“AI has the potential to help address many of our country’s pervasive problems and advance our safety, health and well-being,” said The Future Has Begun, a report on the impact of AI on government by the Partnership for Public Service and the IBM Center for the Business of Government. “That promise has great potential to transform government.”
“The most challenging problems AI may help us solve - from fighting terrorists to serving vulnerable populations - will involve government. More immediately, though not less consequentially, AI will change the way public servants do their jobs. Because of the impact AI could have on how governments ensure our safety and further our well-being, government leaders are likely to play a significant role in dictating future directions for AI.”
To illustrate how government is using AI, the report is focused on four concrete use cases being pursued by different agencies. In-depth interviews were conducted with 14 people from 10 organizations, including the various government agencies involved, as well as their collaborators in non profits and universities. Let me briefly summarize each of these four case studies.
Fighting Crime More Effectively. The first use case describes a decade long collaboration between a USC research team and the LA International Airport on how to use AI to help law enforcement units figure out how to best deploy their limited staff more effectively so they can outsmart terrorists and other criminals. Since the LA Airport police doesn’t have enough officers to staff checkpoints at all times on the eight roads leading to the airport, the USC team developed an AI system to make security schedules unpredictable. After analyzing potential targets, the system recommended randomized police patrol routes and schedules so that terrorists could not anticipate where and when they would run into security checkpoints.
The system has since been used by the Coast Guard to randomize boat patrol routes in major ports like New York and LA, and by the Transportation Security Administration to assign air marshals to flights. More recently, another version of the AI system has been developed to help rangers fight wildlife poachers around the world.
Making Tedious Tasks a Thing of the Past. This use case is about the efforts at the US Bureau of Labor Statistics to leverage AI to relieve employees of tedious, repetitive tasks and save hundreds, even thousands of work hours so they can be redeployed on more important activities. Every year, the BLS collects data on workplace injuries from a sample of around 200,000 institutions. In 2015, for example, there were almost 3 million private sector workplace injuries and illnesses and more than 750,000 from the public sector, resulting in about 300,000 incident narratives. Each narrative must be read and assigned a code to help the bureau analyze how these incidents happen and how to prevent them.
Needless to say, this is a repetitive, time consuming process, - about 25,000 work hours. While tedious for humans, it's the kind of activity that’s perfect for AI and machine learning. In 2014 the BLS started experimenting with using AI to automate the coding tasks and improve their quality and efficiency. By 2016, the most recent survey year, the AI system was able to assign nearly 50% of all codes, and proved to be more accurate, on average, than a trained human coder. This enabled the Bureau to assign employees to more complicated tasks where human judgement was required, such as calling survey respondents to get further details and clarifications.
Helping the Nation’s Most Vulnerable Populations. The third use case describes how Johnson County, Kansas, in partnership with the University of Chicago, is using AI to analyze data from different county departments to identify vulnerable populations, such as those suffering from mental health and/or substance abuse, so that they can assist them with services that could keep them out of jail.
Data shows that over 50% of people in jails around the country have mental issues. Johnson County analyzes data from its mental health center, the emergency medical services departments, law enforcement and court and corrections to identify individuals most likely to be incarcerated. In its initial pilot, a machine learning system identified 200 people among those most at-risk, basing its predictions on 252 different types of information from the combined data sets, including demographics, enrollees in mental health programs and the number of times they’d been arrested.
A little over 50% of the individuals identified by the AI system ended up in jail, a prediction estimated to be about 25% more accurate than those of case workers, who generally have access to fewer features to help them identify at-risk people. The county is working to introduce data from additional sources to improve the accuracy of its predictions.
Conquering the Complexities of Federal Purchases. The last use case describes the efforts of the Air Force to use AI to help its acquisition professionals makes sense of the highly complex regulations governing acquisitions and speed up the process of buying goods and services.
The Air Force is a huge government purchaser, having spent $53 billion on products and services in fiscal 2017, about 11% of all federal acquisitions that year. Working with two contractors, the department is now conducting an AI pilot. Air Force employees and contractors are uploading large amounts of data about the acquisition processes, including the thousands of documents describing the Federal Acquisition Regulations and the equally daunting Defense Acquisition Supplements. The goal is to teach AI systems to analyze the many documents involved to get at the meaning of the regulations, and to thus be able to assist acquisition officers and contractor on how to make good contract decisions more quickly and efficiently.
Finally, the people interviewed for the report were asked to share their lessons learned in order to help government leaders seeking to use AI.
Not every task should be augmented by artificial intelligence. AI is not appropriate for every challenge. “Agencies and project teams should first discuss what role artificial intelligence could play in their work, what tasks could AI make easier and what outcomes they expect AI to help them achieve.”
Do not underestimate the upfront investment needed. Once areas where AI can help are identified, “they need to consider the resources they will need, including experts with knowledge of AI systems and how to use them, and budgets to support implementation of the technology.”
Start small. Like most new technologies, it’s important to start with small scale projects before deploying AI at scale. “Using a pilot program enables people to get familiar and comfortable with the technology and catch errors and correct course. And it enables the system to improve.”
It is always about the data. “AI is data hungry. One of the most common challenges with using AI is data access, availability and quality. The more and better quality the data, the better its performance and accuracy. However, most government data and information is contained in separate agencies and, in many cases, the data is limited. All agencies should ensure quality data and information are available for training, testing, using and refining AI systems.”
Agency expertise in AI is crucial. “Agencies will need a robust federal AI workforce to manage the growth and potential of these technology systems.” However, since agencies will likely encounter challenges with attracting AI experts, “they should prepare for a probable shortage of AI talent in government, and look for ways to work with AI experts in the private sector and academia.”
Government could work with outside experts, particularly at colleges and universities. “Colleges and universities have a tremendous amount of artificial intelligence expertise and ongoing research and development programs and projects.” In addition, working with AI students and researchers is a good way to attract them to the public sector.
Comments