Hierarchies of bias in artificial intelligence architecture: Collective, computational, and cognitive

Kudless, Andrew

doi:10.1177/14780771231170272

Cited by 5 publications

(3 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…19 Architect Andrew Kudless has highlighted similar inherent biases within the Laion dataset used to train Stable Diffusion. 20 To maintain research integrity, we removed cities visually identified as misrepresentative. Hallucinated places can be identified using a tailored CNN similarity model.…”

Section: Resultsmentioning

confidence: 99%

Synthetic Data Meets Architectural Typology: An Exploratory Computational Workflow with a Carbon Footprint Inference Case Study

Koehler

2024

112th ACSA Annual Meeting Proceedings, Disruptors on the Edge

View full text Add to dashboard Cite

This study employs large-scale-text-to-image (LLI) models to analyze building typologies in relation to environmental contexts. Here, types delineate deep compositions of places that intertwine socio-economic histories with physical structures. Rather than mere representations or models, types encapsulate various facets of a site into a specific term, acting as strategic interfaces bridging architectural design and policy-making. Moving forward, can typological thinking assist in understanding generative-AI workflows from an architectural perspective? Moreover, can one redesign types as instrumental interfaces once again linking design to their environmental contexts?The investigation examines the compositional characteristics of AI-generated images of buildings across various cities.A synthetic dataset of 150,000 images was segmented into individual building segments, enabling a statistical analysisof compositional features across 5,600 cities. The paper introduces how LLI models portray diverse local typologies anddifferentiates these using computational metrics for big-data analysis. It explores the LLI’s potential to assess the carbonfootprint of places by analyzing materials, building parts, and construction methods within the generated images throughimage segmentation.Despite only a real-world alignment of approximately 70 percent in the synthetic data, such databases can augmentexisting building datasets. Synthetic datasets are particularly useful in hard-to-access contexts and in past or projectivesettings. They allow for the precise staging of specific content, such as building typologies or perspectives. Computationalarticulation of types reveals specific attributes that transcend linear classification regimes, aiding here in assessing places’carbon footprints through multilayered linkages. The analysis indicates that embodied carbon in places does not align withgeographical carbon classifications, offering more differentiated resolutions. Furthermore, the embodied carbon,composed of multiple materials, is not visually apparent,suggesting a need for local and project-specific adaptations.

show abstract

Section: Resultsmentioning

confidence: 99%

Synthetic Data Meets Architectural Typology: An Exploratory Computational Workflow with a Carbon Footprint Inference Case Study

Koehler

2024

112th ACSA Annual Meeting Proceedings, Disruptors on the Edge

View full text Add to dashboard Cite

show abstract

“…Midjourney AI art platform is a comprehensive platform that can be used for mosque façade design that includes several key features that distinguish it from other platforms (Kudless, A. 2023).…”

Section: Results and Discussion Overview Of The Ai Platformmentioning

confidence: 99%

“…The Midjourney AI art platform emerges as a multifaceted and comprehensive tool, pioneering a revolutionary paradigm in mosque façade design, as noted by Kudless (A. 2023).…”

Section: Results and Discussion Overview Of The Ai Platformmentioning

confidence: 99%

Innovative Integration: Exploring AI Art Platforms in Architectural Education for Mosque Facade Design

Zakariya,

Susanti,

Paryoko

2023

ijess

View full text Add to dashboard Cite

This research aims to investigate the effectiveness of implementing the mentoring model in Mosque facade design presents a unique challenge in the field of architecture due to their importance as the mosque's front-facing feature and their role in representing Islamic identity. In this study, an AI art platform was introduced as a tool to improve the creativity and design process for architecture students and professionals. The study investigated the platform's novel application in the context of architectural learning for mosque façade design. The research methodology was divided into three parts: a description of the platforms used, identification of common use cases in mosque architectural design, and query analysis using NLP methods in Midjourney AI art platform. According to the findings of the study, the AI art platform is a valuable tool for architectural learning, allowing students and professionals to generate diverse and innovative designconcepts. The Midjourney AI art platform has flaws such as difficulty recognizing architectural styles from specific regions.

show abstract

Bibliography

2023

Digitisation, AI and Algorithms in African Journalism and Media Contexts

View full text Add to dashboard Cite

show abstract

Hierarchies of bias in artificial intelligence architecture: Collective, computational, and cognitive

Cited by 5 publications

References 10 publications

Synthetic Data Meets Architectural Typology: An Exploratory Computational Workflow with a Carbon Footprint Inference Case Study

Synthetic Data Meets Architectural Typology: An Exploratory Computational Workflow with a Carbon Footprint Inference Case Study

Innovative Integration: Exploring AI Art Platforms in Architectural Education for Mosque Facade Design

Bibliography

Contact Info

Product

Resources

About