The amount of digitally available data has grown dramatically in recent years amounting, for example, in 2020 to a global Internet traffic of more than 3 zettabytes. These data present new opportunities as they can provide different answers and insights to urgent questions and lead to entirely new questions and hypotheses.
Data Science has become key to our data-driven world with a vast economic, societal, and scientific impact. Up until now, however, Data Science research has mainly focused on advancing methods with the goal to automate processes and to learn from the available data. Therefore, Data Science has often been seen as an unpredictable black box delivering unforeseeable results.
As a consequence, we need to expand the current almost exclusive emphasis on computational aspects by re-thinking and re-creating Data Science with, by and for humans and society. Centre for Human Data Society (CHDS) at the University of Konstanz will establish and extend new ways how humans and society and data science (processes/workflows) interact with each other and shall act with each other.
What We Aim to Find Out
The Human-Society-Data Science-Interdependency research in Konstanz seeks to understand the Human-Society-Data Science-Interactions, especially by identifying and distinguishing sequential and non-sequential, reactive processes, workflows, interactions, and consequences. The CHDS Initiative will focus on three main research areas – autonomy, agency, accountability - which we identified from a multidisciplinary perspective as the core to human-centered data science. This does not yet include that the areas are to be defined in the same way and have the same meaning in all disciplines. Hereto a scratch of definitions within all disciplines is needed.
Research Area 1 - Autonomy
As any action associated with Data Science is based on, initiated from or processed with data, data-related actions require some degree of data and processing autonomy. This raises the question to what degree humans and machines have data and processing autonomy and what is the significance of (not) having it. Neither individuals nor machines operate as wholly autonomous data processing or owning agents. Rather, humans and machines are causally efficacious and the relative magnitude of the contribution to the (co)determination varies depending on the level of agentic resources, spheres of activities, and contextual conditions, as well as other humans and machines involved.
Research Area 2 - Agency
To be an agent is to influence the course of data-related actions and functioning. Therefore, individual's and machines’ capabilities to exercise control over data-related actions (efficacy) is key. This entails not only the physical possibility of influencing these activities, but also the information and understanding of Data Science and its interaction with the individual and the society necessary to make decisions. As a capacity to act and give meaning to the action, agency cannot be localized in institutions, symbolic systems or in human beings but in entangled or hybrid collectives.
While the idea of complete agency is an unachievable ideal, since there are always constraints upon action, the ambition of the initiative is to make these constraints transparent and to put the human in a position of circumspect decision-making. Since ideas of agency are always dependent upon culturally shaped notions of the person, an important aspect is the question how categories such as agency (as well as autonomy and accountability) are socially contributed by individuals or attributed to machines in different ways.
Research Area 3 - Accountability
Humans set themselves goals and anticipate likely outcomes of prospective actions to guide and motivate their efforts (cf., human agency, Bandura, 1997, 2006). Hence, contingencies between data and processing actions and possible outcomes/goals as well as the expectation that actions will result in particular outcomes are also an essential aspect of being an agent accountable for the action. Furthermore, outcomes resulting from human or machine data and processing actions includes being accountable and responsible for results caused by the actions performed. While accountability is complete if these contingencies are necessary and sufficient, accountability is often partial as the contingencies are necessary but often not sufficient due to the increasing number of overlapping data-driven processes and workflows together with the development of artificial intelligent systems including numerous human-human, human-machine, and machine-machine interactions. As a consequence, the responsible human (or machine) for a data-driven workflow result may not to be ascertained.