Automation and Artificial Discretion

Written by: Justin Bullock and Matthew Young

Robert Bifulco (Syracuse University), Justin Bullock (Texas A&M University), Jesse Lecy (Arizona State University), Jeffrey Talley (IBM Center for the Business of Government), and Matthew Young (Syracuse University) discussed pressing issues for the field of public administration related to technology and innovation. Once gathered, we quickly agreed to focus on advancements in artificial intelligence (specifically “deep learning” variants of machine learning [ML] applications), other forms of algorithmic automation, and these systems’ use of big data in Public Administration. We all immediately recognized that the use of these tools in the public sector creates immense and unforeseen possibilities for both social benefit and harm. Thus, the initial debates centered on the relative likelihood these risks or benefits would materialize, and what they could look like in practice.

Given the dynamic state of research and development on these technologies and their related disciplines, it may be unsurprising that there was significant disagreement within the group on their appropriate role in the practice of governance. Levels of optimism and skepticism about the true capacity of these systems, the likelihood of meaningful implementation, and the real benefit:cost ratio in both fiscal and social welfare contexts varied considerably. These early disagreements led to a robust conversation about the different contexts in which AI, algorithms, and big data could and should be utilized in the public sector.

Some members pointed to observable benefits from current uses of these technologies, including efficiency and effectiveness gains from “smart cities” applications, fraud detection, and national security. Others raised concerns about the demonstrable ways that such technologies amplify structural biases and inequalities, the inability to audit decisions made through deep learning ML programs, and the dangers associated with over-reliance on quantitative reductions of complex social phenomena. These debates laid the groundwork for the group to think critically about possible theoretical frameworks for understanding the context and situations in which the benefits of using these technologies are likely to outweigh the risks, as well as how to evaluate their potential use ex ante and observed use ex post. Through these discussions, a framework began to take shape around two major lenses: levels of governance and task context.

Most us agreed that both the task domain in question and the level of governance at which adoption would occur are significant factors in assessing whether the outcomes are likely to be positive or negative. Of these two factors, we concluded that task domain is likely to have the strongest effect. That said, however, our identification of level of governance as a factor also mirrored a separate working group’s call for the field to pay more explicit attention to the level of analysis – micro, meso, and macro – of both the administrative unit of interest and social outcomes. Our group’s choice of evaluative criteria largely mirrored those used in Salamon (2002): effectiveness, efficiency, equity, managerial capacity, and political legitimacy. Although we all agreed on the potential benefits of increasing effectiveness and efficiency of governance by using these technologies to support automation of large and complex tasks, the group had varying levels of concern about the potential impact on equity.

Minnowbrook played a crucial role in bringing us together to discuss a critical topic for governance that the field of Public Administration has mostly ignored outside of a select group of specialty journals. The participants of this working group strongly believe that the issue of artificial discretion, task automation, and the ever-increasing measurement and analysis of individual actions and behaviors should be a central discussion within the field. If scholars of public administration continue to ignore these issues, they do so at the peril of irrelevance to understanding 21st century governance.

References

Salamon, Lester M. 2002. The Tools of Government: A Guide to the New Governance. New York, NY: Oxford University Press.