From Constraints to Cognition: Integrating Bounded Rationality into AI Design for Realistic Decision-Making

This study explores how bounded rationality through the mechanisms of satisficing and heuristics can be systematically embedded into AI system design to enhance human-likeness, cognitive realism, and decision-making efficiency under uncertainty, time pressure, and information overload. While prior work has largely focused on optimizing accuracy or speed, this research addresses the gap in aligning AI behavior with realistic human cognitive constraints. Using a cross-sectional quantitative design, data were collected via structured questionnaires from domain-relevant respondents. Partial Least Squares Structural Equation Modeling (PLS-SEM) was employed to test hypothesized mediation relationships between environmental factors (time pressure, uncertainty, information overload) and AI design outcomes (human-likeness, cognitive AI design, realism) via bounded rationality mechanisms. Results show that satisficing significantly mediates the relationship between time pressure and cognitive AI design (β = 0.312, p < 0.001) and between information overload and human-likeness (β = 0.287, p < 0.001). Heuristics also positively influence realism (β = 0.295, p < 0.01). Trust, transparency, and engagement demonstrated weaker direct effects (β < 0.10, p > 0.05). With 48 percent of the variation in the most important outcomes explained through mediation effects, the model proved to have a decent fit (SRMR = 0.062; NFI = 0.88). This study presents a new, experimental-verified bounded rationality-based AI design model, which can offer practitioners with actionable changes to make their algorithms appear more human and adapt to new experiences. The potential of further research in this framework is to apply it to AI applications related to a particular domain.