Beyond Mimesis: From Imitation to Extrapolation in Architectural Intelligence
Beyond Mimesis: From Imitation to Extrapolation in Architectural Intelligence

Summary/Abstract: The progressive encapsulation of architectural knowledge within technical systems has led to increasing automation of design and decision-making processes. Long limited to the execution of explicit rules, this automation is now undergoing a profound transformation with the rise of machine learning systems, characterised by self-regulation, probabilistic decision-making, and increasing opacity of their internal mechanisms. This article analyses this transformation in three stages. First, it revisits Shannon’s rigorous quantification of information, a prerequisite for its machine processing. Next, it examines the genesis of computability and the conjecture formulated at the Dartmouth conference, according to which intelligence and creativity can be simulated by formal procedures incorporating controlled randomness. Finally, by comparing Mario Carpo’s imitation thesis with recent results in high-dimensional geometry, particularly those of Yann LeCun, the article argues that contemporary AI cannot be described adequately as a simple ma-chine of imitation. Under a strict geometric definition of interpolation, contemporary AI systems are better characterised as operating through extrapolative generalisation than through interpolation in the ordinary sense. This leads to a redefinition of the autonomy of the architectural automaton, understood less as the capacity to reproduce precedents than as the tendency, in high-dimensional spaces, to generate outputs beyond the strict envelope of available data.

• Details
• Contents