Artificial Discretion as a Tool of Governance: A Framework for Understanding the Impact of Artificial Intelligence on Public Administration

In the visa application scenario, decisional discretion is relatively low and street-level bureaucrats are more likely to semi-automatically follow algorithmic output.

“In the visa application scenario decisional discretion is relatively low and street-level bureaucrats are more likely to semi-automatically follow algorithmic output.”

In a case of low task complexity, the algorithm is more dominant and pushes out human discretion.

“In a case of low task complexity, the algorithm is more dominant and pushes out human discretion (Young, Bullock and Lecy 2019).”

The second scenario entailed a street-level bureaucrat who used an algorithm to predict welfare fraud, which reflects the predictive algorithm that is typical of more recent (and contentious) AI applications in government.

“the second scenario entailed a street-level bureaucrat who used an algorithm to predict welfare fraud, which reflects the predictive algorithm that is typical of more recent (and contentious) AI applications in government.”

AI algorithms change street-level bureaucracy: not just low discretionary decisions are automated but also medium to high discretionary decisions are being affected by automation.

“This also reflects that relatively new ways in which AI algorithms change street-level bureaucracy: not just low discretionary decisions are automated but also medium to high discretionary decisions are being affected by automation (e.g. Young, Bullock and Lecy 2019).”

Algorithms have started interfering in the discretion of more complex bureaucratic decisions by prompting decision recommendations to bureaucrats, especially in complex tasks risk assessment.

“studies suggest that algorithms have started interfering in the discretion of more complex bureaucratic decisions by prompting decision recommendations to bureaucrats, especially in complex tasks risk assessment (Bullock 2019; Young, Bullock and Lecy 2019; Zouridis, Van Eck and Bovens 2020).”

Several scholars have pointed out that automation is affecting managerial decision-making, organizational accountability and power constellations, and discretion of street-level bureaucrats.

“Several scholars have pointed out that automation is affecting managerial decision-making (Meijer, Lorenz and Wessels 2021), organizational accountability and power constellations (Zouridis, Van Eck and Bovens 2020) and discretion of street-level bureaucrats (Young, Bullock and Lecy 2019).”

Depending on the value or notion prioritized, the algorithm will reach different results and impose costs and benefits on different individuals or societal groups.

“depending on the value or notion prioritized, the algorithm will reach different results and impose costs and benefits on different individuals or societal groups”

Adoption of ADM has shifted discretion within public administrations from street-level to higher-level officials responsible for system design.

“However, the adaption of ADM has shifted discretion within the public administrations from the street-­ level to higher-­ level officials responsible for system design (Young et  al.  2019).”

Theories previously often tested on street-level bureaucrats also apply to system-level bureaucrats.

“Secondly, the theories previously often tested on street-­ level bureaucrats, also apply to system-­ level bureaucrats, which is important to acknowledge as discretion shifts from the street-­ level to system-­ level officials with the increasing use of AI governing tools (Young et al. 2019).”

Young et al. introduced artificial discretion as a framework for public managers to assess AI's impact and its differences from human discretion in task specificity and environmental complexity.

“Young et al. [128] introduced artificial discretion as a theoretical framework to help public managers assess the impact of AI and how it differs from human discretion with respect to task specificity and environmental complexity.”

Public Administration scholars recognize that information communication technologies affect human discretion and bureaucracy, with recent attention to AI.

“Public Administration scholars have also started to recognize the impact of information communication technologies (ICTs) on the nature of human discretion and bureaucracy with some recent attention paid to artificial intelligence [21,23,80,96,128].”

Higher discretionary tasks that may directly impact an individual or community are typically characterised by fuzzy success criteria and multiple interdependent systems that are difficult to model.

“Higher discretionary tasks that may directly impact an individual or community are typically characterised by fuzzy success criteria and multiple interdependent systems that are difficult to model”

The accompanying negative aspect is distancing public servants from citizens and inhibiting a rich knowledge generation avenue.

“The accompanying negative aspect is distancing public servants from citizens and inhibiting a rich knowledge generation avenue”

Project management practices are required to manage inertia towards sharing of data between government departments, status quo bias, and resistance from unions.

“Project management practices are also required to manage inertia towards sharing of data between government departments, status quo bias, and resistance from unions”

AI architectures such as neural networks are challenging to reverse engineer to determine factors and weights that produced model outputs.

“AI architectures such as neural networks are challenging to reverse engineer to determine factors and weights that produced model outputs”

However, this accuracy comes at the cost of propagating human biases and discrimination inherent in the data used for machine training.

“However, this accuracy comes at the cost of propagating human biases and discrimination inherent in the data used for machine training”

Discretion shifts from street-level to system-level officials with the increasing use of AI governing tools.

“as discretion shifts from the street-­ level to system-­ level officials with the increasing use of AI governing tools (Young et al. 2019).”

The ICT system mimics the high discretion task of the front-line caseworker concerning the assessment of unemployed chances to re-enter the labour market.

“This ICT system mimics the high discretion task of the front-line caseworker concerning the assessment of the unemployed chances to (re-)enter the labour market (Young et al., 2019).”

Deciding how to strike the balance among value trade-offs is necessarily a political and not a purely technical act.

“Deciding how to strike the balance among these is necessarily a political and not a purely technical act: depending on the value or notion prioritized, the algorithm will reach different results and impose costs and benefits on different individuals or societal groups (see also Bovens and Zouridis 2002; Young, Bullock, and Lecy 2019).”

The suitability of different AI tools varies across different policy areas and tasks, and the trade-offs involved need to be well understood and anticipated to guide the appropriate diffusion of AI in government.

“The suitability of different AI tools varies across different policy areas and tasks, and the trade‐offs involved need to be well understood and anticipated in order to guide the appropriate diffusion of AI in government (see Young, Bullock, and Lecy 2019 for such a framework as to the utility of “artificial discretion”).”