Potential Uses of Artificial Intelligence for the Federal Government: A Summary
Today the Disruptive Competition Project (DisCo) released a white paper titled “Potential Uses of AI for the Federal Government.” The paper highlights real and potential applications of AI for the federal government, including: combating waste, fraud, and abuse; improving the deliverance of healthcare; and supporting natural disaster relief. It then provides recommendations for the federal government to prepare for AI through three main strategies: iterating upon the national strategy for AI; modernizing federal IT infrastructure; and sharing datasets.
I. Potential Applications of AI for the Federal Government
a) Combating Fraud, Waste and Abuse
In 2012, fraud, waste, and abuse cost Medicare and Medicaid $98 billion, adding roughly 10% to the total spending of these two programs. Minimizing this loss as much as possible can therefore help make public healthcare more feasible for the federal government. AI-assisted platforms, such as Aetna’s fraud unit, have the potential to employ machine learning programs that are responsible for identifying and investigating insurance fraud. Pondera similarly has an AI-assisted platform responsible for finding improper payments in large government programs such as Medicare.
b) Improving the Deliverance of Healthcare
AI has the potential to transform healthcare by making predictions about patients, improving medical diagnostics, and monitoring patients by utilizing previously untapped but abundant sources of data. Predictions about patients can be established when data from Medicaid, Medicare, or other government programs and data from relevant sources, such as electronic health records, are plugged into machine learning techniques. These can potentially predict patient outcomes, including which individuals are most at risk of being hospitalized and even which individuals are more likely to die. A Google-developed AI platform has, under testing circumstances, outperformed human pathologists in detecting cancer, accurately evaluating biopsies 99% of the time.
Lastly, patient monitoring systems can utilize AI to predict and identify signs of deterioration in a patient’s condition so care can be given accordingly. A Philips patient monitoring system enables hospital nurses to identify deteriorating patients and reports a promising 86% reduction in cardiac arrest and a 40% decrease in intensive care mortality due to a faster response time.
c) Supporting Natural Disaster Relief
Within the last 40 years, natural disasters have caused more than 3.3 million deaths and $2.3 trillion in economic damages. AI can further aid the government before, during, and after these natural disasters by predicting when these extreme weather events will occur, which will help coordinate effective and efficient disaster programs. One Concern is one example of an AI-enabled platform supporting natural disaster relief; it currently monitors earthquakes, assesses the impact of flooding, predicts potential post-earthquake damage, analyzes infrastructure damage, and proposes evacuation routes. A Land Rover project still in the development stages employs an intelligent drone assisting in search and rescue in terrain not navigable by vehicles, broadcasting live footage back to the rescue crew vehicles and monitoring for changes to the landscape.
II. Recommendations for the Federal Government in Implementing AI
The federal government should regularly revise and iterate upon the national strategy for AI, advancing some of the original objectives and recommendations but also expanding upon them to reflect new objectives and recommendations as they become apparent. This includes making long-term investments in AI research, especially in education. Investments need to be made to the current resources in universities and other academic institutions. As an example, the vast number of academic research computers are three to five years old and currently incapable of modern AI development; this inevitably stifles AI development.
a) Iterate upon the National Strategy for AI
The federal government should regularly revise and iterate upon the national strategy for AI, advancing some of the original objectives and recommendations but also expanding upon them to reflect new objectives and recommendations as they become apparent. This includes making long-term investments in AI research, especially in education. Investments need to be made to the current resources in universities and other academic institutions. As an example, the vast number of academic research computers are three to five years old and currently incapable of modern AI development; this inevitably stifles progress.
b) Modernize Federal IT Infrastructure
To fully utilize AI technologies, federal government IT will require some upgrades. Many government agencies spend a significant portion of their IT budgets on maintenance of legacy IT systems, which limits the budget that could be spent on AI and other emerging technologies. For example, the Department of Defense relies on a 1970s system using 8-inch floppy disks.
In order to better implement AI in the future, the federal government needs to modernize the current IT infrastructure, including conducting performance measures to understand the current state of the federal government’s IT infrastructure, then digging into what needs to be modernized and determining how much it would cost to implement a new modernized system
c) Disclose and Share Datasets
To make progress quickly in AI, emphasis should be placed on making available already existing datasets held by the government, those that can be developed with federal funding, and, to the extent possible, those held by industry. The federal government should, therefore, release existing datasets that are known to be valuable as a top priority to the extent permissible by the law. Without these datasets, machine learning algorithms in the U.S. will lag behind other countries in analyzing datasets and providing more in-depth, richer insights, which is crucial to the continuing development of AI.