Why government is the key to unlocking AI’s true potential

With artificial intelligence (AI) conquering realms previously thought exclusive to the human mind, like visual arts and language, it may seem like the promise of machines solving our greatest challenges are right around the corner. 

Unfortunately, AI alone is not going to improve our lives.

It’s certainly impressive that ChatGPT, the most advanced chatbot to date, can churn out mystery novels, write code and pass MBA exams, while Dall-E can draw anything you ask of it. But these are parlor tricks compared to the hard work of addressing our thorniest problems like poverty and inequality.

AI’s true potential has remained elusive because those charged with tackling these issues — governments — have long struggled to leverage technology.

Governments possess the key building blocks to unleash AI’s potential — data — but first, governments must learn how to walk with tech before running.

I led two state labor departments, one during the COVID-19 pandemic, so I saw first-hand how some of our most critical services like public workforce and unemployment programs remain rooted in Great Depression-era practices and powered by yesterday’s technology. But I also saw how technology can and cannot help. 

So, why is government so rarely successful at leveraging technology and what can be done?

Government needs to shift its mindset. Technology is not a standalone solution. It is deeply integrated with processes. So technological solutions must couple with changes in operations — or they fail.

Our unemployment insurance systems are a case study. At the onset of the pandemic, millions of workers were suddenly without jobs and applying for unemployment insurance in record numbers. States did their best to process claims quickly, but long delays were the norm.

Unemployment insurance problems were not due only to old technology, so the fix isn’t just new technology. Proof of this is that no state unemployment insurance program performed well during the pandemic — even those with “modernized” systems on which millions of dollars were spent.

The unemployment insurance systems I’m familiar with are powered by refrigerator-sized mainframes from the 80s that use a programming language few still know. Nevertheless, during the pandemic my team was the first to accept Pandemic Unemployment Assistance claims and second fastest to pay them. 

We were successful because we focused on applications of technology that would complement the process reforms we were making at the same time. Process needs better tech, but process must also change to leverage better tech.

Our nation has been administering unemployment insurance the same way since it was created in the 1930s: claimants provide information about their employment situation; staff manually review each claim, while also manning the call center and answering questions. 

This process is not scalable. During the pandemic to keep our heads above water, we made every operational fix possible to free up staff and streamline the process. Our motto was, “let tech handle easy cases so people can help people.”

While contemporary technology is necessary, it is not sufficient. The latest technology in the world is only as good as its underlying bureaucracy.

Another impediment to government innovation is a procurement process primed to buy easy off-the-shelf tech solutions. Unfortunately, no off-the-shelf fixes exist to do the hard work of actually improving service delivery.

This backward process means government buys a solution but does not work on the problem. That’s why after spending millions of taxpayer dollars, technology rarely drives public outcomes.

Successful projects in both the public and private sector, use an iterative, resident-first approach. That means developers work closely with residents and their state partners to understand problems, test solutions, and continue to refine them.

The good news is that this process becomes easier with contemporary technology because it also frees the data that fuels a virtuous cycle of continuous improvement that transforms service — and this is the key to unlocking the true power of AI.

For instance, states have decades of quarterly wage records. By using machine learning and AI to mine this trove of data, we can transform workforce training programs, which is what I’ve been encouraging states to do as CEO of RIPL, a non-profit civic-tech org that works with governments to use data better.

In Maryland and Hawaii, they now have a digital tool that analyzes an unemployed worker’s resume and provides data-backed recommendations for training opportunities, job openings and new careers proven to lead to higher wages. This is made possible by machine learning working on tens-of-millions of state wage records to understand which careers similar job seekers have switched to, stuck with and earned more in. 

Massive amounts of data are what enable machine learning to do such impressive things. And there’s no shortage of data that only governments have access to that can be used to improve lives — all it takes is the hard work needed to approach technology differently.

Scott Jensen is the former Rhode Island director of Labor and current CEO of RIPL, a nonprofit tech-for-social-impact organization that works with governments to use data, science and technology to improve policy and lives.