Q&A: New Jersey's Chief Innovation Officer Lifts the Curtain on Launching AI Projects

Offering staff an AI-based tool to look for common issues with documents speeds up that process. As a future iteration, offering feedback to a user at the time of upload — for example, “It looks like the document you uploaded may be missing information. Are you sure this is the right file?” — can further save time and empower applicants to fix issues and keep their applications moving quickly.

We’ve also used machine learning algorithms to match data sets from multiple programs to identify eligible recipients of other programs. These kinds of embedded, process-based use cases end up saving a lot of time and improving customer service.

STATETECH: Many governments are struggling to balance innovation and risk. How did New Jersey decide what guardrails needed to be in place before rolling out AI tools to employees?

COLE: We start with the assumption that employees, like many people generally, will be curious and eventually try AI tools. They are also being bombarded with pitches for new AI services, and AI-based features are popping up all over commonly used applications and websites.

So, we want to inform the state workforce about what generative AI is, how it works, and what it is and is not good at doing now. We want people to understand that AI tools may appear to “think” but in fact, they’re attempting to predict acceptable responses to prompts and that they’re trained on massive data sets with unverified information.

On balance, they can excel at synthesizing large amounts of information, finding patterns in data and helping to automate common tasks. But they can also present fictitious information confidently, and they may base outputs on inherently biased inputs, such as all of the content on the internet on which they’ve been trained. We offer these lessons, along with practical use cases for generative AI in public sector work, as a training program made available to the entire state workforce.

Next, we want the educational process to be practical and interactive, so we developed an internal generative AI website, the NJ AI Assistant, that works like a commercially available chatbot but is hosted on state IT infrastructure and has guides and usability improvements tailored to public sector users. Pairing training with usable tools helps enable responsible use and avoids the shadow IT that results from people finding ways to use unauthorized tools without proper security and ethical considerations in place. Ultimately, we want to empower public servants to deliver effective, efficient and equitable programs and services, so we’re working to ensure they have the right tools for the job and know how to use them — or not — in the right way.

READ MORE: Here is a guide to AI governance for state and local agencies.

STATETECH: The biggest challenge with AI may not be the technology but organizational change. How are you helping employees build confidence and literacy around AI?

COLE: We consult with agencies on how to address all kinds of opportunities they see in their work, and when we find a pattern that works, we share it with the next agency. Knowing how peers navigate similar situations can help build confidence. Being a trusted partner that has experience working with emerging technology also helps our colleagues in other agencies separate signal from noise. We want to create an environment for thoughtful experimentation with appropriate guardrails and constraints.

STATETECH: As states move from experimentation to production, what metrics matter most? How do you know an AI initiative is actually delivering value?

COLE: Metrics should align with our core values and goals. For instance, working with the New Jersey Department of Labor, we use generative AI prompts to speed up the rewriting of transactional email templates for unemployment insurance communications, and this work led to recipients responding to notices 30% more quickly, which gets claims processed and benefits issued faster.

In the EDA example, reviewing documents fell from weeks to minutes. That allows businesses to grow and create jobs more quickly.

For call centers we manage, we use AI to analyze transcripts and create better menus. In one large call center, this process has helped agents handle 15% more callers.

When surveyed, users of the NJ AI Assistant estimate it saves them hours of time each week, which frees up time to serve more residents.

STATETECH: Looking ahead three to five years, where do you think AI will have the greatest impact on state government operations and constituent services?

COLE: First, the government collectively needs to ensure it’s protecting the interests of residents — as individuals, consumers and communities — as more powerful technology develops.

Second, we want to ensure the state is supporting economic growth around new technologies, enabling everything from innovative startups to high-value researchers to thrive. And within government operations, we want to ensure every interaction with state government services is accurate and trustworthy, and delivers positive outcomes for residents.

As we develop more specific and refined uses for AI, we can scale our work to be more effective and have a greater impact while holding our values at the forefront.