Unlock AI’s full potential through
human-machine collaboration
September 16, 2020
Virtually every government agency is employing artificial intelligence (AI) in one way or another, and the range of government use cases appears limitless.
The Defense Department, for example, is using AI to predict aircraft maintenance needs for improved readiness. The Coast Guard uses it to analyze satellite imagery in deciding which vessels to inspect for possible smuggling. And the Health and Human Services Department is using AI to analyze departmentwide contracting data to develop smarter procurement practices.
These impressive examples represent only a tiny glimpse of how federal agencies are incorporating AI today. In an Accenture survey, 91 percent of federal executives reported that their agencies are piloting or adopting AI. That compares favorably to only 73 percent of global executives who said their organizations are using AI to some extent.
Yet while the potential for AI in government is enormous, and the use cases are many and varied, experts agree that the vast majority of AI projects today steer towards automating routine administrative tasks. “So far, most agencies have focused on automating simple manual workloads, taking baby steps in exploring AI,” said former Federal CIO Suzette Kent.1
Certainly, automation—which employs robotic processing automation (RPA), data capture, natural language processing (NLP), and other technologies—is a ripe goal for AI initiatives in government. There is no shortage of repetitive, routine, manual, rules-based tasks and procedures that tie up federal employees and contractors for countless hours when they could be using that time for far more important work that better aligns with their human strengths. For example, Accenture found that 49 percent of the average federal worker’s time is potentially augmentable with AI, trailing only the education, health and social work, and financial sectors in this potential.
Today, more and more agency leaders are coming to realize this and beginning to leverage the potential of AI systems to transform not just how they do their work, but also what work humans and machines should be doing separately and together. In short, AI is increasingly becoming a catalyst for change across the government.
1. Billy Mitchell, Suzette Kent: Toughest challenges in federal AI adoption will be in ‘middle’ layer, FedScoop, July 26, 2019, https://www.fedscoop.com/suzette-kent-toughest-challenges-ai-adoption/Consider, for example, how numerous federal agencies are collaborating today to reexamine how they approach disaster response and preparedness. The biggest challenges response teams face during a disaster are to quickly assess multiple streams of information, establish real-time situational awareness, and effectively coordinate and execute responses to limit damage, rescue survivors, and deliver medical and other aid.
Agencies such as the Defense Department’s Joint Artificial Intelligence Center (JAIC), the military services, the Federal Emergency Management Agency (FEMA), the U.S. Agency for International Development (USAID), NASA, the State Department, NOAA, the National Geospatial-Intelligence Agency, the Civil Air Patrol, the Department of Energy and various academic institutions have been working to tackle these challenges using AI-enabled tools and approaches.
Many of these organizations, for example, are partnering to develop an AI-enabled flood and damage assessment that uses overhead imagery from aircraft and satellites to identify areas where water should not be present and then assess damage based on FEMA’s assessment categories. The assessment tool also uses overhead imagery to locate road obstructions and identify routes to safely deliver supplies and remove flood victims. The net effect of this will be to increase disaster resilience, save lives, and lessen the impacts of disasters. The JAIC’s Humanitarian and Disaster Relief team has already conducted a successful first test of a prototype with the Indiana National Guard.
Such a capability can be a game-changer for government response teams in the future. With this, responders can quickly combine numerous data streams—such as weather, transportation, power grid, manpower, commerce, and satellite imagery data—to answer critical questions, such as “where to place response teams during a disaster?”
“ Planners sought to develop entirely new approaches to addressing complex mission challenges by allowing responders to engage with richer information in new ways.
Kent said researchers are now taking this capability a step further to pose the question, “How can we recover faster?”
“Where will there be trees and debris that need to be removed?” Kent said. “And what is the workforce that we need to repair flood damage? So [we can] not only use the capabilities to minimize impact, but to speed up recovery. So when you think about this type of scenario, that fundamentally changes our end-to-end workforce. … So to train our workforce, to leverage the powerful capabilities, we need not only the commitment from the technical side, but [also] mission operations and the business teams who understand and have the insights to help us identify, deconstruct, and reconstruct some of those complex interactions.”
Planners of this effort were not interested in automating existing work processes—instead, they sought to develop entirely new approaches to addressing complex mission challenges by allowing responders to engage with richer information in new ways. In this way, AI is expanding the capacities and capabilities of government disaster response teams.
Too often, the conversation around AI occurs through a narrow and negative narrative that machines and humans are in competition with each other. Instead, we should acknowledge and build upon a different view where AI’s role is to enhance human potential.
Research involving 1,500 companies has found that firms achieve the greatest performance improvements when humans and machines work together. Through their interaction and collaboration, humans and AI can enhance each other’s complementary strengths. For humans, those strengths include leadership, teamwork, creativity, and social skills. Computers are best at speed, scalability, and quantitative capabilities. For organizations to take full advantage of this collaboration, they must understand how humans can most effectively augment machines, how machines can bring out what humans do best, and, finally, how business processes can be redesigned to support the partnership.
So what does it take for government agencies to start reimagining the work itself and facilitate true human-AI collaboration? For one, they must think beyond a linear “command and response” approach and, instead, create an interactive, exploratory, and adaptable relationship. This requires an innovative set of practices that most enterprises aren’t actively building today.
Automation requires replicating specific tasks to get a job done. Using AI for augmentation, however, demands the ability to communicate and iterate with these systems. To foster human and AI collaboration, businesses will need to explore and master the tools and advancements that enable humans and machines to better engage each other—for example, new breakthroughs in natural language processing that translate into improved machine understanding of human speech and syntax.
