Practice based research, School of Communication, Royal College of Art, London UK.
This research was supported by Microsoft Research Cambridge through its PhD scholarship programme.
Supervisors: Prof. Teal Triggs (RCA), Richard Banks (Microsoft Research)
Examiners: Prof. Jussi Parikka, Prof. Anthony Masure
Gitlab
Royal College of Art Repository
This PhD thesis utilises diagrams as a language for research and design practice to critically investigate algorithmic prediction. As a tool for practice-based research, the language of diagrams is presented as a way to read algorithmic prediction as a set of intricate computational geometries, and to write it through critical practice immersed in the very materials in question: data and code. From a position rooted in graphic and interaction design, the research uses diagrams to gain purchase on algorithmic prediction, making it available for examination, experimentation, and critique. The project is framed by media archaeology, used here as a methodology through which both the technical and historical "depths" of algorithmic systems are excavated.
My main research question asks:
How can diagrams be used as a language to critically investigate algorithmic prediction through design practice?
This thesis presents two secondary questions for critical examination, asking:
Through which mechanisms does thinking/writing/designing in diagrammatic terms inform research and practice focused on algorithmic prediction?
As algorithmic systems claim to produce objective knowledge, how can diagrams be used as instruments for speculative and/or conjectural knowledge production?
I contextualise my research by establishing three registers of relations between diagrams and algorithmic prediction. These are identified as: Data Diagrams to describe the algorithmic forms and processes through which data are turned into predictions; Control Diagrams to afford critical perspectives on algorithmic prediction, framing the latter as an apparatus of prescription and control; and Speculative Diagrams to open up opportunities for reclaiming the generative potential of computation. These categories form the scaffolding for the three practice-oriented chapters where I evidence a range of meaningful ways to investigate algorithmic prediction through diagrams.
This includes, the 'case board' where I unpack some of the historical genealogies of algorithmic prediction. A purpose-built graph application materialises broader reflections about how such genealogies might be conceptualised, and facilitates a visual and subjective mode of knowledge production. I then move to producing 'traces', namely probing the output of an algorithmic prediction system—in this case YouTube recommendations. Traces, and the purpose-built instruments used to visualise them, interrogate both the mechanisms of algorithmic capture and claims to make these mechanisms transparent through data visualisations. Finally, I produce algorithmic predictions and examine the diagrammatic "tricks," or 'chicanes', that this involves. I revisit a historical prototype for algorithmic prediction, the almanac publication, and use it to question the boundaries between data-science and divination. This is materialised through a new version of the almanac—an automated publication where algorithmic processes are used to produce divinatory predictions.
My original contribution to knowledge is an approach to practice-based research which draws from media archaeology and focuses on diagrams to investigate algorithmic prediction through design practice. I demonstrate to researchers and practitioners with interests in algorithmic systems, prediction, and/or speculation, that diagrams can be used as a language to engage critically with these themes.