One can hardly go a minute in government technology these days without hearing or talking about artificial intelligence.
But a certain humbleness about the cutting-edge, quickly growing and often controversial technology can go a long way, according to the AI x Talent report from Govern For America, a nonpartisan nonprofit focused on getting more people into public service.
“Leading states have reported impressive results from pilot projects,” the report stated, adding that the use of ChatGPT has saved time for officials and helped public agencies reduce technical debt.
“At the same time, state chief AI officers also caution that many AI pilots will fail at first — and that learning from failure, and iteratively training AI models, is part of the experimentation process,” the report added.
The report’s findings were partly based on responses from AI leaders in more than a dozen states. The report comes as more states create AI task forces and leaders while elected leaders debate how to embrace artificial intelligence.
At least 32 states had AI task forces as of the report’s publication. Even so, the report says that fewer states have installed AI leaders or created “a permanent AI center of excellence in the state’s IT function.”
Another hurdle that could slow AI advancement at the state level involves scale.
According to the report, “deploying large language models at scale will likely require substantial advances in state capacity.”
Even so, the report continues, “the surge in interest in AI has created momentum to invest in underlying data science capabilities, a critical first step in any government’s AI journey.”
Signs of the advance of state-level AI, and a foreshadowing of what’s to come, were as abundant at the recent NASCIO conference as were warnings and cautionary tales.
One example came from Pennsylvania and Harrison MacRae, the director of emerging technologies at the state's Office of Administration and one of the panelists in the session that highlighted the report’s release.
He told Government Technology that a yearlong pilot in his state had 175 workers from 14 agencies using ChatGPT Enterprise — a test that resulted in more than 85 percent of participants reporting positive experiences.
“There’s just a learning curve of learning how to prompt and interact with something new,” he said during an interview separate from the panel discussion.