AI tax targets the wrong signal

As Connecticut lawmakers debate Senate Bill 515, they are asking a question more states will soon face: As artificial intelligence changes work, what happens to workers whose jobs change or disappear?

The bill would create a “workforce and productivity gap” surcharge. If a company’s payroll falls while each remaining worker appears to produce more, the state could impose a new tax. Companies that keep staffing steady and use “collaborative technology” meant to help workers rather than replace them would be exempt.

That sounds like a way to protect workers. But it would likely do more harm than good for Connecticut workers over time. The bill directs the state to design a surcharge around a “productivity gap,” meaning firms could be taxed if they appear more productive while employing fewer workers.

The bill does not specify how this would be measured, but in practice such approaches rely on some version of falling payroll alongside steady or rising revenue. Those figures move for many reasons other than technology replacing workers, including reorganization, shifts in product mix, or higher prices. The result is a policy that risks taxing adjustment rather than harm, discouraging investment and slowing the wage growth that usually comes with a more productive economy.

A drop in payroll does not have one meaning. It can reflect weaker demand for a company’s product, a shift toward different lines of business, work moving to contractors or other locations, or better internal organization. In all of those cases, measured output per worker can rise. The bill treats those very different situations as if they were the same.

Technology usually reshapes tasks before it eliminates entire jobs. Despite common claims, firms do not typically replace an occupation all at once. They change pieces of work. Some tasks disappear, some become more valuable, and new ones emerge. The real question is not whether a firm has fewer workers than it did three years ago. It is whether workers are moving into more valuable roles as the work itself changes.

The bill tries to account for that by rewarding “collaborative technology,” meaning technology that helps workers do their jobs instead of simply replacing them. The instinct is sound. The trouble is that the line is hard to observe from the outside. The same software can reduce the need for some roles while raising the value of others. A payroll statistic cannot tell you which is happening.

What the policy can do is change behavior. If firms need to stay close to an old staffing baseline to avoid a surcharge, many will manage to that line. They may delay layoffs, change hiring plans, or move work outside payroll. That may make the numbers look better without leaving workers better off.

There is a second problem, and it goes straight to workers’ pocketbooks. When firms face unclear rules around new technology, they hold back on investment. That means less experimentation, slower adoption, and weaker productivity growth. Over time, weaker productivity growth usually means weaker wage growth.

There is also a broader economic point. When technology makes a service cheaper, people often use more of it. Economists call this Jevons paradox. Lower-cost AI legal research, for example, can reduce the time spent on each task while expanding demand for legal services overall. A tax built around preserving current payroll pushes in the opposite direction. It nudges firms to hold onto existing roles instead of helping workers move into higher-value work, often in different roles, firms, or sectors.

A functioning labor market depends on exactly this kind of movement. Workers change jobs. Firms expand and contract. New tasks show up in places the old jobs did not. A policy that tries to hold the labor market still ends up reducing opportunity.

None of this diminishes the concern that motivates the proposal. If the goal is to support workers through these changes, the most direct tools remain the most reliable. Strengthening training and career mobility does more than penalizing firms based on imperfect signals. It helps workers move into the new roles that technology creates. Policy can also do more to align incentives. Today, it is often easier to expense investment in machines than investment in people. Allowing immediate expensing of employer-provided training would put worker skill development on similar footing. Connecticut largely follows federal rules here, so progress would require action at both levels.

Connecticut’s proposal recognizes a real challenge. But treating lower payroll as clear evidence of harmful displacement is too blunt for a dynamic economy. The goal should not be to freeze today’s jobs in place. It should be to help workers move into better ones over time.

Revana Sharfuddin is a research fellow at the Mercatus Center’s Labor Policy Project at George Mason University.