The Missing AI Skill That's Hindering Government Innovation

PHILADELPHIA — The arrival of generative artificial intelligence has created a stark division: some users champion its powerful capabilities, while others feel the much-discussed "hype" doesn't align with their experience. While there are many nuances to the topic, a foundational element is the prompt itself. That was a common thread at the National Association of State Chief Information Officers (NASCIO) Midyear Conference this week as state tech leaders identified a gap in prompt engineering skills in the public sector.

Prompt engineering is the science of crafting effective instructions or “prompts” to guide artificial intelligence models and provide it with the essential data it needs to make informed decisions. Users who give artificial intelligence language models overly simplistic commands are more likely to receive what's termed "AI slop." Conversely, an expertly crafted prompt with detailed data about the task at hand and desired output can massively enhance the result.

The challenge for governments is scaling a workforce of non-technical staff to understand the difference. An expert prompt engineer will provide necessary context, dictate tone and set output constraints, further refining responses to meet specific expectations.

While many states have wrapped up initial AI pilots, they’re finding that moving forward, building skills in how to actually use generative AI to its full potential will include a focus on training and education for prompt engineering.

In a first-of-its-kind pilot in Pennsylvania, over the course of a year, 175 employees from 14 agencies used ChatGPT Enterprise. While the results were promising, as over 85 percent of employees reported a “somewhat positive” or “very positive” experience, the pilot also found that generative AI “requires our employees’ expertise and judgment to be used effectively.”


“There’s just a learning curve of learning how to prompt and interact with something new,” said Harrison MacRae, director of emerging technologies at the Office of Administration in Pennsylvania. “We experimented with different ways to provide different training and opportunities to set people up for success. As we’re coming out of the pilot, anchoring on some of those give-offs and barriers we saw is an area of focus for any AI tool that might exist.”

Jennifer Lorenz, executive deputy chief information officer for the New York Office of Information Technology Services, told Government Technology there’s also a need for a more experienced workforce in the field in her state, a detail that will likely inspire training programs for public service workers.

“Prompt engineering is something we’re seeing a need for. The way you conduct a search with AI, the more context you can give it the better the results will be, granted, the data that’s pulling has to be accurate,” said Lorenz.


However, she added that looking 10 years into the future, it’s unlikely that states will still need to proactively seek or create training programs for employees with this specialized skill, as by then it may be a natural ability for a much larger part of the population.

“Between here and there some training will be necessary, but I don’t know for how long,” she said. “The younger folks, the people that are in school now, they’re learning this as a part of their everyday lives, so I don’t know if in the future there will be so many programs.”