Computational Creativity in Architecture:
An Exploration of the Potential for Artificial Creativity using Generative Adversarial Networks to create Original Architectural Spaces from Text Descriptions.
What is creativity in architecture, and can the process be carried out artificially? What could artificial architecture look like? This dissertation strives to understand and measure artificial creativity’s potential to assist the architect’s creative process, particularly in the generation of architectural concepts and ideas. This understanding was found by assessing human creativity, neural networks’ underlying processes, and artificial intelligence’s current creative capacity for architectural applications. To address this, a Generative Adversarial Network, based on AttnGAN, was trained to generate images of internal and external architectural space based on text descriptions of the space, exploring computational interpretations of architectural space.
The human and artificial creative processes were compared and put into their cultural and societal context to assess the output, process and repercussions of artificial architectural ideas on the domain. We interpret that creativity is a co-creative process, where we cannot decouple human input from the network’s output. The artificial creative process is similar to our own and has been demonstrated by this paper to produce creative ideas; the model has created something humans could not. How we as humans choose to interpret this output and take it further into our design thinking is up to us.
To be published soon
Tutor: Ian Knight