Artificial Intelligence and Cognition: Between Functional Performance and Structural Limitations

Recent advances in deep learning have led to the emergence of artificial intelligence systems capable of generating complex content and interacting flexibly with humans, most notably large language models (LLMs). This development poses a dual challenge for cognitive science: assessing the relevance of these systems as models of human and animal cognition, and examining their role as experimental and applied tools that are reshaping research practices. Building on an explanatory definition of cognitive models, we argue that current generative AI systems exhibit strong functional performance, but also structural limitations related to disembodiment, the absence of developmental learning, and difficulties with meaning and symbol grounding. We outline an articulation between emergent representation mechanisms and more explicit forms of symbolic reasoning. Finally, we discuss the cognitive and societal impacts of widespread AI adoption and argue for cognitive resilience, particularly through education, as a key strategy for preserving autonomy and critical thinking in AI-saturated environments.