Artificial Intelligence Thinking, Learning, and Generation: Cognitive and Neural-Inspired Foundations of Machine Learning and Deep Learning

Abstract

Artificial intelligence (AI) systems are often discussed as if they think and learn, and are capable of generating, but these claims are typically presented as metaphors rather than analyses. The development of machine learning and deep learning, particularly neural and generative models, has sparked debate over whether artificial systems truly resemble human cognition in any meaningful way or merely reproduce its features through statistical computation. In this paper, the working principles of modern AI systems are examined by situating machine learning and deep learning within the intellectual traditions of neuroscience and cognitive science. Based on recent surveys, theoretical studies, and critical views, the paper discusses how learning processes in artificial systems are motivated by and differ radically from biological thinking. It compares supervised, unsupervised, and reinforcement learning paradigms, the formation of representations in deep neural networks, and how generative models can produce outputs that seem creative but lack comprehension or intention. The paper critiques neural metaphors, cognitive analogies, and assertions of machine intelligence, arguing that interpreting what AI systems can and cannot do requires careful, specific analysis. Finally, the paper offers an interdisciplinary synthesis that helps clarify the conceptual underpinnings of contemporary AI, where anthropomorphic interpretations are flawed. It underscores the need to synthesize insights from cognitive science, neuroscience, and philosophy.