Successful AI Stems from Human-Centered Design, Fed Leaders Agree
Federal artificial intelligence (AI) technology leaders agreed today that a human-centered approach to data management and automation generates stakeholder buy-in and improves agencywide perceptions of AI projects.
At CXO Tech Forum: AI and Robotics Process Automation (RPA) in Government on Dec. 5, government officials speaking on several panels emphasized the importance of human-centered design in the development of AI capabilities.
Anil Tilbe, Director of Enterprise Measurement and Design at the Veterans Experience Office (VEO), said that human-centered design is “extremely important” in AI development. “Using human-centered design, you’re prioritizing human intelligence,” he said.
At VEO, Tilbe and his team are using information gathered from veterans and Veterans Affairs employees to build the AI environment. He suggested that the human intelligence element can help mitigate ethical concerns around AI.
Department of Labor CIO Gundeep Ahluwalia added that his agency also uses a human-centered approach to AI to improve citizen and employee experiences.
“The biggest promise for artificial intelligence, RPA, and some of these newer technologies is to allow our citizens to get better service from the government and to move our folks from low value to high value work,” the CIO said. “It’s not a solution looking for a problem. You have to start from a problem and make sure you stay focused on solving that problem,” Ahluwalia added later.
Ahluwalia said that a successful example of AI development at Labor was launched to code injury data. Before the program, employees were spending hundreds of thousands of work hours reading and coding data on workplace injuries. The agency understood that these employees had the potential to use their time more effectively, so it deployed an AI-based coding service with nearly 80 percent accuracy.
When Department of Health and Human Services (HHS) CTO Ed Simcox joined the agency, he said he also prioritized employee satisfaction in his approach to data management and AI. The CTO did a listening tour of the agency and found that HHS employees were seeking out data from other components of HHS, but regulations around health data and a strong sense of data stewardship made it difficult for employees to share information.
“We had to go to the people first,” Simcox said, “We put into practice human-centered design…to find out what [the stakeholders] needed from a data science platform. We spent many months gathering requirements at the people level…within HHS to figure out how we could create a platform that served as much of the need from an enterprise level as possible.”
Simcox and his team discovered, for example, that requirements like data use agreements (DUAs) were particularly time consuming, but a necessary party of the process to request data. To mitigate this, HHS is now building a way to automate DUAs into its data management platform.