Human defenders adapt differently than ai in cloud cyber defense games
Bounded Reasoning: Cognitive Hierarchy in Human-versus-AI Cyber Defense
Cryptography and SecurityHuman-Computer InteractionMachine Learning
Summary
This paper examines how humans and AI systems defend cloud computer networks from attacks in a game setting. The researchers found that humans change their defense choices based on past success or failure, while AI defenders do not show this pattern. Adding information about the attacker’s likely moves improved human performance late in the game but did not increase overall protection scores significantly. The results highlight that just looking at average performance scores can miss important differences in how humans adapt and think during cyber defense.
What this means in practice
- •For cybersecurity teams: Improve human-in-the-loop cyber defense strategies by recognizing the distinct ways humans adapt over time compared to AI defenders.
- •For game designers: Develop better training simulations for cyber defense by incorporating attacker-aware information that enhances human decision adapting during gameplay.
Authors
Zahra Aref, Sheng Wei, Narayan B. Mandayam
Abstract
Human-agent evaluations often compress interaction into a single performance score, even when human and automated policies adapt differently over time. We study this in a sequential cyber-defense game on an attack graph, where a human or reinforcement-learning defender protects cloud assets against a Deep Q-Network (DQN) attacker. We compare four defender settings: a human reward-only game operationalizing the DQN information structure, a human reward-plus-transition game operationalizing the Cognitive Hierarchy Theory-driven DQN (CHT-DQN) information structure, an automated DQN defender, and an automated CHT-DQN defender. In the reward-only game, participants receive payoff and reward feedback. In the reward-plus-transition game, they also see attacker-aware transition probabilities from the CHT-DQN model. Across 80 Mechanical Turk participants and matched automated simulations, human defenders adapted to outcomes in a way the automated defenders did not: they were more likely to reselect a node after a successful defense than after a failure, in both games, an asymmetry we interpret as consistent with Prospect Theory and Cumulative Prospect Theory. The 40-round average also mixes early rounds, where the DQN attacker acts mostly at random, with late rounds, where it mostly exploits; in the final stage, mean protection ranks the reward-plus-transition human game above the reward-only human game, then the automated CHT-DQN defender, then the automated DQN defender, an ordering the overall mean hides. By contrast, the reward-plus-transition game does not produce a statistically reliable overall gain in weighted data protection over the reward-only game. These results suggest that evaluating human-agent cyber-defense systems only by an averaged task score can miss behaviorally meaningful differences in adaptation, action allocation, and bounded human reasoning.