Architectures of Learning

This chapter traces the evolution of artificial intelligence architectures, from the early perceptron to contemporary multimodal systems, emphasizing their structural logic and implications for architectural design. Beginning with artificial neural networks (ANNs), the text explains how simple neuron-like models developed into more sophisticated architectures such as convolutional neural networks (CNNs), recurrent neural networks (RNNs), and long short-term memory networks (LSTMs), each designed to capture different dimensions of data. The discussion highlights the revolutionary impact of generative adversarial networks (GANs), which introduced adversarial learning, and diffusion models, which now dominate generative AI imagery. The rise of transformers and large language models (LLMs) is presented as a major shift in natural language and multimodal reasoning, extending AI beyond text to spatial and visual domains. Emerging approaches such as Neural Radiance Fields, graph neural networks, and prototypes like SpatialGPT are explored as early steps toward spatially aware AI. The chapter frames these architectural innovations not simply as technical advances but as epistemological shifts that redefine how architects perceive, model, and collaborate with computational systems.