Machine learning and statistics uncover the stories that data tell by extracting complex patterns from data. We often use those ’stories’ to make predictions about the future — for example, to categorize defendants according to the likelihood they will re-offend. In so doing, we assume that the stories told by those past data tell us something about the future — at least in part because we also assume that those data accurately and completely reflect the world. Both assumptions, however, are fragile at best — and in order to understand this fragility, we need to consider the history of the data we use: What is missing from these data? What assumptions and biases are built in to the data? How do changes in the world affect the data we record? A focus on, and understanding of, data histories will help us to make better — and better-informed — use of the stories that data tell.
Bio: Jacquelyn Burkell is currently Associate Vice-President (Research) at the University of Western Ontario. She holds a PhD in Psychology (Cognitive Science) from Western and is an associate professor in the Faculty of Information & Media Studies at UWO. A highly collaborative scholar, Jacquelyn is a co-investigator on two SSHRC partnership grants – one examining artificial intelligence in the context of justice, the other focused on youth equality and privacy online. More broadly, her research focuses on the social impact of technology and examines how technological mediation changes social interaction and information. Her recent publications include Remembering Me: Big Data, Individual Identity, and the Psychological Necessity of Forgetting (Springer, 2016).