Whether bold and showy or working quietly behind the scenes, AI tools help agencies, and society at large, be more efficient, effective and often more creative. Technology that once seemed fanciful has become ubiquitous. But good AI relies on good-quality data and planning — and therein lies the rub for many government agencies.
“There are a lot of entities that view data as just part of the backend systems of record,” said Dr. Craig Orgeron, Executive Director of the Mississippi Department of Information Technology Services (ITS) and state CIO. “I don’t know that they view data in a strategic way. And I think the AI conversation does that. It’s meant to carry that issue forward.”
Trusted Sources, Truer Answers
An AI model built on reliable datasets will beat any clever AI model sitting on poor data, Orgeron said. Yet agencies often fixate on generative AI’s transformative potential, for instance, but pay scant attention to its underpinnings.
“We’re awash in data,” he said, “but I’ve been struck that in the different sectors, it’s rarely pristine, and I think there needs to be much more effort in that space if we want these AI models to perform. There’s probably more work to do than folks really let on.”
That work often includes assigning ownership, or stewardship, of priority datasets; using valid sources of record; data profiling; and metadata that makes discovery and interoperability easier. Guarding against security and privacy threats and implementing strong data governance also are vital to cultivating good data, Orgeron said.
Because think about service delivery, an area in which AI technology has become popular. An AI tool built with low-quality data can tell people the wrong deadlines, incorrectly explain legal requirements and share misleading eligibility rules, among other hazards. But with reliable sources, AI can be a real asset.
For example, Mississippi recently made its chatbot — MISSI — smarter by adding AI capabilities. Now people can more easily learn about applying for licenses and permits, state park closing times, and other straightforward details.
In the future, agentic AI embedded in enterprisescale software packages will interact with agentic AI in other software, Orgeron said. In other words, an autonomous agent in your enterprise resource planning software could work, without human interaction, with the agentic AI in your customer relationship management system. That would exponentially boost the powers of each agentic AI.
There also will be more opportunities for AI-driven analysis and auditing, summary preparation, and document intake, among other improvements. “‘Do more with less’ is an enduring government mantra, and AI tools can be very, genuinely helpful in that space,” Orgeron said.
4 Best Practices
About 80% of AI projects fail — roughly double the failure rate for IT projects without an AI component. Root causes include inadequate training data and misdirected goals, with public-sector initiatives particularly at risk from poor data management. There may be no fail-safes for developing government AI innovations, but Orgeron believes that certain best practices encourage success.
First, let use cases lead the way. Identify pain points, then build the use cases to address them. Look for lower-risk use cases that would deliver high value. From a service-delivery perspective, how can you give residents and agency employees straightforward answers faster? What AI changes will lead to visible improvements in how people request and receive services?
Next, keep humans in the loop. Some people may interpret human involvement as anti-innovative, but Orgeron doesn’t see it that way. For example, employees can play key roles monitoring for privacy and security risks. Think of AI as a digital colleague — a “force multiplier for what public servants can do in the near term,” he said, rather than a replacement for agency staff.
Build necessary precautions in early. Beyond security and privacy, plan for potential model drift, which is when AI models degrade over time because of behavioral, environmental or other changes. Responsible and explainable AI are less discussed than they used to be, but bias remains very real, Orgeron said, and AI models should be designed to address it.
In addition, build partnerships. By bringing together entities — state agencies, industry, higher education and other stakeholders — you can drive progress, create momentum and accelerate safe adoption of AI solutions. For instance, the Mississippi Artificial Intelligence Network and the state’s AI Collaborative provide statewide AI leadership, education and information sharing.
“I think that in the future, whether it’s agentic or retrieval-augmented models, the [data] that you feed it will be way more relevant to the solutions,” Orgeron said. “That’s why bringing the data conversation to the fore is the exact right thing to do.”
This article appeared in our guide, “Building Government’s Data Toolbox: Practical Uses for Data and Insights.” For more on how agencies are making real change using data, download it here:
Leave a Reply
You must be logged in to post a comment.