In February 2017, Bill Gates made headlines by proposing a new "robot tax" to ameliorate the effects of automation on workers. At the time, this proposal generated much controversy but little substantive policy. Since then, tax authorities across the globe have been considering proposals for a robot tax, in some form or another.
In August, South Korea announced that it will reduce tax incentives for industry automation equipment, in what may be a precursor for a direct robot tax. Earlier in the year, the European Parliament considered a proposal that would classify robots as “electronic persons” on whose behalf employers would have to pay taxes. That proposal was rejected, for now, but could be resurrected at any time.
The proposal has gained traction in the United States as well. A San Francisco county supervisor is leading a statewide effort in California to impose a tax on businesses who install “robots, algorithms, or other forms of automation” that displace workers.
These proposals stem from legitimate concerns about automation squeezing human workers out of their livelihoods. On the other hand, there are strong policy objections to a tax that disincentivizes innovation in a cutting-edge industry. There are also competitive concerns if the United States hinders automation while other countries reap the benefits of using robots to produce goods and services.
Putting aside the policy implications, a robot tax raises basic tax administration questions that bear examination. To begin with, it is surprisingly difficult to define what a “robot” is. Legislatures and courts have struggled when faced with formulating or applying any definition.
Is a robot specifically a programmable machine involved in manufacturing? Does it include algorithms run on computers? Does it include any technology that displaces a human worker, which would include ATM machines (bank tellers), word processing software (typists) and other common modern conveniences?
As Professor Ryan Calo recently wrote, “jurists on the whole possess poor, increasingly outdated views about robots and hence will not be well positioned to address the novel challenges they continue to pose.” There is no reason to believe the IRS, or state tax authorities, will have any more success.
A close analogy is the distinction between an employee and an independent contractor for employment tax purposes. In theory, this issue has arisen so often, and for so many years, that there should be simple, straightforward classification rules. In reality, this area is riven with controversy and ambiguity.
There is enough trouble sorting out whether a human worker is an employee or an independent contractor. It is not difficult to imagine the controversy and ambiguity that would result from trying to sort out whether a particular machine or algorithm is a taxable robot.
Even assuming a workable definition of a taxable robot can be reached, who would bear the tax? Robots can't actually pay taxes. The robots' owners would be the ones paying the tax, and they would likely pass along the costs to consumers. Or they would just move the robots to a friendlier jurisdiction — robots don't complain about relocating.
Finally, what would the tax be imposed on? Robots don't earn wages, so the base for a payroll tax cannot be used. A flat per-robot excise tax sounds simple, but would be regressive and burdensome for small businesses. There are several other possibilities, but each poses its own difficulties from a tax administration perspective.
One proposal is a tax based on the number (or salary equivalent) of human workers displaced by the robot. The amount of payroll tax that the displaced workers would have earned is assessed as an excise tax against the business. This comes closer to approximating the harm that is intended to be addressed.
It is not clear how the number of displaced workers per robot would be calculated. The most straightforward fact pattern would involve a business that employed five workers on an assembly line, then places a robot into service on the line and fires all five workers. The business would be assessed an excise tax equal to the amount of payroll taxes for the five displaced workers.
Not every fact pattern will be so simple, however. Suppose a business wants to begin creating a new product. Instead of hiring workers, it places robots into service. No workers are laid off. What is the correct amount of tax? How many workers would have been employed but for the robots? How would you calculate the amount those hypothetical workers would have paid in payroll taxes? This is precisely the sort of facts-and-circumstances issue that consumes a tax authority’s resources and invites controversy.
An alternative would be to impose a tax based on the amount of income attributable to the use of automation. This only exacerbates the problems with the first proposal. Suppose our business fires three of the five employees, and places a robot into service to replace the other two. What percentage of the income is attributable to the robot? The problem is even more complicated for large enterprises, for whom allocating income between human and robotic workers would be a compliance morass.
A more modest proposal would follow the example of South Korea and limit or eliminate tax incentives related to automation. Unlike South Korea, there is no specific tax incentive in the United States for automation. A possible alternative would be to limit or eliminate depreciation or business expense deductions for robots. This is more difficult than it sounds, because of the difficulty of defining what should be covered by such a "robot exception."
An expansive definition could cover many necessary business expenses and capital investments that have nothing to do with displacing workers. A more tailored exception, for instance one somehow linked to the number of workers actually displaced, runs into the same administrative problems as the other proposals.
Even the potential for robot taxes complicates planning for businesses that rely on automation. A taxpayer considering entering or expanding in a jurisdiction that is contemplating such a tax must add to its other concerns whether there might be a "robot tax" enacted in the future.
This uncertainty creates opportunities for tax competition between jurisdictions. For example, if one state considers a tax on automation, a neighboring jurisdiction could expressly reject doing so in the expectation of inducing businesses to move. In that case, the state would have neither the jobs, nor the revenue from its contemplated robot tax.
Perhaps these problems can be resolved eventually. If other policy alternatives fail, and the worst fears of job displacement from automation come to pass, popular support for a robot tax in some form could become widespread. At present, however, it is likely that the haphazard imposition of robot taxes would do far more harm than good.
When the internet was young, and governments began eyeing it as a new source of revenue, Congress stepped in to block internet taxes and protect the fledgling sector. If the robot tax proposals now being discussed at the state level become a reality, Congress could intervene in the same fashion. At any rate, the robot tax poses serious administrability issues that should be addressed before any such tax is seriously considered.