AI Pilots in Government Seeing Positive Results

Machine learning AI modernization

Pilots and projects involving artificial intelligence (AI) technologies are producing some early wins for Federal agencies in procurement, evaluation, and other areas, government officials said on Thursday.

One of the keys to success is a pillar of pilot projects: test quickly, and fail quickly.

“You say OK, I want you to build me this item, and then I’m going to plug it into my sandbox, I’m going to drop it on my laptop, and I’m going to test it. If it breaks, I’m going to say ‘thanks, but no thanks,’” said Harrison Smith, acting chief procurement officer at the Internal Revenue Service, speaking at Government Executive’s Genius Machines event.

“When you do that in a sandbox and when you do that in a restricted environment small enough, it’s OK to have that failure, because what you’ve done is failed quickly enough to identify that there’s a risk,” Smith added.

Currently, the IRS uses automation to help with speeding up procurement and contracting procedures, with limited data availability to test the functionality of tools, and with the intent to expand the amount of data over time.

“Right now, if our bot tanks, if it does not function–which I really don’t expect it to do at all–but if it broke today, it would be worth its weight ten times over. We have learned so much in such a short amount of time,” he said.

Smith noted that the IRS will deploy robotic process automation next month to handle publicly available processes and information with an estimated return on investment of 10 times.

In another example of piloting AI, the National Science Foundation (NSF) is looking to reduce bias in finding peer reviewers, said Erwin Gianchandani, deputy assistant director for the Computer and Information Science Directorate at NSF.

“We’re using AI–this is a strategy we’ve started over the course of the last year or so, it’s a pilot effort that’s currently underway–to actually go into our database and look at past awards that we’ve made and proposals associated with those awards,” said Gianchandani. “We can tap into that and do some natural language processing and machine learning, we have the ability to potentially be able to start to identify perspective reviewers who are not conflicted with proposals, and help our program officers, who are searching for these reviewers, be able to do their work faster, more efficiently, and more effectively.”

To support the usage of AI, agencies need to get the data right. Efforts to create better data, like at the Department of Veterans Affairs, create the potential for strong use cases in the future.

“You’ve got two trillion records, there’s genomic information for 750,000 individuals–that’s several times larger than the next largest databases–and then there’s imaging and other data, so what can you do with that data?” asked Gil Alterovitz, Presidential Innovation Fellow and research affiliate with the VA. “We want to be very cognizant of the veteran’s preferences–what are the needs and questions they would like to invest, or diseases they would like to investigate? Taking all of that into account, a number of different AI solutions are being developed that are scalable, and different principles are being developed around that,” he added.