Embracing AI in Government with Humility, History & Optimism

A primer for governments and entrepreuners on what's next, based on what's come before.

The rapid evolution of artificial intelligence (AI) has sparked a global conversation, filled with both excitement and concerns, about its implications for society. As AI technology becomes increasingly integrated into our daily lives, it is crucial to not only amaze at its potential but address the technology’s underlying issues: privacy, security, bias, and explainability — issues made more acute in the public sector. As a kind of primer and guide, this article charts the (brief but relevant) history of “AI” in government to inform the concerns and opportunities ahead.

Looking Back: “AI” in Law Enforcement

Indeed, AI is not a new phenomenon in government. It has been used, debated, and even litigated in the context of criminal justice and law enforcement for years. From facial and sound recognition algorithms to “black box” decision-making machines, AI has been both a source of controversy and a catalyst for progress.

Let’s take a closer look at an example that underscores the importance of transparency and fairness in AI systems. Developed by Northpointe in the 2010s, COMPAS (Correctional Offender Management Profiling for Alternative Sanctions) is an algorithm used across the United States to predict the risk of a defendant committing future crimes. However, it has faced scrutiny for alleged racial biases. ProPublica’s study revealed that the algorithm had a high rate of incorrectly predicting future criminal behavior, with black defendants being disproportionately labeled as high risk. This highlights the urgent need for fair and accountable AI systems, particularly in consequential contexts such as criminal sentencing.

The Rise of Generative AI & City Policies

The AI landscape has been revolutionized by the explosive growth of generative AI, with Large Language Models (LLMs) like OpenAI’s GPT-4 and Google’s Bard leading the charge. Trained on vast amounts of text data, these models possess the ability to generate human-like text, unlocking immense creative potential and opening doors to countless applications.

In this space, two cities have emerged as leaders: Boston and Seattle. Boston has actively encouraged its employees to embrace Google’s Bard LLM system, recognizing the transformative power of AI in improving government effectiveness and efficiency. The focus is not just on how to govern AI, but also on exploring how AI can be used to enhance governance itself. Seattle has adopted a generative AI policy that acknowledges both the opportunities and risks presented by this technology. By requiring permissions for accessing or acquiring generative AI products, validating their output, and avoiding sensitive data inputs, the city demonstrates its commitment to responsible AI use. Furthermore, the formation of a Policy Advisory Team signals their dedication to formulating a comprehensive policy on generative AI.

Protecting the Public Trust

While generative AI holds incredible promise, it is essential to address the significant privacy and security concerns it poses, especially in public institutions and regulated industries. These concerns include accidental data sharing, heightened biases, security threats, and intellectual property challenges.

  • Accidental Data Sharing: Generative AI systems may inadvertently share sensitive data if not properly controlled and monitored. This could result in unauthorized access or exposure of personal information.
  • Heightened Biases: Generative AI models learn from vast amounts of data, including potentially biased sources. If not carefully managed, these biases can be perpetuated and even amplified in the generated content, leading to unfair or discriminatory outcomes.
  • Security Threats: Outputs generated by generative AI, such as code or video, may contain hidden security threats or vulnerabilities. Adversaries could exploit these weaknesses to compromise systems or gain unauthorized access to sensitive information.
  • Intellectual Property Challenges: The use of generative AI raises legal complexities, particularly in the public sector. Protecting intellectual property rights and ensuring compliance with copyright laws becomes a challenging task when AI systems generate content that may infringe upon existing works.

Lessons from New York City’s AI Law

As we grapple with these challenges, we can draw inspiration from existing legislation that paves the way for responsible AI use. According to the New York Times, NYC’s 2021 AI law offers valuable insights and practical guidelines. The law mandates that companies using AI software in hiring notify candidates, conduct annual independent audits for bias, and disclose the data collected and analyzed. Violations of these regulations are subject to fines. This groundbreaking law has even spurred the growth of the AI audit business, emphasizing the importance of accountability and transparency.

One notable aspect of the New York City law is the introduction of the “impact ratio.” This calculation measures the effect of using AI software on protected groups of job candidates, sidestepping the intricate details of algorithmic decision-making. As AI systems grow increasingly complex, achieving full explainability becomes more challenging. The impact ratio approach strikes a balance between understanding the effects of AI — through AI audits — and avoiding an overwhelming emphasis on explainability.

Embracing Opportunities for GenAIs and LLMs

As we navigate the complexities of AI integration in government, it is crucial to recognize the vast opportunities that generative AI and Large Language Models (LLMs) offer. These transformative technologies have the potential to make a significant impact in several areas:

  • PR/Communications: Public Information Officers (PIOs) can leverage LLMs to generate engaging social media posts, compelling press releases, and captivating web content efficiently. For example, ZenCity has already harnessed LLMs to enable PIOs to create content at the click of a button, amplifying their outreach.
  • Transparency: Generative AI can enhance government transparency by simplifying and organizing complex city council meetings. ClerkMinutes by HeyGov is a remarkable platform that employs AI to summarize and present easily digestible summaries of these meetings, making them more accessible to citizens while saving valuable time for City Clerks.
  • Budget Books/Strategic Plans: The laborious process of creating budget books and strategic plans can be streamlined with generative AI. By automating the generation of these documents, AI can save valuable time and resources, allowing governments to focus on crucial decision-making. MySidewalk’s powerful data retrieval and publication systems illustrate the emergent potential.
  • Customer Service/311: By automating responses to common queries, generative AI can revolutionize customer service in the public sector. Efficiently handling and resolving inquiries through AI-driven systems enhances the overall customer experience and optimizes the efficiency of 311 services.
  • Municipal Codes/Laws/Charters: The interpretation and application of municipal codes, laws, and charters can be simplified with generative AI. By providing clear and concise interpretations, AI systems assist in navigating complex legal frameworks, ensuring accurate and consistent understanding. Companies such as Camino and UpCodes are leveraging advanced tools to speed up the permitting process by parsing municipal codes, a particularly timely innovation as communities race to take advantage of funding for new infrastructure, ranging from semiconductor plants through CHIPS to bridges and roads from ARPA and IRA.
  • Grant Writing/Matching: Generative AI can automate the grant writing and matching process, helping public sector institutions secure funding more efficiently. By analyzing vast amounts of data and generating grant proposals, AI reduces the burden on organizations and increases their chances of success. The GovTech consulting firm, Ad Hoc, has leveraged AI to support agencies’ grant management programs, minimizing duplicate data and organizing grants around reliable performance metrics.
  • RFP Creation & Responses: Responding to Requests for Proposals (RFPs) is another area where generative AI can make a significant impact. By automating the generation of RFP responses, AI systems streamline the process, enabling public sector institutions to secure contracts more efficiently.

Looking Ahead & Finding Balance

As promising as those concepts may be, I suspect they merely scratch the surface of what is possible — for better or for worse. As AI continues to evolve and become more integrated into our lives, we must approach its use thoughtfully, especially in the public sector. The opportunities presented by generative AI and LLMs in government are vast, and with careful planning and implementation, they can significantly enhance efficiency, transparency, and accessibility. And at the same time, careless use of machine learning and generative AI can threaten privacy and security, all the while undermining democratic values of fairness and explainability.

So we must find a balance: a balance of principles and practicalities, of policies and personalities, and of means and ends. And finding that balance will be the challenge for good governance — and for good stewards, both public and private — in the age of AI.