PJ Dick Innovation Fund Project Grant: From Archive to Generative: Beyond Prompting Toward a Search-Centric Pedagogy for AI-Assisted Architectural Design
Mock-up interface of the search platform. Image credit: Jimmy Wei-Chun Cheng.
From Archive to Generative: Beyond Prompting Toward a Search-Centric Pedagogy for AI-Assisted Architectural Design
Project Lead: Jimmy Wei-Chun Cheng, Special Faculty, Carnegie Mellon Architecture
Project Team: Lynn Kawaratani, Liaison Librarian to the School of Architecture (CMU)
The rapid expansion of text-to-image generative AI is transforming visual production in architecture, creating an urgent pedagogical challenge: how to integrate these systems into the curriculum while ensuring the credibility of their outputs and addressing their opaque, black-box reasoning. In conventional studio pedagogy, architectural literacy is cultivated through precedential research, iterative critique, and dialogic negotiation between students and advisors. Achieving an image with architectural rigor is a multi-stage process that requires interpretation and judgment. By contrast, text-to-image systems accelerate production but compress this cognitive process into a single prompt, often resulting in outputs with reduced architectural coherence.
Proprietary AI platforms such as Midjourney, OpenAI, and Google Gemini further embed platform-specific censorship and generalized aesthetic assumptions. These models frequently flatten complex architectural concerns into simplified visual clichés. For example, sustainability or climate-responsive design is commonly depicted as greenery-covered buildings, while socially oriented architecture is represented through equally generic tropes. Such biases risk misguiding early-stage students who have not yet developed the architectural literacy required to evaluate AI-generated work critically.
This project proposes a search-driven framework for integrating generative AI into architectural education, linked to the Carnegie Mellon Architecture Libraries' knowledge infrastructures that relate to architecture. By combining intelligent critic agents, multimodal search, and an image generator, the framework contextualizes AI-generated images within architectural discourse. Funding from this grant will support the development of a pilot prototype and its deployment test in the undergraduate Poiesis studio. The outcomes will provide the foundation for seeking subsequent funding to scale the framework into a comprehensive, school-wide educational platform.
Image: Mock-up interface of the search platform. Image credit: Jimmy Wei-Chun Cheng.
About the Project Lead
Special Faculty
-
Established in 2023 by PJ Dick Trumbull Lindy Group, the Faculty Grants Program will award a total of $400,000 over four years beginning in 2024. The program supports faculty research and teaching innovations that address the School’s three pedagogical challenges of climate change, social justice and artificial intelligence. The proposals were assessed on their impact in furthering a faculty member’s research and teaching, their contribution to interrogating the School’s challenges, and their viability to garner further research support, make an impact on the discipline and expand the pedagogy of the School.