Charts and graphs are often the first or most important contact that people have with data. As such, the designers of these charts have a great deal of control over what information people get out of them. We assume that these designers are well-intentioned, but what if they are not? What damage can "black hats" do in the visualization space? What are our ethical responsibilities as chart designers?
In this talk, I discuss the space of adversarial visualizations: visualizations that distort or deceive. Drawing on results from my own work in graphical perception and statistical communication, I present examples of charts that, despite faithfully encoding the underlying data, lead to cognitive and perceptual biases, or just fail to reliably present patterns of interest in the data. I will present examples of sinister scatterplots, evil error bars, and malicious maps, and discuss alternate designs or strategies that result in improved understanding.
I will conclude with a discussion of open problems in black hat visualization, and a call to action for ethical and responsible data science.
Michael Correll is a research scientist at Tableau Software. He received his PhD. in Computer Sciences from the University of Wisconsin-Madison in 2015. His research focuses on information visualization, and more specifically on ways to present statistical information to general audiences. His other interests include graphical perception, visual rhetoric, and the digital humanities.