Many commercial enterprises are also going down this path. Lemonade, a startup natively designed to use human-AI collaboration to disrupt the insurance industry, is one example with considerable applicability for customer-facing government agencies. At Lemonade, AI is embedded in the organization and present in nearly every workflow. In particular, the company’s claims payment process was designed to play to the strength of AI and humans working together. Customers file claims with a chatbot that both logs the claim details and instantly compares the claim to others within the Lemonade database—a first wave of defense against fraud. If everything looks okay, the claim can be paid out immediately to the customer. If a claim is too complex or problematic, the AI shares the information with a human agent, who steps in to manage the case.
Fraud and the administrative costs of complex bureaucracies are two of the largest costs to insurance companies, and the company solves both by making AI a key part of the process. Meanwhile, it also provides the customer with a simplified, seamless insurance experience while making a human touchpoint available when it is needed.
The notion that AI’s greatest value is not in replacing humans, but rather in enhancing the capability and potential of human employees is not new. We saw this with the advent of personal computers in the last quarter of the 20th century. While computers did assume many rote, manual tasks, such as clerical and record-keeping functions, their greatest value by far was in augmenting government and commercial employees with powerful tools—such as productivity, collaboration, advanced computation, and database software packages—to make them more informed, capable, adaptable, and independent. Federal executives get this. In our research, 85 percent of federal executives (compared to 79 percent of executives globally) acknowledge that collaboration between humans and machines will be critical to innovation in the future.
The problem, however, is that many agencies struggle with how to do that exactly. Only 18 percent of federal executives (vs. 23 percent of executives globally) reported that they are preparing their workforces for collaborative, interactive, and explainable AI-based systems. And only 22 percent of federal executives report that they have inclusive design or human-centric design principles in place to support human-machine collaboration.
of federal executives reported that they are preparing their workforces for collaborative, interactive, and explainable AI-based systems.
Agency leaders should evaluate AI’s potential in two ways—opportunities where AI can work independently to streamline operations and where it can be used collaboratively with human workers to improve their performance.
For the former, AI can play a role in preprocessing or prioritizing critical information to reduce the administrative burden. For example, the Social Security Administration is using AI to identify medical documentation that is most useful in supporting a disability claim. In many cases, agencies will need to rethink legacy process flow with AI used earlier to minimize bottlenecks and shift workers to more complex analysis or roles requiring empathy.
For the latter, it’s about considering how AI can aid the worker’s decision making and analysis. A good example is the role that AI is playing in many call centers. In addition to prepopulating forms, it can monitor a conversation to coach and recommend potential solutions to a call center representative. It’s not hard to imagine similar approaches being widely applied in fields like medicine and law enforcement.
By taking these steps, federal leaders can unlock the full potential of their workforce to operate with greater agility and effectiveness to achieve better outcomes.
So how can federal agencies do more to bring out the full power of their people? They can start by moving beyond deploying AI for automation alone and push into the new frontier of co-creation between people and machines. Natural language processing (NLP), explainable AI, and extended reality (XR) are among the tools that can unlock new ways for humans to interact with machines and for machines to interact with us.
01 Natural language processing
Meaningful collaboration always begins with communication, and traditional language barriers between humans and machines are disappearing for both written and spoken text through advancements in NLP. In many cases, these advances are pushing AI outside the realm of data scientists alone and into the core operations of the organization, giving average users the ability to command powerful systems that were unimaginable just a few years ago.
By leveraging these advances, businesses can deepen human-AI collaboration. Google’s BERT and Baidu’s ERNIE—which are both open-source frameworks—enabled AI systems to move from understanding just one word to understanding phrases in context.
02 Explainable AI
Collaboration can’t just be one-way—organizations must complete the feedback loop and build capabilities that allow humans to better understand machines. The growing field of explainable AI is letting humans de-mystify the output of previously “black-box” AI systems—making human-machine collaboration possible even if the AI wasn’t designed to explain its decision-making process, through approaches like counterfactual explanations. If a citizen is denied a loan or benefit, for example, the system needs to be able to explain the reasons for the denial and offer the smallest number of changes the applicant would need to make to have the application approved. Making AI explainable turns a human-AI interaction into a relationship.
03 Extended reality
Likewise, machines can be valuable collaborators when they can understand the physical context of humans and can sense—and make sense of—a person’s surroundings. For example, image recognition and machine learning allow Microsoft’s HoloLens 2 mixed reality headset to not only see, but also understand the wearer’s physical environment. This contextual understanding of the environment unlocks new capabilities for the device, like being able to identify dangerous equipment and warn the wearer if the equipment is operating hazardously. This significance will grow dramatically with the deployment of 5G and adoption of edge computing models.
On the Exploring AI in Government podcast series, Dr. Tim Persons, GAO’s chief scientist, shared “I think we're still underestimating how much we're going to get out of [AI] over time as it evolves. I think it's going to surprise us . . . we're going to look back and say, I can't believe we used to do things that way.” Federal leaders will need to approach AI with a similar sense of both wonder and ambition to realize its full potential.
The mixed response to the COVID-19 pandemic has underscored both the criticality of transforming available data into real insight and our current gaps. It also demonstrates the need and potential for automated systems that can address dynamic requirements. Here are considerations to keep in mind:
Federal workers will need to make decisions faster and increase their capacity and AI can help. But in our haste to deploy, we should not overlook our commitment to transparency and appropriate use, as any misuse can set user trust back significantly.
The future is likely to be more virtual, highlighting the need to get human + AI collaboration right. Federal workers will become more reliant on AI-based systems. In fact, Accenture research finds that AI tools may impact as much as 30 percent of the average federal employee’s time by 2028. However, our research also suggests that many agencies may face gaps in reskilling their workforces to effectively collaborate with AI.
