Return to Home
Next Story

Exploring AI Use In Government

In 2023, the Biden Administration released the first ever Federal AI Use Case Inventory. See how the federal government is using AI.

By Austin Steinhart

Fall 2024

In October 2023, the Biden Administration issued an executive order on AI to address the "extraordinary potential for both promise and peril that Artificial Intelligence (AI) possesses." The executive order establishes new standards for "AI safety and security, protecting Americans’ privacy, advancing equity and civil rights, and promoting innovation and competition, and more."

In addition to these new standards, the executive order requires that "the Director of [the Office of Management and Budget] shall, on an annual basis, issue instructions to agencies for the collection, reporting, and publication of agency AI use cases."

To comply with this order, the federal government released the first ever Federal AI Use Case Inventory in 2023. As technology continues to develop at a rapid pace, this inventory marks an important moment in prioritizing how advancements can be utilized for the public good in ways that are safe, secure, and transparent.

However, the inventory is not without its flaws. The site only provides a CSV file, neglecting to provide any summary statistics or a way to view the inventory such as in a table. As the federal government's own principles outline, this data is not accessible and is not "available to the widest range of users for the widest range of purposes."

use case download option

What exactly is a "use case of AI"? The term Artificial Intelligence (AI) has been used since at least the 1950s and generally refers to technologies that "make computers do things that are thought to require intelligence when done by people." A preferred term by many in the field is machine learning , which better captures the underlying mechanism in which computers use math and vast amounts of data to find patterns and apply it to new data.

In recent years, the meteoric rise of large language models, such as ChatGPT, have dominated the popular understanding of AI. Underlying these models is not magic, but rather mathematical concepts from the mid-century that have been applied to vast amounts of data, recently made possible by advancements in computing power. Companies in all industries have jumped on the hype and mystery around AI to begin labeling as many new technologies and services as AI as possible in order to capture a sense of super intelligence that models like ChatGPT invoked.

The Federal AI Use Case Inventory contains some of this over generalization. Technologies that have likely been used for decades in the federal government that do not use new advancements in machine learning or AI are included in this list alongside use cases that do genuinely use new technologies and AI. These misclassifications can harm our ability to establish "strong guardrails to ensure AI keeps people safe and doesn't violate their rights" as the Biden Administration has set out to achieve.

In an effort to make the inventory more digestible and explore more of these issues, scroll along below to investigate how the federal government has reported its use of AI. If you would like to view the inventory in full, skip to a sortable table of the entire inventory. Let's begin.

The Federal AI Use Case Inventory contains 710 use cases of AI as of September 1st, 2023. Each box represents one use case.
For the inventory, each department and agency submitted their own list of use cases. For example, the Department of Treasury included a Collection Voice Bot, a model to help direct a taxpayer down a certain call path given their verbal responses over the phone. Hover over a box to learn more about the specific model at any time.
The use cases are spread across 21 departments and agencies, including all 15 executive departments, illuminating that AI use has become widespread in the federal government. Just two departments, the Department of Energy and the Department of Health and Human Services, makeup almost half of all reported use cases in the inventory. The Department of Commerce, Department of Homeland Security, and Department of Veterans Affairs round out the top 5 departments. Use the drop-down below to highlight a specific department in the Other Departments category.
As discussed above, what exactly is considered a use case of AI is murky. The collected inventory shows that the federal government struggles to provide a clear understanding of this term, a common challenge for many people outside of the field. In the inventory, the term AI is incredibly expansive. It includes use cases that fit the traditional definitions of AI such as machine learning models that use large language models for chatbots or neural networks for facial recognition but also includes analytics tools, and data infrastructure, technical tools that most experts would not consider AI.
The inventory requires departments to list a technique to categorize each use case, meaning what technical tools were used. However, categories are not standardized across departments making them difficult to group. In addition, despite it being labeled as mandatory in the provided inventory instructions, over 50% of use cases have no listed technique. 10 of the 21 departments did not provide a technique for a single reported use case, including the Department of Commerce, the Department of Health and Human Services, and the Department of Veteran Affairs.
Using the use case titles and summaries, I attempted to standardize and fill in missing techniques. While this is not a perfect categorization, it does give insight into how the federal government is using AI. Many use cases listed multiple or ambiguous techniques and were categorized as unspecified machine learning. See all the created categories using the drop-down.
▸ Learn more about the techniques listed:

With the help of ChatGPT, see short definitions of each technique:

  • Machine Vision - A model focused on processing and interpreting visual information, often used in tasks like image recognition and object detection. Modern techniques often use convolutional neural networks.
  • Neural Network - A computational model inspired by the human brain, consisting of interconnected layers of nodes (neurons) that learn to recognize patterns in data.
  • Natural Language Processing (NLP) - A field of machine learning focused on enabling machines to understand, interpret, and generate human language.
  • Unspecified Machine Learning - A general categorization used when a specific technique was not clear.
  • Automation - Tools to perform repetitive tasks automatically without human intervention, often to improve efficiency and consistency. May or may not involve machine learning/AI techniques.
  • Classification - Machine learning model that categorizes data into predefined classes or labels based on input data.
  • Analytics - Using data to extract actionable insights for decision-making. May or may not involve machine learning/AI techniques.
  • Regression - A type of model (often statistical) that predicts continuous numerical values based on input data, often used for forecasting or trend analysis. May or may not involve machine learning/AI techniques.
  • Time Series Forecast - A model that predicts future values in a sequence of time-ordered data points, often used in financial markets, weather forecasting, and demand planning. May or may not involve machine learning/AI techniques.
  • Large Language Model (LLM) - A type of deep learning model, often a neural network, trained on vast amounts of text data to perform tasks like text generation, summarization, and translation.
In addition to technique, just 2% of the use cases have publicly available source code. This raises concerns about the transparency of AI tools being used by the federal government.
Also included in the inventory is information indicating a use case's stage in development. Once again, despite being a mandatory field in the provided instructions, roughly 50% have no provided development stage.
Of the use cases that did include a development stage, roughly 50% are currently in use, 25% are in development, and 25% are in planning. As expected, AI use in the federal government will only continue to increase in the years ahead.
There is much more to explore in the Federal AI Use Case Inventory. As with the first release in September 2023, we may expect an updated inventory to be released in September 2024. In the meantime, you can explore the inventory by hovering over the boxes or view the full inventory as a table below.

Return to the beginning of the story
or explore the inventory in table form below.

See code for this project here.