From Ohio's "baby bot" to driver's ed in Delaware: How states are using AI

State governments are eyeing a wide range of use cases for AI-powered tools, a new survey shows, but they face major hurdles like legacy IT, siloed data and a lack of ethical guidelines.

As in many other industries, the public sector is dipping its toe into the world of AI with chatbots. According to a new survey, nearly a quarter of US states are already deploying chatbots or digital assistants for a range of use cases. These robotic process automation (RPA) tools are helping with everything from enrolling infants in health programs to tutoring new drivers. 

executive guide

What is AI? Everything you need to know about Artificial Intelligence

A guide to artificial intelligence, from machine learning and general AI to neural networks.

Read More

At the same time, these simple tools are providing state technology leaders with an easy introduction to intelligent automation, laying the groundwork for more ambitious AI deployments. Nearly nine out of 10 states are using or have some planned use for AI, according to the survey published Tuesday by the National Association of State Chief Information Officers (NASCIO) and the Center for Digital Government, with support from IBM. Conducted in August, the survey is based on responses from CIOs and other technology leaders in 45 US states. 

"AI is an integral part of our vision going forward," Utah CIO Mike Hussey says in the NASCIO report. "For now, we're just getting our feet wet as we try to increase our understanding of this new technology. But as AI continues to mature, the technology will become pervasive throughout everything we do."

While state CIOs see the longterm potential for AI to transform how they deliver government services, there are many major hurdles they'll have to overcome before moving beyond chatbots. These include a lack of ethical frameworks -- just ine percent of respondents said they have a clear policy to ensure the responsible use of AI regarding values, ethics and laws. It also includes questions about the quality of their states' data and legacy IT that's incompatible with modern applications. 

Chatbots: A logical starting point 

Chatbots have proven to be a logical starting point for state governments, given they're easy to deploy and typically focused on single tasks. The survey showed how states have deployed them for a wide range of purposes. For instance, Ohio's Disability Onset Alert Bot helps determines eligibility for public assistance. In one five-week period, the bot cleared a backlog of 3,000 cases, the report said. 

After the success of that project, Ohio is rolling out a "Baby Bot" to enroll infants in medical coverage. The bot is part of the state's larger effort to address its higher-than-average infant mortality rate within African American communities. 

"Baby Bot is also allowing us to maintain, or in some cases reduce, the amount of staff needed to respond from a frontline perspective," Ohio CIO Ervan Rodgers,said in the report. "That ensures we can offer high-touch services for citizens versus answering basic questions that we can program into a chatbot."

In Delaware, chatbots have been deployed for a range of services, including IT help desk assistance and assistance for new drivers practicing for an upcoming exam. 

Meanwhile, in Texas, a digital assistant helps residents create accounts for the Texas.gov portal, which connects citizens to 1,100 online services and processes. Over time, officials hope that about half of Texas's 28 million citizens will create an account on the site, enabling the state to use AI to mine the data and spot trends among users. 

"The intent is to compare what services each individual is using the most and then, based on their profiles, alert people to all the available services that may be of value to them," Texas CIO Todd Kimbriel said in the report.

Future applications

Like Kimbriel, the survey respondents saw a variety of applications for AI. Nearly half (49 percent) said AI would be most valuable for analyzing the large volumes of new and existing information collected across state departments and agencies. 

aistates1.png

"Today, it may be easiest to roll out chatbots, but I think we'll see a bigger bang for the buck with machine learning, which will let us ingest and learn from all the data we own," Utah CIO Mike Hussey said. 

In an example of this, Utah has launched a pilot project with a private AI vendor to apply machine learning to video feeds from cameras mounted along freeways. The goal is to detect accidents and automatically dispatch responders. 

Another 22 percent of survey respondents said AI could be valuable for gathering and delivering information. Other valuable applications include managing personnel, translating languages, and evaluating information to determine eligibility/compliance. 

Challenges to overcome

While most states are planning to adopt AI-driven tools, there are plenty of challenges to overcome. Forty-five percent of respondents said overcoming the constraints of current IT infrastructures is the biggest hurdle. Other challenges include a lack of IT professionals with the right skill sets, as well as cultural concerns within the government.

Meanwhile, nearly three quarters of respondents (73 percent) said they are only somewhat confident in their ability to manage AI risks. More than half (59 percent) don't yet have a framework for evaluating the risk of AI implementations. There are also issues of visibility: 32 percent of survey respondents saod departments or agencies are deploying AI without central IT's involvement. 

More than nine out of 10 respondents lacked a clear policy with respect to the ethical and legal parameters for AI deployments, and only 20 percent of respondents said those policies are under development. 

Data quality and governance is also a challenge. Forty-three percent said they don't have data organized in a way that will make AI successful. In addition, only 17 percent have assessed whether data is usable and accurate enough to successfully leverage AI.