AI Summary: The article explores the contrasting paradigms of symbolic systems and neural networks in the context of information compression and reasoning. It highlights that symbolic systems, akin to high-pass filters, discretize information into clear categories and rules, exemplified by legal codes, while neural networks function as low-pass filters, capturing global structures through smooth representations. The discussion emphasizes that both approaches serve as mechanisms for compressing complex realities, yet they differ fundamentally in their methodologies and implications for understanding. The article raises the question of whether integrating symbolic components into AI systems is necessary, given the effectiveness of current neural network models.
