As information visualizations are increasingly used to engage citizens on social and political issues, it is worth thinking about the ethics of visualization. The digital humanities are well positioned to bring an interdisciplinary and critical perspective to such ethics. How can we approach ethically fraught historical visualizations? How can we think about the opportunities for using AI to quickly generate visualizations? In this short paper, we describe a case study approach that an interdisciplinary group of us have taken to understand the ethics and the teaching of visualization. In particular, we will briefly describe how we have used the development of six concrete cases as a way of thinking through the ethics of specific visualizations. We take this methodological approach, drawing on traditions of philosophy and acknowledging the pedagogical potential of case studies for teaching ethics. We close by discussing key ethical issues that emerged without trying to resolve them, outlining the problems that we believe (future) practitioners should reflect on.
The first historical case study contrasts two examples of early data visualization—the line graphs of British trade data included in William Playfair’s Commercial and Political Atlas (1786) and the Diagram of a Slave Ship (1789). The latter was created and circulated by a group of British antislavery activists–in order to explore questions about the ethics of visualizing people as data, and the violence that can accompany the reduction of lives to numbers. In the first example, there are no people depicted; instead, the data of the slave trade–and therefore its violence–is masked as financial data. In the second, enslaved people are directly represented on the chart. However, in neither case are enslaved people, formerly enslaved people, or their descendants involved in the creation of the chart. This prompts a second set of questions about the relationships between those doing the visualizing and those being visualized, with implications for how we approach community-based projects today.
The second case explores what it means for data or designers to be “tainted” and the duties and responsibilities around such visualizations or the data they contain. Specifically, it deals with the legacy of the 20th century Topographische Anatomie des Menschen, often called the “Pernkopf Atlas” after its creator, Eduard Pernkopf (Czech et al 2021). The atlas is considered an inimitable work in anatomy, praised for its detail, accuracy, and artistic beauty. Yet, the work and its authors are entangled with the horrors of the Nazi regime, and the cadavers used of dubious and potentially horrific provenance.
From anatomy to physiognomy, and bridging the historical with current practice, the third case study looks at a visualization of composite or averaged faces of women across a large dataset of paintings, in a study investigating shifts in standards of beauty [Rosa and Suárez 2015]. The method itself is very similar to a process developed in the late 19th century by eugenicist Francis Galton for characterizing racial prototypes; it’s also found in computer science research (Turk and Pentland 1991) and in digital art, most famously Jason Salavon’s Amalgamation series ( ‘Jason Salavon’ n.d.). This raises questions over the degree to which such visualizations concretize, rather than merely historicize, the male gaze they attempt to describe.
The next two cases describe the first-person experiences of two visualization designers, one placed in a newsroom and another in academia, both trying to communicate complex issues to the general public in clear and accessible ways. Climate change science has been challenged due to the inherent uncertainty of scientific modelling and the difficulty of comprehending how scientific knowledge gets built (Edwards 2013). The case study describes the design process of a data designer aiming to represent a 125-page report on how proven low-carbon solutions can be scaled up globally, and in that process coming into ethical debates with their editor-in-chief on the level of simplification expected by the newspaper’s audience. Similarly, modelling the epidemic data, such as that from the COVID-19 pandemic, is not an easy task due to the complexity of the data and their algorithms and the fact that these complexities could also cause the visualization outputs to be hard for the public to understand which could lead to misinterpretation and confusion or even deception. There are multiple ethical discussions and guidelines discussed on the web on what would be the best way of presenting such information to the public (Foard 2020, Hepworth 2020, Makulec 2020, TableauFit 2020, TableauFit 2020b). An example would be using the logarithmic graphs to represent the COVID-19 deaths although experiments have shown that the readers may not understand them fully ( LSE COVID-19 2020, Romano et al. 2020).
The final case study challenges our expectations that visualizations should faithfully represent data that have not been manipulated. What if we used visualizations to find or manipulate data? As Johanna Drucker has pointed out, “Visual expressions serve not merely as representations of existing knowledge, but as primary modes of knowledge production.” (Drucker 2020). We have inherited scientific traditions of visualization that emphasize representation and consider the manipulation of data as unethical cherry-picking. Consider the DIKW (Data Information Knowledge Wisdom) pyramid, itself a visualization of sorts (Frické 2009, Rowley 2007). It bears assumptions about the relations between data and knowledge that need to be confronted. We can do this by experimenting with humanities-inspired techniques that produce data or find data through sketching. Visualizations are not just the trace of information, they can be information, or a way of making it.
We argue that the critique of visualization needs to be multifaceted, understanding visualizations as outputs of scientific processes, cultural artifacts, entertainment, and even as perceptual objects for external cognition with their unique affordances and design criteria among others. We thus hope that through these case studies, designers can reflect on their relation to data visualizations, and detangle the ethical issues embedded in their design, understanding that even seemingly small design choices carry ethical implications. We aim to present this preliminary exercise on ethics and envision to further develop it as an educational activity for students and other visualization designers and practitioners.