TOWARDS ADVERSARIAL ARCHITECTURE
ASSOCIATION FOR COMPUTER AIDED DESIGN IN ARCHITECTURE (ACADIA) 2022.
VANGUARD PAPER AWARD RUNNER UP
Co-Authors: Antonio Furgiuele (WIT Architecture), Memo Ergezer (WIT Computer Science), Cagri Hakan Zaman (MIT Architecture)
A key technological weakness of artificial intelligence (AI) is adversarial images, a constructed form of image-noise added to an image that can manipulate machine learning algorithms but is imperceptible to humans. Over the past years, we developed Adversarial Architecture: A scalable systems approach to design adversarial surfaces, for physical objects, to manipulate machine learning algorithms.
Adversarial Architecture explores the application of adversarial images to the built environment and develops a new method of design agency to directly engage artificial intelligence. Embedding a layer of information to physical surfaces that is only perceptible to machines has many potential applications, such as uniquely identifying and tracking objects, embedding accessibility features directly to surfaces, and counter-surveillance systems in different scales. To construct an adversarial architecture a series of objects were selected: a cup, a banana, and a table. These everyday objects were chosen because their geometric, material, and spectral differences offer complex characteristics repeatedly found in the built environment. To transform the built environment to become adversarial, a workflow was developed: scanning existing objects, unrolling objects, generating adversarial noise, printing noise onto surfaces, wrapping objects with noise, testing objects, and reception of feedback from neural networks. The research offers a new framework of design possibilities to construct and apply adversarial surfaces to control how the built environment is captured by computer vision, understood by machine learning, and acted upon by neural networks. While AI has and will continue to profoundly affect the built environment, the future of design can critically inform AI.