Evaluating AI “Understanding” with Cognitive-Psychological Criteria: Evidence, Gaps, and Structural Limits

Abstract

With the rapid development of large-scale language models, the performance of artificial intelligence in language understanding and reasoning tasks has become increasingly strong. As a result, many people have gradually regarded the success of tasks as the same as true understanding. However, from the perspective of cognitive psychology, understanding is not merely about giving correct or fluent responses; it is actually an internal psychological process. It includes aspects such as meaning construction, context model update, reasoning generation, and metacognitive regulation. Under such circumstances, this paper systematically studies the cognitive understanding ability of artificial intelligence from the perspective of cognitive psychology. It employs the methods of literature review and qualitative case analysis. First, it clarifies the concept definitions and evaluation criteria of understanding in cognitive psychology by combining classic theories and experimental evidence, and then uses these criteria as the analytical framework. Study the performance of “similar understanding” in contemporary artificial intelligence systems. It was found that although artificial intelligence can approach human understanding at the behavioral level, it has not reached the core cognitive psychological standards. AI lacks experience-based situational models, causal and goal-oriented reasoning mechanisms, and inherent metacognitive monitoring. These limitations are structural, not quantitative issues. It reflects the fundamental difference between artificial intelligence systems and human cognition. This article can help clarify the theoretical distinction between surface manifestations and true understanding, and also provide a cognitive psychology foundation for a more cautious explanation of artificial intelligence capabilities.