By: Theresa A. Pardo, Ph.D. and David A. Bray, Ph.D.
The value of public service - and the combination of citizen services and stabilizing governance process it provides - rests in the dynamic aggregation of an enormous volume of decisions and actions performed by individuals at all levels of government. Over the last forty years and more, information technologies have augmented the ability of those in government - entrusted to perform functions consistent with the Constitution, laws passed by Congress, and Executive Orders and directions provided by the Executive Branch - to communicate, process information, and make decisions.
As argued in the National Academy of Public Administration’s May 2023 Call to Action, to improve public services through the use of AI, that is to fully realize public service transformation, we must focus attention towards decision making. To do so requires both public servants and members of the public to understand the full spectrum of decision types, and increasingly to build understanding of the role AI can play in creating new capability for public value through improved decision quality.
Appreciation for the requirement that decisions makers, whether institutional or individual, have a keen understanding of decision types and the related decision processes and risks is growing in the context of AI. This is reflected in the October 30, 2023, Executive Order on Safe, Secure, and Trustworthy Artificial Intelligence which states, among other priorities that Federal Agencies must provide, “… appropriate human consideration as part of decisions that pose a high risk to rights or safety. Agencies should identify AI functionality that plays a role in decisions that pose a high risk to rights or safety and ensure that the AI functionality is not permitted to intervene directly in such situations without appropriate human consideration and accountability.”
Such human consideration requires deep understanding of which types of decisions and specific decision contexts can benefit from the use of AI and in those cases, does that use in fact pose a high risk to right or safety. The level of effort and expertise required to build that understanding across decisions will vary greatly by decision type, with the most complicated and complex, representing the greatest challenges to decision risk identification and management.
Consider three generalized types of decisions in public service:
Type 1 decisions in public service are fairly constrained and prescribed, meaning those in government that have a combination of legal and policy guidance as to what decisions they may make based on information received. In terms of applying artificial intelligence (AI) to such constrained and prescribed decisions, the range of outcomes an AI could recommend should match the similar range that humans may make in similar situations. Comparison tests between humans-alone vs. humans-plus-machines can be done to show both a consistency in decisions and one or more decision quality improvements either in terms of speed of responding, ease in which a citizen can request a decision, ease in which a citizen can understand the decision generated, or some other outcome metric.
Type 2 decisions in public service represent decisions that are more open-ended and not constrained. In these cases, both public servants and members of the public need to understand the capabilities, limits, and methods by which improvements in decision-making can occur. Understanding the capabilities of decision-making in these more open-ended decisions in public service requires first identifying methods to baseline the latitude, scope, and quality of decision making without an AI. For any organization, public or private, this may initially prove difficult however context-specific metrics to include the speed, inclusiveness or depth of public participation, cost effectiveness or other benefits of a decision-making process can be assessed or estimated to establish such a baseline. Then, as AI is added to improve a public service decision-making process, improvements in these metrics can be measured and reported.
Type 3 decisions in public service are both incredibly complicated and complex, as in they possess feedback loops, as well as lack a shared, consensus target outcome. These Type 3 decisions include decisions associated with resolving the challenges of economic growth, improving public health and health care, as well as remedying climate change impacts both within the United States and in an interconnected world. These challenging decisions will require new pilots and efforts to incorporate greater awareness both among public servants and with members of the public of the capabilities and limits associated with AI to help improve human decision-making.
For all types of decisions, increasing awareness of AI to include expert systems, decision support systems, generalized neural networks, and now generative AI (to name a few) among both public servants and the public is key. Data literacy gaps in the workforce and among the public must also be filled and data management capability at the institutional level created. Open-ended, complicated, and complex decisions as well as consideration of decision risk require decision makers who can challenge the models and the data, for example, when the AI generates insights or recommended actions that run counter to what is known or seen by the decision maker. Before employing AI in decision making, public servants and the public must understand the level of autonomy being delegated to the AI. They must consider whether the delivery systems for the AI are appropriate to the decision-making context, for example, as part of policy development by a committee or crisis response in an emergency operations center.
Ultimately, to realize improve decision quality through the use of AI, we must draw on the foundation of knowledge about human decision making and decision types and context and build understanding of AI and how it can be best employed with and on behalf of the people to increase decision quality and ensure the transparency necessary to understand and manage risk and build trust.