AI Unsheathed: Testing Human–AI Collaboration Through Deck Construction in Competitive Strategy Games

his paper investigates the role of artificial intelligence as a collaborator in scientific reasoning by framing a known-answer question in the domain of competitive
strategy games. Using Flesh and Blood (FaB) and the hero Kassai as a test case, the study evaluates whether AI can navigate both formal constraints and contextual judgment through four hypotheses: metagame identification, mainboard construction, card evaluation, and sideboard design. Results show that while the AI reproduces broad descriptive patterns and aligns well with community consensus, it systematically underestimates dominant strategies, misclassifies card roles, and fails to anticipate dynamic shifts in the competitive environment. Its outputs reflect biases inherited from online sources, revealing limitations in structural reasoning and contextual adaptation. Nonetheless, the AI demonstrates competence in organizing information, synthesizing consensus knowledge, and producing structured outputs. These findings highlight both the promise and the limits of human–AI collaboration: AI can serve as a valuable assistant in knowledge synthesis, but expert oversight remains indispensable to ensure accuracy, contextual awareness, and strategic depth.