The Connectionist Turn: How Contemporary Generative AI Reshapes Architectural Rationality

This study examines how connectionist AI reshapes architectural rationality, focusing on the under-theorised epistemic implications of generative technologies. It positions latent space as the convergent medium of representation, cognition, and computation to investigate how learning-based models reorganise architectural reasoning. Employing a qualitative hermeneutic methodology suited to interpreting epistemic transformation, and analysing four emblematic cases, the study identified a tripartite shift: representation moves from symbolic abstraction to probabilistic, feature-based latent descriptions; cognition evolves from individual, rule-defined schemas to collective, data-inferred structures; and computation reorients from deterministic procedures to stochastic generative exploration. In this framework, type and style emerge not as fixed classifications but as continuous distributions of similarity, redefining the designer’s role from originator of form to curator of datasets, navigator of latent spaces, and interpreter of model outputs. Ultimately, the paper argues that connectionism introduces a distinct epistemic orientation grounded in correlation and probabilistic reasoning, thereby prompting critical reflection on the ethical, curatorial, and disciplinary responsibilities of AI-mediated design.