Complexity, understandability, and compatibility: A comparative study of AI advisory systems for National Security

Government Information Quarterly, vol. 42, pp. 102088

Abstract

Artificial Intelligence (AI) advisory systems are being implemented in the public sector for more efficient and effective decision-making. Yet, there is a lack of in-depth qualitative and comparative research focusing on how decision-makers in a real-world setting use different types of AI advisory systems. By asking “How do different AI advisory systems affect use by national security decision-makers?”, this study reveals through a qualitative case study and using scenario-based interviews that decision-makers are more likely to use relatively simple AI systems over complex ‘black box’ systems. Additionally, factors such as accountability concerns and compatibility with existing decision-making processes influence their willingness to use AI advisory systems. Ultimately, a more technically advanced AI system is not necessarily perceived as more competent, as decision-makers view processes like data analysis as integral to nuanced and effective decision-making. This suggests that the fit between the perceived competences and compatibility of the AI system and the decision-making task at hand is highly important for the successful implementation of AI advisory systems.

Authors 3

1. Brecht Weerheijm corresponding Aachen Institute of Public Administration

   Leiden University

   Affiliation as printed

   Institute of Public Administration, Leiden University, Turfmarkt 99, 2511 DP Den Haag, the Netherlands

2. Sarah N. Giest Aachen Institute of Public Administration

   Leiden University

   Affiliation as printed

   Institute of Public Administration, Leiden University, Turfmarkt 99, 2511 DP Den Haag, the Netherlands

3. Bram Klievink Aachen Institute of Public Administration

   Leiden University

   Affiliation as printed

   Institute of Public Administration, Leiden University, Turfmarkt 99, 2511 DP Den Haag, the Netherlands