Breaking AI’s Black Box

Workers must have a hand in implementing new technology.

There is a temptation to treat artificial intelligence policy as something distant—white papers, safety labs, summit communiqués. The real frontier is closer to home. It is in the quarterly calls where company after company, to Wall Street’s cheers, announces that AI will let it shrink its workforce. It is in a warehouse in Memphis where the algorithm has decided someone is moving too slowly, and in a delivery van in Newark where the screen tells a driver they have been deactivated and offers no human to call.

Technology has reshaped work since the invention of the wheel, and every generation of American workers has lived through volatile transitions—mechanization, containerization, the personal computer, the internet. But AI is different, for three important reasons.

The first is speed. Systems that were research curiosities three years ago now write performance reviews, screen resumes, and write complex code. The second is scope. AI does not target one industry the way the shipping containerization targeted longshoring—it touches white-collar and blue-collar, professional and service, full-time and gig work en masse. The third is opacity. The systems now deciding whether a worker is hired, scheduled, paid, surveilled, or fired are genuinely indecipherable, even to the employer who deploys them. That combination is both unprecedented and unsettling.

At its core, AI is a neutral tool, which can either be good or bad for workers depending on how it’s deployed and for what purpose. Whether workers have genuine influence in the process will be a huge determinant of how this plays out.

Our focus is on what AI is doing, and about to do, to the working lives of ordinary Americans. Our goal is to ensure that when AI arrives in a cubicle, warehouse, hospital, or delivery van, the workers there have a meaningful say in how it lands. With proper input, worker organizations can make sure that the AI rollout helps rather than hinders the workforce. Technology can be used to improve worker productivity and job quality rather than just raise profits, a fact developers should keep central to their thinking as they create the tools and applications of the future.

But in a country where unionization is minimal and the tradition of works councils shallow, we must build an American response quickly, because AI is arriving now. There is little historical precedent for our federal government moving to protect workers in such a rapid transition. However, the size of the challenge is without precedent as well, so if we want AI to meet the promises of its promoters, workers must be heard and their concerns must be addressed.

A winning approach begins where the impacts for workers are most direct: their workplace.

Rights and Engagement at the Largest Employers

Our largest employers, where AI deployment is likely to be fastest, should immediately begin engaging with their workers while a broader national framework is formed.

Specifically, companies with 500 or more employees—a group representing more than half of the U.S. workforce—must give their workers a collective right to advance notice and meaningful consultation prior to the deployment of any AI system that affects their wages, hours, and working conditions.

Employers should be required to disclose the system’s purpose and expected impact, engage in substantive good-faith discussions, and issue a written final decision that fully considers their workers’ concerns. As the system’s effects become clearer, employers need to provide updates and revisit the conversation. AI implementations need to wait until after these consultations have been held.

Corporate America already has large organizations like the Business Roundtable and the Chief Human Resources Officer Association through which they coordinate their actions. There’s nothing to stop all Business Roundtable members from committing to this process and coordinating their resources. While it’s fair for any one employer to object on the grounds that they’d be left behind their competitors, that concern would be nullified if they all did so together.

Where a union or worker organization is present, it will serve as the democratic representative for affected employees, consistent with its traditional role in workplace matters.

In workplaces without existing representation, an independent, employer-free process should be put in place to elect non-management representatives. This process must secure a minimum of five representatives (more for larger employers), granting them reasonable time off, necessary resources, and full access to the pertinent information needed to fulfill and improve the execution of their duties.

To help these bodies, relevant local, state, federal, and private entities—such as the Bureau of Mediation, the American Arbitration Association (AAA), and Workforce Investment Boards—may develop model templates for institutional and workplace use. Smaller workplaces should be required to adopt similar processes over time.

Individual Workers’ Rights

All workers—regardless of union status, employer size, or whether they are employees or independent contractors—should have basic rights concerning AI. They should receive a plain-language explanation of the technology’s role and have the right to challenge AI-generated outcomes before a human reviewer with authority to override the system. States including New York, Illinois, Colorado, and California have adopted pieces of this approach; Congress should establish a clear federal baseline.

Every worker also needs basic AI literacy. Employers—or, where they do not, the federal government—should provide free training on how AI systems work, how they are used in the workplace, and how workers can use them to build new skills and opportunities that will make them more productive and better paid. The program should be available to all workers, and for large employers should be delivered on paid time within 180 days of implementation. A workforce that understands AI will be better equipped to adapt to technological change and take part meaningfully in decisions about its use.

These proposals are not a substitute for collective bargaining where workers want it. They are a framework for the vast majority of American workers who lack a union to have a voice in the most consequential workplace transformation in a generation—and for those who do have a union to bargain over AI deployment and algorithmic management, as the law plainly requires.

Empirically Grounded Sectoral Collaboration

The public conversation about AI and the future of work is more often than not based on guesswork. Workers cannot predict when or how their roles will change; employers struggle to forecast workforce needs; policymakers are left guessing whether to expand the social safety net, a training pipeline, or both. Crucially, workers—who are best positioned to see AI’s incremental arrival—are rarely consulted.

To ground this discussion, the federal government should mandate a collaboration of government, employers, and workers across the nation’s largest employment sectors. A joint approach ensures credibility, as together they can publish what the country lacks: an honest, shared map of where work is going.

These sectoral bodies will carry out two essential tasks. The first is transparent forecasting which will map out how AI is reshaping work today and how it will do so in the future. The second is the establishment of minimum standards for critical workplace practices. For example, they can require that AI-generated patient acuity scores be explained to health care staff or require human-accessible audit trails for AI used in warehouse discipline.

A federal sectoral framework would inform workplace-level consultations, rather than replacing them. It could provide enterprises and employees with solid, empirical data and recommended standards that they can adopt, adapt, or exceed. Crucially, these national standards must serve as floors; individual states and employers would be free to enact higher standards.

Resources

AI deployment that leads to higher productivity and wages can become a win for everyone: Higher wages lead to higher demand, which leads to rising output and more jobs. But where it doesn’t, workers need protections.

Workers whose jobs are redesigned or eliminated need more than a process. Funding must be addressed separately because it is the hardest—and most often avoided—question.

Two needs are central. First, workers need portable benefits and access to retraining so they can move into new opportunities. Second, they need income support—through enhanced unemployment benefits or direct cash assistance—to help them retrain, start a business, move, or support their families during the transition.

How to pay for this remains an open question, but there is growing agreement across ideological lines that AI’s gains should be broadly shared—including by those who will be using its advances in the workforce and anyone displaced in the process. Proposals include a “robot tax” on automation, public wealth funds, universal income payments, and a small tax on AI usage. The mechanism is debatable; the principle is not. A transition of this scale cannot be financed out of workers’ own pockets.

The question is not whether AI will reshape American work. It already has. The question is whether American workers will have a meaningful role in shaping the systems that now shape them. Employers must prioritize long-term growth that benefits everyone over extractive short-term gains, and they must include workers in that process.