Chatbots and robotic process automation can help agencies rack up quicker wins than more complex deployments of artificial intelligence, but they shouldn’t distract from IT modernization.
Three-quarters of government decision-makers struggle to select the right artificial intelligence solutions for their projects, a new report found.
Still, 61% of respondents to a KPGM survey said AI is moderately to fully functional in their organization, according to “Thriving in an AI World,” a report the professional services firm released March 9. And in the next two years, respondents said they plan to use AI to improve process automation (48%) and analytics (40%).
To determine the best AI solutions, agencies must first define their use case, said Rob Dwyer, KPMG advisory principal specializing in technology in government. Robotic process automation is a common entry point to AI in the public sector because vendors in that area are well established, and it’s relatively easy to earn small wins that can drive support for other AI efforts, he said.
There are several classes of AI, ranging from the simpler, easier-to-implement RPA and chatbots to complex natural language processing and computer vision. Use of some basic AI technologies increased during the past year: 75% of states deployed chatbots to help with questions on unemployment, COVID-19 and other topics, according to a report from the National Association of Chief Information Officers. Gartner predicted that 70% of customer interactions will involve emerging tech such as chatbots by 2022, up from 15% today.
“Do you need something to orchestrate everything and fire off automations … like an automation backbone, or do you need something to quickly interface with legacy systems in a pretty easy way? They’re not all the same,” Dwyer said of solutions. “All the providers -- even for something like RPA -- have a bit of a different approach.”
When considering their options, agencies should make use of existing technological and workforce resources, referring to technical reference models for guidance, Dwyer said. Sometimes the problem is with how to access technology, but sometimes agencies have the tools but don’t know how to use them. Additionally, more complex AI requires more modern technological foundations, like cloud.
“If you don’t have the data and you don’t have it in an accessible place, it’s hard,” he said. “When you get into the more sophisticated technologies like computer vision and actual, real cognitive [AI], that becomes a different problem set” because data has to be in the cloud but accessible to the applications – even to remote devices on the network edge, he said.
Starting small, rather than creating a detailed overarching plan, can help, too. “The state of play and the capabilities of the industry change so quickly that – I’m not saying don’t plan – you can spend a lot of time planning something that becomes out of date pretty quickly,” Dwyer said.
Overall, survey respondents are optimistic about the technology, with 77% calling for a more aggressive approach. Specifically, 79% said they believe the Biden administration will advance AI adoption, 74% said government is dedicated to training employees to use AI and 71% said workers already have the skills needed for it.
Although more advanced AI requires more advanced foundations that agencies might not yet have, opportunities for process automation are limitless, Dwyer said. But he’s concerned about agencies using automation as a reason to avoid or delay overall IT modernization.
Piecemeal automation projects might “actually inhibit true digital transformation from a process and from a constituent perspective,” he said. “You’re tying things together and doing it on the cheap and generating efficiencies, but are you really transforming a process or transforming an experience?” he asked. “I think the answer to that question is no.”
The KPMG study, conducted in January, surveyed about 150 government IT decision-makers with at least some knowledge of AI.