Technology plays a critical role in helping the government meet its mission of serving the public. Within the federal government, trends such as fiscal constraints, increased reliance on non-federal partners, and changing demographics are redefining the workplace. Employees must be increasingly adept at learning and using new technologies that allow them to work more effectively to meet their agencies’ evolving mission requirements. Artificial intelligence (AI) is one technology that many federal agencies are using to address these trends and challenges in new and creative ways.
What is AI, and how did we get here?
At the heart of the field of AI is the idea that machines can be used to simulate human intelligence. Since its introduction by Alan Turing in the 1940s, AI has attracted the interest and funding of the federal government and the private sector. The early 2000s saw the advent of the first virtual personal assistant and the beginning of a commercial boom. More recently, Google Duplex beat the so-called Turing test of a robot behaving in a way indistinguishable to a human.
When discussing the development of AI, I often refer to a 3-waves conceptualization developed by the Defense Advanced Research Projects Agency (DARPA), the federal agency that first funded AI work in the 1960s. The first wave involves expert knowledge and logical reasoning, such as that used in online tax preparation. The second wave involves “machine learning.” This stage is much more complicated and less transparent. This is when we start to rely on a computer’s ability to make countless decisions for us, like “what is my fastest route to work?” The more we rely on machine learning, the less we feel in control. In this wave, more research is needed before we, as a society, will feel comfortable letting it make decisions and take risks for us. Building on these first two waves, the third wave will add contextual sophistication and abstraction, and as such, will require decades more of research and development to come to fruition, if ever.
Opportunities and challenges
To the players in this increasingly crowded field, AI offers both opportunities and challenges. This duality is true of any disruptive technology, and AI is no exception. Some of AI’s opportunities include improvements in productivity and economic outcomes, better customer service, improved security, reduction in crime, and enhanced mobility. In one example, drug companies spend 10-15 years bringing a single drug to market, averaging a cost of between $600 million and $1.4 billion. AI could reduce the time and expense by finding new insights in large biomedical or health-related data sets.
Challenges related to how we give our consent to be governed also come into clearer focus for us as a constitutional society. There are challenges related to citizens’ rights to privacy and other civil liberties. Think about AI and facial recognition technology that may be scanning us and tracking our movements without our knowledge or in a manner outside our control.
There are also challenges around the accuracy of AI results, safety assurances, and algorithmic oversight. These challenges include ensuring that software upholds our society’s values and produces fair, unbiased results.
How is AI being used today?
Every day, we are giving up some of our decision-making control in thousands of micro-decisions that rely on algorithms to make our lives easier. AI is changing the federal workplace as well, allowing government employees to work more productively and cost efficiently by reducing routine tasks. As employees gain more technical and data skills, their insight into complex problems grows. At NIH, for example, AI is improving the way PubMed presents search results to its more than 2.3 million daily users. At NOAA, AI helps forecasters recognize potential severe storms more quickly and accurately. At VA, AI matches veterans with clinical trials and experimental therapeutics.
In the private sector, AI is increasingly used in many fields, including automated vehicles, financial services, criminal justice, and cybersecurity. In healthcare, AI is being used to help track flu outbreaks and extend the reach of medical providers.
As part of its work in supporting Congress, GAO has conducted assessments of the opportunities, challenges, and implications of AI. In a recent assessment of AI in drug development, GAO developed six options that policymakers could consider, including promoting basic research to generate more and better data, and creating mechanisms or incentives for increased sharing of high-quality data while also ensuring the protection of patient data.
These options have the policy goal of creating better drugs under a faster timeline, with researchers able to screen more chemical compounds and identify promising drug candidates in less time than the current process.
An agile approach to examine and leverage the real-world benefits of AI
GAO is not only assessing AI. We are also gearing up to use it ourselves. GAO has hired a Chief Data Scientist and is building an Innovation Lab that will provide a “sandbox” testing environment where analysts will be able to collaborate and apply AI to large federal data sets. In the future, one potential project could involve using AI to identify improper payments made to deceased individuals who are still receiving Medicare payments. To date, GAO has identified these payments within one data set. Being able to use advanced search and data mining analysis on multiple, very large data sets simultaneously would be an enormous step forward. As GAO’s Comptroller General has pointed out in congressional testimony, since the fiscal year 2003, cumulative improper payment estimates government-wide have totaled about $1.5 trillion.
GAO also wants to use technologies such as AI in its own auditing. For example, we see opportunities to use data science to untangle the Department of the Treasury’s data on the General Fund. This might include creating machine learning algorithms to parse a large number of disparate and unstructured financial documents, potentially generating the first-ever complete audit of this complex fund.
The successful techniques that emerge from the Innovation Lab have the potential to save billions of dollars for taxpayers across the federal government.
Moving forward into the future
If AI were embedded in every federal agency’s operations – freeing up employees to focus less on routine tasks and more on tackling grand challenges – the government could save billions of dollars more while improving service to the public in yet-to-be imagined ways.
Even if the technology stopped advancing today, the transformation to date would have far-reaching effects across our society and government. Continued attention and care will be needed to ensure the benefits are maximized while also mitigating any potential harm that may come from this transformation.
In the coming months, GAO will collaborate with leaders from across the federal government and data science community on the development of a framework for accountability and transparency in AI. This is part of GAO’s commitment to balance the technology and data science issues with the need for robust AI governance oversight – all without hindering innovation.
We’ve seen technology redefine daily life before. Think of how the computer and internet revolutions evolved, quickly pervading our lives and making the need to be computer-fluent an accepted reality.
AI will be just as pervasive, with the average person expected to understand the basics of probabilistic statistics and advanced computing. In that more complex world where we give up some control to computers to make decisions for us, we will again learn to adapt and balance opportunities with challenges. Although it will be messy and bumpy for a while, we will learn how to do this as well.
Now is the time for us to plan for that new reality.