The moral of the Terminator movies is this: Don't let something as powerful as artificial intelligence run amok. For government organizations looking to improve their operations with AI, that requires a comprehensive strategy.
AI is essential technology. Our Deloitte Center for Government Insights estimates that the federal government could free up 1.2 billion work hours annually through AI-powered automation. One large state government we examined could free up nearly 34 million person-hours, saving taxpayers more than $900 million.
We're only beginning to tap AI's potential. Businesses use it to revolutionize everything from customer interactions and logistics to predicting outcomes. The public sector has also used it to automate chatbots, track disease outbreaks, filter satellite images and expand a growing list of analytic capacities.
But AI can end up growing unpredictably, from a robotic hand learning to trick cameras rather than grasp objects to artificial-life simulations designing a fast animal by evolving a very tall creature that, at high velocity, falls down. A comprehensive AI strategy can ensure that AI grows quickly, to effective use, like a vine guided by a trellis. A strategy needs to incorporate both technology and management choices; you can't get much out of technology if you can't manage it. Here, as is explored in a recent Deloitte report, are five elements for an AI strategy for government:
How ambitious is your organization? A vision should address not just an organization's AI requirements but also it's specific goals for AI and how those will fit in with the larger mission. Whether you're approaching a distinct problem or proposing agencywide transformation, a specific AI vision helps determine which initiatives to undertake.
For smart cities, this has meant considering many possibilities. Take Medellin, Colombia, which has installed its hills with soil sensors to anticipate mudslides and identify weaknesses in the municipal drainage system. Medellin also has more than 800 cameras monitoring its highway system. Many of them are "smart," predicting traffic jams and informing drivers as well as automatically calling emergency services to accidents and issuing tickets with photo evidence. Medellin has seen an 80 percent drop in traffic violations over five years.
At the focus stage, government leaders determine which initiatives should receive AI investment. This stage includes asking which applications to develop and what technological and human resources will be needed to develop them. The project might address back-end technologies, mission-focused needs or customer engagement.
Pittsburgh has focused on AI-enabled traffic lights, cutting travel time where they are installed by 25 percent. NASA plans to set up a bot management office to explore robotic process automation, hopefully assisting any employee who would like to automate a repetitive task. The Australian Navy has tested an automated gunnery assistant and a cognitive weapons engineer, reducing human effort by 85 percent. Each focus has different requirements. This is the stage to decide who will be the end user of the AI, whether it should assist or augment human intelligence or operate more independently, and how to prepare for future developments in the underlying technologies.
3. Defining Success
It's important to determine how the AI deployment will create value. For instance, the U.S. Department of Defense has recognized that many AI applications create value at the "forward edge," where users invent uses that designers never imagined. With that in mind, DoD designed its system to increase AI adoption and communicate new uses.
Strategy architects should identify appropriate performance metrics for the AI, adopt data ethics guardrails, and explicitly define performance standards in terms of accuracy, explainability, transparency and preventing bias.
An AI upgrade may require platform, data and technical upgrades. The U.S. Department of Health and Human Services, for example, recognizes that its multitude of data sets would yield more information if they were standardized, comparable and stored in a central repository.
Excellent planning still requires execution. Public-sector organizations will have to attract talent by offering the opportunity to do important work and to grow professionally, as well as training current employees in the skills needed to develop AI, or at least to work with it.
Governments have many options besides recruitment. Public-sector organizations might crowdsource through prizes and challenges, engage in private-sector partnerships, or share data across agencies with accelerators or an AI "guild" model.
5. Management Systems
Implementing an AI strategy will require the flexibility to adapt and grow with a rapidly changing, possibly unpredictable technology. Strategies will need to address technical scale-up as well as change management.
The U.S. military's Joint Artificial Intelligence Center has steeled its management for the AI future by developing a Center of Excellence, a hub that enables organizations across the Pentagon to share data, ask questions and request standardized approaches to common AI problems. Shared management approaches make it easier for talent to swap between initiatives. Established systems for scaling, pilots and deployment can grease the rails for projects as they grow. Good communication will be crucial for multiple AI initiatives to complement each other.
It's a jungle out there. That knock on the door may not be a T2000 sent from the future to eliminate you from the timeline, but any deep-learning project could be discovering ways to meet its objectives while embarrassing yours. For all the difficult decisions and rapidly evolving variables in AI, stability and clarity will accompany those who begin prepared. Designing a comprehensive AI strategy is the firmest way for government agencies to start.
Governing's opinion columns reflect the views of their authors and not necessarily those of Governing editors or management.