Who should teach society how to use AI? - WRAL

On June 30, the North Carolina Department of Information Technology published a statewide AI Strategic Roadmap, authored by the NC AI Leadership Council, which had been assembled by the governor’s office. Like many strategic plans, it paints an ambitious picture of where the state hopes to go in the coming years. It speaks to responsible AI, workforce preparedness, government modernization, education, privacy, and public trust.

These are all worthy goals, and perhaps more importantly, they acknowledge that artificial intelligence has moved beyond being solely a technology concern. It is rapidly becoming an economic development concern, a workforce concern, and increasingly, a civic concern. As I read through the roadmap, however, I found myself less interested in what was included than what was absent.

Other WRAL Top Stories

The document spends considerable time describing what state agencies, educational institutions, employers, and communities should do to prepare for an AI-enabled future. Yet it says relatively little about the organizations creating the technology itself. That omission led me to a much broader question, one that extends far beyond North Carolina.

Who, exactly, is responsible for teaching society how to use artificial intelligence?

That question struck me because just a day earlier, on June 29, California announced a statewide partnership with Anthropic, the company behind Claude. The agreement was notable not because California selected one AI provider over another, but because the arrangement reportedly extends beyond discounted software licenses. It includes training, technical assistance, and implementation support designed to help public employees learn how to incorporate AI into their daily work.

In other words, California did not simply purchase software. It purchased capability.

This is not a subtle distinction. It is a significant reframing of the relationship between a technology provider and government customer. I suspect this will become one of the defining questions of the AI era. For decades, we have thought about software as a product. A company builds an application, sells a license, and perhaps provides documentation or technical support. Beyond that, success largely depends on the customer. If the organization fails to transform itself with the software product, that is viewed as an implementation problem rather than a vendor problem.

For example, when a county purchased Microsoft Office 20 years ago, it was buying software. Employees might attend a class or watch a tutorial, but no one expected Microsoft to partner with the county to rethink permitting, budgeting, emergency management, or citizen services. The value of the software came from the features inside the application.

Enterprise software has traditionally been additive. Governments purchased databases to manage records, spreadsheets to analyze budgets, and word processors to write reports. Each application made an existing task somewhat faster or easier, but it rarely changed the underlying nature of the work.

Artificial intelligence is fundamentally different.

Unlike traditional software, AI is not simply another application to install. It changes how work is performed, how decisions are made, how information flows, and increasingly, how organizations themselves are structured. Buying access to a large language model is becoming less like purchasing enterprise software and more like connecting to a new form of infrastructure. And infrastructure has always carried broader societal expectations.

When roads were built across America, we did not simply hand citizens the keys to a car and wish them luck. We developed driver's education, licensing systems, traffic laws, safety standards, and public investment in teaching people how to operate safely within the new transportation network. The government at all levels takes responsibility for the effectiveness and safety of the transportation system.

Healthcare, another sector rooted in the public interest, follows a different model. Much of the responsibility for safe adoption rests with private industry. Pharmaceutical companies conduct clinical trials, produce extensive documentation, educate physicians, monitor adverse events, and continue supporting products long after they enter the market. The government establishes the regulatory framework, but the private sector bears substantial responsibility for ensuring that complex technologies are used appropriately.

Consumer-facing industries often take another path, placing considerably more responsibility on the individual user to achieve their own success. IKEA and Lego expect you to “do it yourself” with enclosed assembly instructions. KitchenAid doesn't provide a culinary instructor with every stand mixer, nor does Weber send a pitmaster home with every new grill. DeWalt doesn’t send a carpenter to your home after you purchase a new table saw. Apple and Samsung don’t spend resources to teach you how to get the most from your smartphone.

Consumer companies provide manuals, safety information, and customer support, but the responsibility for achieving success lies with the owner. And so does the liability of product misuse.

In most cases, a buyer beware strategy works well enough for society. After all, toys, appliances, tools and furniture are not products that often intersect with the “public interest”. Platform technologies like the smartphone sit in more of a gray area. Neither phone manufacturers or software app providers that leverage them have been required to provide broad public education on the risks of surreptitious private data collection, disinformation, distracted driving and phishing attacks, even though all of these are phone-related challenges to society.

It is worthy to debate when product manufacturers should have more accountability than they currently carry. The firearms industry is another illustrative example. Manufacturers have not been required to provide broad public education for gun use, analogous to how we provide driver training through public schools. Nor do they conduct certification or licensing analogous to the DMV. But misuse of guns has created all manner of challenges for society to manage and liabilities to adjudicate. Public safety is foundational to the public interest.

