Learning Algorithms, Design, and computed space

The paper analyses and speculates on what opportunities and challenges will arise from the introduction of learning algorithms (machine learning, neural networks, etc.) in architectural and urban design. The penetration of such class of algorithms in cities and design disciplines is rapid and profound increasing both the thirst for gathering ever larger and more accurate datasets and raising the prospect of automating tasks currently performed by humans. Whilst it is understood that learning algorithms are essential tools to analyse large datasets, design disciplines have paid far less attention to how such processes are carried out, how spatial data are reformatted by algorithms which largely operate on statistical bases and, most importantly, what image of the city emerges from such processes. To unravel the complexity of the issue, it is first necessary to retrace the ideas informing the emergence of numerical procedures at beginning of the twentieth century and Artificial Intelligence in the 1950's as they allow us to project a different paradigm of how space can be analysed, structured, and changed. Finally, the paper will offer some points for speculation and further reflection on how the methods put forward through learning algorithms compare to current approaches to digital design; this will foreground their disruptive potential for a radical transformation of urban design, one that could be deployed to tackle some of the most pressing urban issue.