Where federal AI is headed next

Artificial intelligence use cases are so numerous, federal agencies need to move fast. But they also need to use caution, especially in the generative branch of...

Tom Temin

February 5, 2024 1:00 am

3 min read

Governmental adoption of artificial intelligence looks like cloud computing adoption did around 2010. Agencies are figuring out the most important use cases. Simultaneously, they’re exploring the technology options.

Dan Kent, the chief technology officer at Red River Technology, said he’s encountered hundreds of use cases, each of which is addressable with one of three basic types of AI: predictive, or classic, AI; AI ops; and the more recent generative AI, also called large language model AI.

“So, three really big areas there that we help our customers with,” Kent said.

A typical, and growing, use case for classic AI, Kent said, is chat bots. Agencies “help customer service by leveraging automation to answer customers with the proper answer to any of the questions they may have,” Kent said. “Pretty much every agency is looking at some form of chatbot to deploy.”

AI-powered chatbots, he added, can produce more accurate and faster answers, and they can anticipate queries as they gain experience.

“The more you learn from the questions and your answers that you get, the more predictive you can be. And that’s really where we want to go with AI,” Kent said.

The classic AI technology also applies widely to surveillance and identification systems. Used in common office collaboration tools, AI can “listen” to meetings, summarize them, call out possible meanings and identify who spoke the most and the least, Kent said. Applied to case management systems, AI can point out trends and patterns, and help sort cases according to criteria important to the agency, such as legal complexity and therefore potential difficulty or time required to adjudicate.

The AI Ops model, Kent said, means continuous analysis of phenomena such as data flowing through an IT network and automatically invoking alterations or corrective measures. These measures would have been specified by people ahead of time. But the algorithmic system works on its own, so events occur faster and with less human labor. Kent said this is especially important as agencies try to counteract increasingly sophisticated cyber attacks.

“It’s really about taking all the data, whether it’s coming in from network flows or coming in from logs,” Kent said. “You have all the historical data, then you see what’s happening real time across the network. And these tools, whether a cyber tool, a configuration tool or a customer satisfaction tool, they leverage all of that to ensure that if any anomaly happens, they’re going to first detect the anomaly, [then] remediate that anomaly in real time.”

He added, “It’s the marriage of big data plus the analytics, and then the instrumentation to allow you to do something with the analytics to automate the process. If you have any type of anomalous behavior or issues with performance, it gets addressed without the user even knowing it’s happened. That’s the goal of AI ops.”

Generative generation

Generative AI, despite its reputation for “hallucinations” and other weird outcomes, is finding its way into federal applications, Kent said. The key to effective and safe deployment, he said, lies in careful selection of the language models used to train generative algorithms.

“I think there’s going to be a big push in the government to leverage what we call private AI,” Kent said, “having your own AI stack that would be on your premises or in your environment, and then taking and using your own large language model.” He added that many AI tools are open source and in the cloud, causing federal would-be users to proceed with caution.

Kent said generative AI also takes considerable computing resources, whether on premises or using commercial clouds.

“You have different tools that you need to build,” Kent said, including “more GPU [graphics processing unit] based systems, versus classic CPU systems. That requires a whole new way of thinking and whole new way of delivering services.”

Kent said he anticipates “the next three to six months is about understanding all the benefits of AI, and also the considerations about AI and then realizing what’s out there.”

He added, “It’s happening so fast in terms of the development and the solutions created.”