an exploration of critical data visualization for sustainable AI
Coded by Zeya Chen: https://observablehq.com/@08dfc3a842de1ba9/ai-incidents-forest
Abstract
As data visualization has grown and matured as a field of study and practice, and the definition of "good visualization" has become richer than ever. While data visualization has already been considered as not only pragmatic visualization for statistical purposes but also artistic visualization for discursive purposes, it is still often technically limited to defined patterns: maps, bars, lines, doughnuts, arrows, circles, etc. to build a traditional data visualization through simplification. In an increasingly quantitative world, the complexity and absolute amount of information surrounding us can be overwhelming, which in turn leads to new ways of translating this information into easily digestible forms – a recent trend in Artificial Intelligence (AI) and Machine Learning (ML)-based data visualization. The need to rethink these popular tools and patterns becomes necessary and urgent when the data does not fit the available representations, or when the available representations are too complex for the participants.
Thus, this article presents an exploration of the implications of visualizing complexity in order to move beyond the expected paradigm, not to overturn any dominant ideology in the field, but to provide a look at other possible avenues of exploration. This paper looks at a complex subject (the sustainability of AI) around the construction of recent main visualization paradigms, examining the following problems through an exploratory data visualization practice --
How to deal with a data visualization problem in which data is too complex for the data itself to be presented in an honest way?
By applying a critical or speculative design approach, can the resulting visualization provoke more thought about the subject matter than simply an accurate depiction of numerical data?
How should critical data visualization be a way to research complex design problems?
This paper broadly follows a strategy of critical and speculative design proposed by Matt Malpas (2019). Using design as a medium of inquiry, it aims to make the end result more effective than explanatory (Malpas, 2019). In this way, the goal is to encourage the audience to think critically about both the subject being visualized (sustainable AI) and the idea that visualizing complexity as a method of design research. In the words of Mazé and Redstroth, the aim is to diversify rather than simplify the understanding of the issues raised (Mazé & Redstroth, 2009). The intended end result of the practical data visualization in the paper is a post-optimal object as envisioned by Dunne, one that goes beyond what existing norms of optimal user experience and towards user unfriendliness in order to provoke thinking (Dune &Raby, 2013).
The paper consists of four parts. Firstly, I explore current discourses of data visualization theories, mainly focusing on Sarah Williams’s Data Action: Using Data for Public Good(2021) but also includes the opinions of Loukissas’s local data and D'Ignazio’s feminist data. This literature review provides a basis for the practical visualization project of this paper - grounding it in the practice of data visualization, providing perspectives for what would be called experimental within the prevalent norms of the field, and providing a starting point for the critical inquiry embodied in the practical work.
Secondly, I study some AI-related data visualization projects to understand the current state of computational data visualization. By comparing the theories of critical data visualization and practical AI-data visualization projects, it aims to uncover the need to question some of the domain modes of information production, management, and presentation. The comparison describes the gap of an actionable new type of data visualization for socio-technical design research.
Thirdly, I present my practical production part of the visualization work, describing each part of the algorithm logic and the reasons behind them, binding back to the first and second parts, and iterating the prototype based on the feedback from a short presentation.
Lastly, I conclude by assessing the success of the prototype in relation to the posed research questions. I reflect on the results and rethink the relationship between data visualization and design study.
Key Words: #data visualization, research through design, critical study, AI ethics, #sustainability
Comments