Artificial Intelligence and Public Service: Key New Challenges

This post is part of a National Academy of Public Administration's Standing Panel on Technology Leadership recent call to action on responsibly using AI to benefit public service at all levels. What follows is an original piece I wrote for NAPA - with particular emphasis added on the parts that I think are key for success. Comments, thoughts, and additional ideas welcomed.

In May 2023, the Executive Office of the President announced actions to promote responsible AI innovation, having previously announced in October 2022 a "Blueprint for an AI Bill of Rights" to include safe and effective systems, protections against algorithmic discrimination, data privacy, notice and explanation, and alternative options to include opting-out of such systems. Such efforts raise interesting questions about (1) where can advances in AI improve delivery of Services to Citizens and (2) what changes in how public service organizes and operates are needed to make such improvements a reality?

I. Where can advances in AI improve delivery of Services to Citizens?

For FY2024 and FY2025, we can look to see signs of adoption of the following at the federal level of the U.S. government as well within state and local governments:

1. Increased use of AI-supported assistance for individuals seeking government information. For example, several cities already have “311” telephone lines and mobile apps to assist individuals with non-emergency city services as well as to provide information on programs, events, and activities in the city.
2. Increased use of AI-supported assistance for talent management and skills matching. AI will help community members find new jobs and tailor training to hone and improve their skills for upward mobility in their jobs.
3. Increased use of AI-supported review of public applications and filings. AI assistants will provide more tailored support to individuals, to better understand what they are applying for and pre-review a public application or filing prior to human approval.
4. Increased use of AI-supported legal, financial, and ethical reviews. An AI assistant will do the initial review, let an individual know if more information is required, and provide a preliminary result for final review by a human.
5. Increased use of AI-supported assistance for analyzing geospatial data. AI can assist in making sense of geospatial information— for example, data from drones for civilian purposes, private cube satellites, and sensors associated with the “internet of things” -- as well as identifying patterns of importance to improve the delivery of public services.

For the longer-term, beyond just two years, we can look to see improved Services to Citizens to include:

1. Use of AI-enabled delivery of materials and provision of transportation. This would allow public services to be paired with AI-enabled autonomous vehicles to include fire and emergency services.
2. Use of AI-enabled robots to offset repetitive and manually intensive work. This will include using AI for civil construction efforts, disaster response, healthcare, or other public functions.
3. Use of AI-enabled digital assistants to detect and help understand biases. AI will help hold up a “digital mirror” to compare decisions and help to understand where biases are and what to do.
4. Use of AI-enabled “digital twins" of real-world dynamics. AI will allow public service organizations to build models of real-world dynamics—either of actual physical assets or social interactions. Such models will create highly accurate “digital twins” that would allow individuals in public service to experiment with certain scenarios in a digital environment.
5. Use of AI to match humans into different ad-hoc teams to fit a specific public service goal or problem set.

II. What changes in how public service organizes and operates are needed to make such improvements a reality?

Consistent with the goals of a “Blueprint for an AI Bill of Rights”, public service needs to consider how to involve the public in a participatory process involving AI service deliver that does not become itself either an overly burdensome or politically fragmented process.

• Governments will need to implement something akin to “public review boards” that look at the diversity, consistency, and appropriateness of the data used.
• Public service will need to solve growing cybersecurity challenges, to include the growing challenges of disinformation and misinformation as these can erode the services of government agencies in ways akin to cyber-related disrupts too.
• AI cannot be treated as “bolt on” to existing efforts – true leadership is required to evaluate the why, what, and how associated with the mission sets of departments and how they interact with the public, with non-profits and businesses, and with other government partners.

In addition, public service will need to do extensive work on engagements with the public on how data is curated, used to train AI, and governed to ensure it is used responsibly, to include:

• Discussions of AI and Citizen Services cannot overlook the essentialness of data curation, training, and government associated with AI systems.
• Without diverse or consistent data, the AI trained by the data may make decisions that erode public trust.
• Without appropriate use of data, public trust may also erode. Such activities should involve outreach efforts by public service organizations, to increase both digital literacy and understanding of AI systems and what they can do.

III. How Can We Proceed in Advancing AI Services to Citizens in 2023 and Beyond?

Cumulatively, the points raised in the earlier two sections demonstrate that we cannot treat employment of AI Services to Citizens as “just” a technology issue – in fact improving AI and Citizen Service is best viewed as a collective set of process improvement, workforce transformation, and leadership decision-making challenges and opportunities to solve. Furthermore, we cannot sit on the sidelines as technology moves forward – we will have to “learn by doing”. This includes specific Presidential-level AI projects ideally spanning multiple departments, with top leadership support, to ensure AI delivers Citizen Services more effectively than before.

To ensure public success, Presidential-level AI projects should include intentional work on whom to include in the governance and oversight of those AI and associated data systems as well as how to ensure the diversity, consistency, and appropriateness of AI’s activities. This intentional deliberativeness will be messier than more autocratic societies, who don’t have a plurality of different perspectives and view – yet we must proceed as such and show the world that the U.S. and other open nations can do this necessary work.

To rise to this challenge, we also must develop a new science of understanding the resiliency, and by extension the brittleness, of AI apps to disruption by false data, data poisonings, jailbreaking, and other exploits if both the public and the public service workforce is to trust interactions with AI. This is a role where the National Academy of Public Administration can help alongside multiple partners in developing such a science linked to public service, administration, and engagement.

Onwards and upwards together!