b'Forum: Evolving Use of Artificial Intelligence in Government More than Meets AIPart II: Building Trust, Managing RiskEdited by Michael J. KeeganThe next contribution to this forum explores the benefits of AI, but also underscores the importance for government agencies to manage real and perceived risks associated with AI. What follows is excerpted from the report More Than Meets AIPart II: Building Trust, Managing Risk, with afocus on significant challenges such as bias, security, transparency, employee knowledge, and federal budgetand procurement processes. IntroductionMany Americans have questions about effects AI technologies may have on aspects of their lives. According to an October 2018 survey, 59 percent of respondents are very concerned or somewhat concerned with job loss and displacement worries ranking highest. They also conveyed concerns about data privacy, security, hacking, and the safety of AI systems. Although these risk factors also affected public perceptions when other technologies were introduced, leaders now need to also address these concerns to foster trust as agencies rely more on AI to carry out missions. Through an executive order, an AI summit, and the creation of a website and a White House Select Committee on AI, theIt is important for federal organizations to move forwardOffice of Management and Budget and the Office of Sciencewith implementing AI technologies as they address AI risks. and Technology Policy are leading a governmentwide effortTheir approach to lessening AI risks also must evolve rapidly to maximize AIs benefits, while laying the groundwork forif they hope to use AI to address governments mostagencies to address risks responsibly. To increase the trust thepressing challenges.public and federal employees have in governments use ofBias. Bias in AI outcomes can stem from a number of AI tools, the governments strategy deals with transparency,issues, including poor-quality data, limited amounts security, technological know-how, procurement, budgeting,of data, or data that doesnt fully represent all aspects and risk management.of a matter. Knowing that biased data may lead to Understanding and Addressing AI Risksbiased results, agencies need to pay special attention to what information is being used with these new As agencies integrate AI into their work, they will havetechnologies. To address AI bias, federal organizations to pay attention to issues ranging from the ethical to theneed employees with technical acumen and data analysis practical. Top challenges include bias, security, transparency,and interpretation skills who can detect data bias and employee knowledge about AI technology, and federalinaccuracies. Experts in government need to understand budget and procurement processes. Each of these challengesthe theory behind AI, how the algorithms work, and is discussed below, along with recommendations for howhow conclusions are reached. Under the White Houses agencies could address potential concerns and developFebruary 2019 AI executive order, the National Institute strategies to mitigate them. of Standards and Technology (NIST), researchers are 62 www.businessofgovernment.org The Business of Government'