Private industry, government and consumer responsibility are three different approaches to answer the same question: when a technology has profound societal consequences, who is responsible for helping people use it well? Artificial intelligence has not answered that question.

Most discussions about responsible AI understandably focus on privacy, bias, copyright, cybersecurity, or model safety. These are important conversations. But they all assume that organizations already know how to adopt AI responsibly in the first place. I would argue that most do not. And this is particularly true for local governments.

It is easy to imagine large state agencies building internal AI teams, hiring specialists, and developing governance frameworks. It is much harder to imagine a rural county with a small IT budget doing the same. Many municipalities across North Carolina struggle to recruit technology staff, if they even have the budget for it. Few will have prompt engineers, AI governance specialists, or dedicated implementation teams anytime soon. Yet, these may be the very communities positioned to benefit the most.

Artificial intelligence offers the possibility of automating administrative work, simplifying permitting, assisting with grant writing, improving citizen communications, accelerating planning, modernizing procurement, and making government services more accessible without dramatically increasing headcount. For smaller communities that have long operated with limited resources, these productivity gains could be transformational. But only if someone teaches them how.

For years, policymakers focused on closing the digital divide by expanding broadband. That investment was necessary and overdue. Yet as connectivity becomes more widespread, another divide is emerging that has little to do with fiber optic cables or wireless coverage.

The next divide may be competency.

A county may have broadband, cloud infrastructure, affordable AI subscriptions, and modern computers, yet still find itself left behind because no one has developed the institutional knowledge needed to redesign government around these new capabilities. That is not merely a technology challenge. It is a workforce challenge. It is also an economic development challenge.

North Carolina has invested heavily in attracting data centers, semiconductor manufacturing, biotechnology, and advanced industries. Those investments create jobs, tax revenue, and long-term economic opportunity. But if we stop thinking about infrastructure once the buildings are constructed, we risk repeating a mistake that should already feel familiar.

For decades, we relied primarily on the private sector to expand broadband access. The economics worked well in dense population centers. They have worked far less effectively in many rural communities. As a result, portions of our state spent years on the wrong side of the digital economy (and some still are), not because the technology lacked value, but because the incentives did not always align with universal access.

I have previously referred to these regions as the "digital rust belt." The economy has largely failed in these regions because requisite technology infrastructure doesn’t yet exist, or is underperforming. And even where connectivity is possible, there have been insufficient programs to train the workforce on how to best leverage it for e-commerce, remote education, telehealth and other benefits.

Artificial intelligence presents an opportunity to avoid deepening the digital rust belt. But this requires us to rethink what we expect from the companies building foundational AI models. If these organizations increasingly describe their technologies as general-purpose infrastructure capable of transforming every industry, I argue that they should be evaluated not only by benchmark scores, reasoning capabilities, or subscription pricing, but also by how effectively they enable society to use their technology responsibly.

Imagine if partnerships like California's became the norm rather than the exception. What if every statewide AI deployment included structured workforce development? What if enterprise AI contracts bundled implementation coaching, ethics training, governance templates, and continuing education alongside the software itself? What if AI companies competed not only on model performance, but on the success of the communities they helped modernize? That would represent a very different relationship between technology companies and society.

The government would still have an essential role. Public education, standards, procurement policy, and accountability remain squarely within the public sector's responsibility. But expecting AI providers to participate actively in building an AI-capable workforce does not seem unreasonable. In fact, it may become one of the defining competitive advantages of the next decade.

North Carolina's roadmap establishes an important vision. It identifies many of the right challenges and sets a direction that deserves broad support. The harder work now begins.

As the state moves from strategy to implementation, the next question should not simply be how the government adopts artificial intelligence, but how government and industry together create an AI-literate society, beginning with an AI-competent public sector itself.

We should not make the same mistake with AI that we made with broadband. Building the infrastructure is only half the job. Building the human capability to use it is the other half. Whether that responsibility ultimately falls on the government, industry, individuals or some combination, we should answer that question deliberately, not accidentally.

We must get this right. I believe it is just as important to train people on how to use AI in ways that protect their own data, privacy and digital security as we might consider it important to learn to safely drive a car. Or to take medication. Or to safely handle a gun.

In the end, the future of artificial intelligence will not be determined solely by the quality of the models we build. It will be determined by the capability of the people who use them. If AI is destined to become infrastructure on the scale of electricity, highways, healthcare or the internet, then teaching society how to use it safely and effectively is not an optional feature. It is part of the infrastructure itself.