysis to examine whether and how managers' perceptions of AI's role in task delegation predicted perceived task appropriateness, above and beyond the effects of demographics, AI use, AI literacy, and attitudes towards AI.  

Our factor analysis revealed three main task clusters: Factor 1 — tasks related to data, perceived as most appropriate for AI integration; Factor 2 — tasks related to planning and administration; and Factor 3 — tasks related to disaster recovery and response. Both Factor 2 and 3 tasks vary in perceived AI appropriateness (Misra et al., under review). Our hierarchical regression analysis identified AI task delegation levels as differential predictors of task cluster appropriateness. Managers' conceptualization of AI's role predicted their judgments of AI appropriateness. Specifically, those who perceived AI as strong in problem identification rated it as more appropriate for data tasks; those who saw AI as capable of proposing options rated it as more appropriate for planning and administration; and those viewing AI as an implementation tool rated it as more appropriate for disaster response and recovery. Our qualitative findings indicated clear boundaries around tasks requiring noncodified knowledge — experiential, interpersonal, and ethical capacities — which managers identified as beyond AI's capabilities (Misra et al., under review).  

Our results demonstrate that managers hold nuanced, role-specific views of AI appropriateness rather than uniform attitudes toward AI adoption. These findings can inform organizational strategies for responsible AI integration by highlighting the importance of aligning managers' conceptual understanding of AI's capabilities with task requirements in emergency management.

Authors

• Norhan Abdelgawad, Virginia Polytechnic Institute & State University
  Presenting Author

• Shalini Misra, Virginia Polytechnic Institute & State University
  Non-Presenting Co-Author

• Benjamin Katz, Virginia Polytechnic Institute & State University
  Non-Presenting Co-Author

• Tuan Pham, Virginia Polytechnic Institute & State University
  Non-Presenting Co-Author

• Patrick Roberts, Virginia Polytechnic Institute & State University
  Non-Presenting Co-Author

• Muhammad Awfa Islam, Virginia Polytechnic Institute & State University
  Non-Presenting Co-Author

• Daniel M Friedberg, Virginia Polytechnic Institute & State University
  Non-Presenting Co-Author