Machine Expertise in the Loop: Artificial Intelligence Decision-Making Inputs and Cyber Conflict

Abstract

How will national cybersecurity stakeholders react to increasingly sophisticated artificial intelligence (AI) products in the decision-making loop? Diverse AI applications, already a component of military operations, stand to further revolutionize threat-intelligence analytics, curation, and presentation in years to come. In doing so, they portend heightened efficiency and higher tempos of security operations while minimizing risk potential. And yet, challenges abound. AI inputs to deliberative processes may be perceived as more or less robust, accurate, or generalizable according to a range of factors that influence decision-makers. Then these habits and preferences might be reintroduced to the information loop in several ways, as the algorithm, the design/implementation process, and institutions adapt to accommodate the human element. Malicious actors are even likely to target this action-reaction loop to influence AI systems. This two-phase cycle is perhaps most worrisome in relation to cyber conflict, where informational ambiguity, functionally diverse workforces, and an expansive and fragmented threat environment combine to produce an immense opportunity for baking bias into the loop. This paper presents the results of two experiments designed to explore different manifestations of AI systems in the cyber conflict decision-making loop. Though findings suggest that technical expertise positively impacts respondents’ ability to gauge the potential utility and credibility of an input (indicating that training can, in fact, overcome decision-maker bias), the perception of human agency in the loop even in the presence of AI inputs mitigates this cautionary effect and makes decision-makers more willing to operate on less overall information.