Machine Apophenia: The Kaleidoscopic Generation of Architectural Images
Alexey Tikhonov, Dmitry Sinyavin
TL;DR
Addresses how AI can generate architecturally meaningful imagery from random inputs and how audiences perceive AI-driven design. Introduces a kaleidoscope-inspired, machine apophenia framework and an iterative pipeline that seeds keyphrases from $50$ images, expands to $408$ terms, and refines images from $512\times512$ to $1024\times1024$ using an ensemble of CLIP-interrogation, GPT-3, BLIP2, Stable Diffusion v1.4, and SDXL. Key findings include ablation studies showing each component improves aesthetic and technical metrics, plus an observational study linking more engaging keyphrases to higher engagement (up to about $63$ percent). The work outlines implications for AI creativity and the distribution of AI-generated architectural content across social channels, highlighting both opportunities and evaluation challenges.
Abstract
This study investigates the application of generative artificial intelligence in architectural design. We present a novel methodology that combines multiple neural networks to create an unsupervised and unmoderated stream of unique architectural images. Our approach is grounded in the conceptual framework called machine apophenia. We hypothesize that neural networks, trained on diverse human-generated data, internalize aesthetic preferences and tend to produce coherent designs even from random inputs. The methodology involves an iterative process of image generation, description, and refinement, resulting in captioned architectural postcards automatically shared on several social media platforms. Evaluation and ablation studies show the improvement both in technical and aesthetic metrics of resulting images on each step.