Machine Learning for Architects and Designers: Implementing Machine Learning into the Digital Design Process
MetadataShow full item record
In 1962, engineer Douglas Engelbart proposed overlapping the creative mind with artificial intelligence to create designs which could not be created by either entity alone (Engelbart 1962). Today Machine Learning (ML) has entered the public consciousness emerging as an important tool in many industries. Architects should understand these tools to be able to create new and innovative design ideas to meet complex design criteria. According to Hebron (2016) traditional design algorithms rely on the information programmed into the design software combined with a specific user input/workflow. These systems allow the computer program’s behavior to be defined as a finite set of rules that will behave in a predictable manner and thus conform to the programmer’s or user’s intentions. In comparison ML can detect patterns inside observed workflow data and provide mechanisms for imparting experiential knowledge upon computer systems. In the specific case of parametric design, rules are established by the user by defining a sequential step-by-step instruction set of geometrical operation tasks upon a set of input data. However, establishing these rules can be a time consuming and complex task. ML can help create those specific rules, if the user can define and provide the necessary input-date and desired output-data. This could lead to faster simulation and optimization methods, as well as the discovery of new parametric design rules. This paper aims to break down basic ML concepts and proposes how they could be implemented in the architectural digital design process. The focus will be put on supervised machine learning as a tool in aiding and complementing parametric design tasks. A prototype project will be showcased. The foremost aim of this paper is to lay out the hypotheses of how ML could be further implemented inside the digital design process. Further, an overview will be given of basic ML and parametric design principles, as well as demonstrating the need for architects and designers to implement ML in their design workflow.