Home – Home – Jornals – Philosophy of Sciences 2025 number S5

The article examines the philosophical foundations and limits of artificial intelligence rationality within the context of decision-making. The author analyzes the distinction between epistemic and practical rationality, emphasizing the latter as the fundamental basis for rational agent functioning. Central to the discussion is the concept of bounded rationality, according to which decision-making occurs under conditions of incomplete information, cognitive constraints, and limited computational resources. The article argues that ideal rationality is unattainable, making bounded rationality the most appropriate model for artificial intelligence. Four types of rationality are explored: ideal, computational, bounded rationality, and bounded optimality. The study concludes that bounded rationality is practically most applicable in developing intelligent systems. Furthermore, it highlights the necessity of considering these limitations when designing adaptive algorithms and underscores the significance of human involvement in the decision-making process to enhance reliability. The «human-in-the-loop» model is interpreted not merely as a technical mode of interaction, but as an embodiment of situational rationality, taking into account context, moral implications, and the uniqueness of each specific situation. Human participation adds a value-based and interpretative dimension to algorithmic reasoning, restoring the connection between rationality and phronesis. Given that algorithms are limited by resources and prone to errors, it is precisely the human who is capable of identifying contextual nuances and providing meaningful corrections. Thus, human inclusion in the loop emerges as a critical factor in enhancing the reliability of artificial intelligence-driven decisions.