AI Implementation - Best Practices
The best practices for implementing artificial intelligence (AI) in a United States government agency are to include relevant stakeholders in design and monitoring, shift employees to higher-value tasks, consider ethical risks and regulate appropriately, update IT infrastructure to accommodate AI systems, and identify data heavy areas for initial implementation.
INCLUDE RELEVANT STAKEHOLDERS IN DESIGN AND MONITORING
- In implementing AI systems, government agencies should include relevant stakeholders in the design and monitoring processes by inviting employee and citizen participation and feedback, and by collaborating with academia and the private sector.
- AI experts Hollie Russon Gilman and Ari Wallach say that successful AI implementation requires a genuine, participatory approach from relevant stakeholders. This approach demystifies AI, cultivates trust, and generates user feedback.
- In North Carolina, the Charlotte-Mecklenurg Police Department partnered with an academic institution to develop an AI tool that detected officers that were more likely to commit adverse events, and the use of this tool greatly improved on the department's previous system.
SHIFT EMPLOYEES TO HIGHER-VALUE TASKS
- Government agencies should train employees to use AI in order to augment their work and shift their time to higher-value activities.
- Employees can use AI to augment their work through automation or augmentation. Automation is the process of automating mundane and repetitive tasks. Augmentation involves breaking down a job into many pieces, using AI for as many components as possible, and leaving the higher-value tasks to humans.
- Despite the fear of AI replacing human jobs, in reality, AI works best in collaboration with humans.
- Furthermore, AI can perform mundane tasks more quickly and accurately than humans, which allows human workers to focus on more complex or sensitive aspects of their jobs. This practice can also save government agencies time and money.
- The United States' Citizenship and Immigration Services department recently began using an AI chatbox that is able to answer high numbers of customer questions and can communicate in English or Spanish. By using this form of augmentation, the AI chatbox answers high-volume and direct questions, while human employees are able to shift their focus to more complex questions.
- The United Kingdom's Revenue and Customs Agency was able to reduce handling times by 40% and processing costs by 80% after automizing mundane tasks with AI and shifting their employees to higher-value tasks.
CONSIDER ETHICAL RISKS AND REGULATE APPROPRIATELY
- Agencies should focus on implementing AI that protects public values, is value-sensitive, and is frequently monitored.
- Even though in the private sector the use of AI is often protected as a trade secret, government agencies should maintain transparency.
- AI that uses predictive algorithms is susceptible to bias based on demographic categories like gender and race, and this bias must be avoided, especially in government agencies.
- If AI is not properly monitored to avoid these types of biases, it could exacerbate problems instead of alleviating them.
- AI should not be charged with making important government decisions about citizens. Because this practice is preventative, fewer published stories exist about its successful implementation. However, there are programs like Quixote, which teaches human values to AI software in order to create "ethical" AI.
- Accenture is creating a tool that would identify racial or gender bias in AI software and provide feedback on how to adjust the software to eliminate this bias, which may greatly aid the successful use of this practice.
UPDATE IT INFRASTRUCTURE TO ACCOMMODATE AI
- Government agencies should evaluate their IT systems to identify how to replace outdated systems to accommodate AI, or how to incorporate AI into existing platforms.
- Many United States government agencies continue to use and maintain outdated IT systems, which is an expense that will only continue to grow as the systems age.
- Additionally, these older systems were built with an orientation towards individual transactions instead of big data.
- Introducing AI software offers an opportunity to consider IT infrastructure holistically. Furthermore, 75% of modern business software uses AI features, so the adoption of AI is almost inevitable.
- The United States' Federal Communications Commission (FCC) created an inventory of its IT infrastructure and realized that it used numerous outdated systems. By using this holistic approach to consolidate, replace, and modify its infrastructure, the FCC was able to reduce its IT budget from spending 80% on operations and maintenance to 50%.
IDENTIFY DATA HEAVY AREAS FOR INITIAL IMPLEMENTATION
- To implement AI effectively, government agencies should identify existing data-heavy areas, which would be naturally suited for AI integration, and ways in which to collect better data for areas that are less data-heavy.
- AI systems rely on large data sets to work well. Some government agencies already have big data sets that facilitate the quick adoption of AI. However, many agencies either lack the data necessary or have failed to collect functional data that would result in a product AI system.
- Many states in the country have used AI to combat the data-heavy problem of fraudulent tax claims.
- In New York, government officials used an AI system to identify suspicious IP addresses, which were then further analyzed to see if the tax return claim was fraudulent. This practice increased tax returns from $10-20 million a year to $40 million a year.
To identify these five best practices, your research team compiled a series of reports on strategies for AI implementation in government agencies and analyzed them for recurring patterns substantiated by case studies and statistical evidence. The reports come from credible sources with demonstrated expertise in AI. Through this analysis, we were able to select five best practices that most accurately and thoroughly captured the best practices for government agencies looking to implement AI systems. Furthermore, we identified additional resources that are updated regularly in order to provide a contemporary picture of a process that is constantly growing and changing.
The following resources provide additional information on recent implementations of artificial intelligence systems in government agencies —
- Artificial Intelligence Strategies: This article summarizes the implementation strategies of AI for dozens of countries' governments.
- A forum called AI World Government brings together experts on implementing AI in government agencies, and its website provides information on the topic.
- The company Emerj, provides market research on AI, including its implementation in government agencies.