States still on a ‘learning curve’ amid generative AI’s promise
Chris Teale for GovExec
Leaders acknowledged the technology’s promise at the recent National Association of State Chief Information Officers conference, but said data management is a major obstacle preventing widespread use in government.
Generative AI has taken hold in state government, but some states have found that using the technology is harder than they thought.
Most state chief information officers, or CIOs, said their states have created advisory committees and task forces (78%), implemented policies and procedures on the tech’s use (72%), put in place practices around responsible use, flexible guardrails, security and ethics (67%), and taken inventory of uses in agencies and applications (61%), according to the National Association of State Chief Information Officers annual survey.
Meanwhile, 53% said employees in their CIO organization use generative AI tools in their daily work, while just 29% said they do not. Already, states said they are using the technology for virtual meeting transcription, cybersecurity operations, document generation and management, and software code generation. States are also piloting its use in other areas like data analytics, with a view to using it to make their employees’ work lives more efficient.
While it may be tempting to roll out generative AI across state government, leaders urged caution. Washington CIO Bill Kehoe said during a session at NASCIO’s annual conference in New Orleans last week that governments cannot just do “AI for AI’s sake.”
One of the biggest stumbling blocks for states looking to use generative AI is making sure the data the technology uses is accurate. North Carolina CIO James Weaver said having clean data is the “fuel for generative AI.” “It’s what makes the engine work,” he added during a panel discussion.
But it’s proven difficult for states. Missouri Deputy CIO Paula Peters said her team launched automated chatbots to answer residents’ questions and thought it would be easy to index all the state’s web pages and documents to provide correct information. It was not, she said, due to the sheer scale of the task. Training the chatbot was also tough as the state could not afford to put out wrong information. It made for quite a “learning curve,” Peters said.
NASCIO has previously warned states to make sure any AI models are trained on high-quality data. A joint report with EY earlier this year found that states are “grappling” with establishing data management programs, and urged them to put a strategy in place to manage their data, ensure its quality and be ready for AI.
While some states have invested in data management, others are much further behind. Nebraska CIO Matthew McCarville said during one panel discussion that the state is only now building a central data office. Not having one before his tenure began in April “blew my mind,” McCarville said.
Not investing ahead of time in data governance but plowing ahead with generative AI anyway could be catastrophic, warned Chris Estes, EY’s US state and local technology leader and a report co-author, who was previously North Carolina’s CIO. He compared it to a decade ago when states suffered high-profile data breaches, and policymakers realized they needed to invest in cybersecurity.
“As soon as a generative AI technology makes a bad decision, it'll be front page news,” Estes said in an interview. “After three or four of those front-page stories about how it made a bad decision, policymakers will say, ‘Oh, wait a minute, we need to fund this, just like we need cybersecurity.’ I think the funding problem will solve itself in time, but I think it's just a matter that policymakers don't understand that the technology requires clean data to make good decisions.”
It can be tricky to focus on data management, however. Weaver, the North Carolina CIO, said it is incumbent on states not to just chase the “shiny toy” of AI but find the best use cases and use it in the right way. John Evans, chief technology advisor at tech services company World Wide Technology, said that could create a “tension between innovation and preparation,” as some may be desperate to roll out new tools, but others may prefer a more cautious approach.
Evans said organizations can recover from spending too much time on preparation. Recovering from moving too quickly is much more difficult, he said.
“In a lot of cases, you spend a lot of time on your data and your stakeholders start to say, ‘Well, we've heard about all this AI stuff. We're investing all this money, where's our outcomes?’ Spend too much time on preparation, you’ve got unhappy stakeholders,” Evans said in an interview at NASCIO. “Spend too much time on innovation without doing the data work on the front end, you end up with bad AI. You end up with AI that's hallucinating. You end up with AI that gets confused.”
NEXT STORY: For rural transit agencies, a rocky road to zero emissions