In their quest to modernize systems and improve service delivery, federal agency IT and program staffs are enlisting artificial intelligence (AI) and robotic process automation (RPM). These technologies, different but related, are finding application in a variety of situations, find their way into new applications. AI and RPA also find their way into the process of developing new applications themselves.
Jerry Ma, the director of emerging technology at the U.S Patent and Trademark Office, noted that the scope of prior art examiners must check in the course of processing patent applications, has increased perhaps a thousand-fold since the advent of the internet. Without technical aids, no examiner can hope to ferret out all possible instances of relevant prior art.
“AI, among other things, is what we see as absolutely essential to helping examiners,” Ma said.
For Ma, there’s no hard line between AI and RPA, and in fact the two types of software can complement one another.
“For example,” Ma said, “you could have an AI rule or an AI model that serves as the trigger to launch an RPA bot. Sort of vice versa, you can have an RPA bot launch an AI inference workflow.”
At the Securities and Exchange Commission, officials envision AI and RPA as augmenting people carrying out mission related tasks.
“Augmentation, that’s primarily our focus,” said Dr. Tanu Luke, the commission’s chief strategy and innovation officer. She said AI and ML apply across the board for the mission of investor protection. Luke said she seeks to apply RPA in document-intensive operations, to spare people from repetitive tasks such as reviewing stacks of paper for particular pieces of information.
Luke said that where RPA is “prescriptive,” AI is “descriptive,” applicable when “we don’t really know what’s in the data.” Luke cited natural language processing applied to narrative documents as an example of where AI can help the reviews and investigations processes of the SEC.
Months to minutes
Government Accountability Office Chief Scientist Dr. Tim Persons said that agencies across the board adopt RPA for rules-driven processes with what he called deterministic outcomes. At the GAO itself, he cited a labor input computation that RPA cut from months to seconds.
For such processes, Persons said, “you know your rules upfront and how to do it. You just have to process engineer and code an algorithm.”
The next wave of automation Persons said, involve larger and more varied data sets, and less deterministic “two plus two equals four” outcomes. He said facial recognition is an example of wave two as defined by the Defense Advanced Research Projects Agency (DARPA). Systems for filtering job candidates is another example of wave two. Persons added that every agency is using wave one or two.
Wave three, he said, is the “stuff of movies like a fully autonomous, thinking system.
Still another differentiator between RPA and AI, according to Melissa Long Dolson, the vice president of worldwide technology sales for AI Ops and Integration at IBM, is whether the question of ethics comes into the deployment. She noted that AI applications can bring in biases that may produce unacceptable or skewed outcomes.
“When you start thinking about AI, you’re starting really to build in human intelligence into the logarithms into the models,” Dolson said. “That changes the dynamic because AI then becomes really a person in and of itself.” Therefore the design of algorithms and the data sets used to train them require care to avoid bias.
“IBM does a lot of workshopping with our customers to make sure that we are helping them map the processes that they would be automating,” she added. “And make sure that we have the right logarithms built, and that we can train both the AI capabilities as well as the bots to do it the way a human would do it.”
Set an AI priority
Agencies on our panel apply a rigorous methodology to prioritize their RPA and AI projects in order to get the greatest returns on investment.
Luke said the SEC sees three main buckets. First come operations and document processing, and their associated repetitive tasks. Second is compliance with executive orders, in particular the application of zero trust principles to SEC networks. Third are various specific mission areas within the agency, such as examinations, enforcement, credit risk management and liquid cash management.
“Our AI applications are more focusing on how can we first and foremost to get folks within SEC enabled to do the things that they need to do in a faster way,” Luke said. Business logic for onboarding new employees or offboarding departing employees is on the horizon for AI, she added.
Because USPTO is mostly funded by patent and trademark application fees, Ma said, the highest priorities for AI and RPA “always focusing on what delivers maximum value to our stakeholders and the innovation community. That means augmentation, he said, not only of examiners’ ability to see all possible prior art but also to help dispute adjudicators make accurate and supportable decisions.
As Luke noted, AI and RPA apply to information technology operations themselves, in the SEC’s case to help establish zero trust. Persons underscored the point, saying, “We’re seeing there’s no future in cybersecurity, of dealing with things, that does not involve some AI and machine learning.”
Dolson cited another example. She said that across the board, agencies are seeking AI-powered applications to help them figure out which workloads are most suitable to cloud deployment.
“We’re helping our customers’ workload assessments. We not only provide them suggestions on which workloads should be moved and when, but also help them figure out what that work entails,” Dolson said. The algorithms take into account the variables of infrastructure needs of an application, data it requires, even the nature of the source code.
Ultimately, the GAO’s Persons said, by always keeping the mission in the forefront of thinking, the priorities for what and how to automate and speed up with AI and RPA will manifest themselves.