The Computational Turn in Evidence-Based Design: A Systematic Review of Artificial Intelligence Integration in Architecture (2022-2026)

The Computational Turn in Evidence-Based Design: A Systematic Review of Artificial Intelligence Integration in Architecture (2022-2026)

Saman Mohri1

1) Adjunct Lecturer, Department of Architecture and Urban Planning, Technical and Vocational University (TVU), Shiraz, Iran,

IOI Code : AACZ-ABHCBC
Publication : The 5th International Conference on Seismic Retrofitting(5icst.ir)
Abstract :
    This study presents a systematic review of the paradigmatic metamorphosis that occurred between 2022 and 2026 as a result of integrating artificial intelligence (AI) into the architectural design process. The primary objective of this review is to formulate a coherent conceptual framework for understanding the transition from rule-based parametricism toward autonomous, agentic systems, while explicating the fundamentally human-centric nature of this transformation. Grounded in the standard PRISMA protocol, the methodology involved screening 1,284 records retrieved from the Scopus and Web of Science databases, from which 61 seminal studies of the highest methodological quality were selected. The findings indicate that this metamorphosis is structured around five distinct yet interconnected pillars: generative design and diffusion models; spatial analysis and feasibility automation; multimodal large language models (LLMs) and agentic workflows; environmental and performance-driven optimization; and, finally, environmental psychology and perceptual metrics. By exploring the mathematical mechanisms of diffusion models, the multi-agent syntactic decomposition of Building Information Modeling (BIM), and particularly the quantification of eye-tracking metrics—such as fixation duration and saccade frequency—this review demonstrates the emergence of a profound technological convergence. Within this paradigm, probabilistic visual AI seamlessly intertwines with deterministic structural and environmental verification engines, as well as human perceptual metrics. The primary theoretical contribution lies in explicating the cognitive-spatial optimization loop, wherein bio-perceptual data is transformed into a feedback signal that dynamically guides AI-driven generation. Ultimately, the results indicate that the nascent paradigm does not portend the replacement of the architect; rather, it augments human agency and fosters a responsive, human-centric, and resilient built environment, contingent upon the meticulous management of data synchronization challenges and ethical considerations.

Keywords : generative ai large language models (llms) architectural design performance optimization systematic review