AI in government: AI law, use cases, and challenges

Government agencies are diverting more of their limited budgets to AI because they recognize its potential to streamline and advance current processes and systems.

For government specifically, potential AI/ML use cases include:

• Operations and management: AI can perform spend analysis, demand forecasting, and market intelligence to help teams plan, allocate resources, and determine budgets.

• Task automation: AI automates tasks and reduces repetitive busywork, such as reviewing data, monitoring suppliers, and drafting grants. 

• Cybersecurity: AI can automate incident response and threat detection, conduct risk assessments, boost vulnerability detection, and improve visibility.

• Data analysis: AI enables faster insights and decision making by collecting and analyzing data.

• Predictive analytics: AI can analyze large amounts of data to make predictions and take preventative actions. For example, it can identify real-time traffic patterns to reduce congestion at peak hours.

• Constituent support: AI-powered chatbots and voice bots help constituents find answers to frequently asked questions and get assistance faster. Call centers and 311 lines are common uses for generative AI.

Agencies can also use AI/ML for things like transportation safety, medical support, space operations, and first responder awareness.

Government agencies often use legacy systems that aren’t designed to work with AI/ML implementations.

To overcome this challenge and use AI effectively, organizations will need to modernize their data, network, cloud, and cybersecurity capabilities. This includes modernizations and improvements across:

• Data management, data cleansing, and data tagging

• Data security and Zero Trust Architecture

• Cloud infrastructure, engineering, and cloud services

AI technology advances every day, making AI law and governance a moving target. In general, though, the White House’s Blueprint for an AI Bill of Rights outlines five key principles to follow when building, using, or deploying AI systems.

1. Safe and effective systems: Systems need pre-deployment testing and ongoing monitoring to ensure they’re safe, effective, and proactively protect users. 

2. Algorithmic discrimination protections: Designers, developers, and deployers must design and use algorithms and systems in an equitable way to prevent discrimination.

3. Data privacy: Designers, developers, and deployers must create built-in data privacy protections and give users agency over how their data is collected and used.

4. Notice and explanation: Automated systems should provide clear explanations about how they’re used and how they determine outcomes that affect the user.

5. Human alternatives, consideration, and fallback: Users should be able to opt out of automated systems and work with a human instead, especially if the system fails, creates an error, or the user wants to contest the output.

The White House also released AI implementation guidance for federal agencies specifically. This includes three main pillars: 

1. Strengthening AI governance: Designate Chief AI Officers who are responsible for coordinating their agency’s AI use, advising leaders on AI, and managing AI risks.

2. Advancing responsible AI innovation: Develop an AI strategy and remove barriers to responsible AI use and maturity, such as outdated cybersecurity approval processes.

3. Managing risks from AI: Determine AI uses that impact rights and safety, follow AI risk management practices, and provide transparency.

For more guidance, check out:

If an AI model pulls from data sources with biased or inaccurate information, its output will also be biased or inaccurate. Because of this, data accuracy can be an issue for government agencies.

To mitigate this risk, agencies can use sources of information they can control, like their own websites, to train their generative AI models. They can then limit searches to these controlled sources.

Unfortunately, even controlled sources of information can be inaccurate. For example, a website may be outdated or missing certain information. Organizations that plan to power their AI tools with their website or similar sources need to ensure these sources are always up to date and accurate.