From AI voice assistants to delivery robots, the pandemic’s acceleration of automation has carved a noticeable presence all around us. Although many people see automation as an eventual replacement for human talent, there is an evolving human aspect that is essential to moving AI and machine learning forward. In a conversation on federal progress in AI and machine learning, Gil Alterovitz, Director of AI at the Department of Veterans Affairs (VA), and Kurt Steege, CTO at ThunderCat Technology, discussed the importance of human talent in driving AI impact.
Human interaction with AI and machine learning is a key part of the 2020 follow-up to the 2019 Executive Order on the American AI Initiative. Federal investment in data collection and integration created a need for human talent to navigate policies, make strategic investments, and drive research and development to make this data actionable. Because of this need, a stronger focus on AI and machine learning could actually lead to a talent gap. As Steege addressed, “We’ve created more data now than in the entirety of human history. It’s not just double, it’s 100 times more.” The amount of unstructured data itself has gone up from 50 percent to 80 percent. To map this data to business or mission value, human input is necessary to understanding the information, where it’s coming from, and how to establish a solid framework for it.
Alterovitz agreed with this and added, “Artificial intelligence makes recommendations to augment some decisions that people will make but having the person in the loop is an important aspect to ensuring that we really have trustworthy AI moving forward.” Alterovitz shared the VA’s current focus in exploring ways that clinicians in the field can practice medicine with AI. A human element is necessary to assess the relevance and success of the AI recommendations. “Think of it as a colleague who gives different suggestions,” said Alterovitz. “AI needs to be able to explain why it’s thinking the way it is so clinicians can decide if they want to leverage that information.”
Steege and Alterovitz both said that federal agencies must be open to collaborating with other agencies, businesses, and academia to move AI forward. “Sometimes it’s an odd mix of things,” said Steege in reference to ThunderCat’s unexpected but applicable partnerships that have arisen from data collaboration. “We had a conversation with the NIH and cancer research and compared it to sensor data at the DOE. This helps to set a different light on things.”
As the largest integrated healthcare system in the country, Alterovitz discussed how the VA is receiving its data at a very rapid pace. With over 1 billion images per year, over 9 million veterans in the healthcare system, and programs such as the Million Veteran Program and VA Data Commons Pilot as well as new sources such as wearable devices, the VA’s ongoing influx of data has created a need for computing resources to leverage this data. The VA is currently exploring new technologies and potential partners to expand its health and wellbeing initiatives.
As advice to colleagues in the federal community seeking to invest in AI and machine learning, Alterovitz emphasized the importance of having existing data sets to build upon. “Leverage what’s already been done,” said Alterovitz. For agencies in the early stages of their AI initiatives, Steege advised using existing public data sets such as those from data.gov or seeking help from a federal IT consultant such as ThunderCat.
You can watch the full discussion between Alterovitz and Steege on the state and future of AI and machine learning in the federal space here.