If Domino's has its druthers, deliverymen would soon be going the way of the dodo bird.

The pizza conglomerate is just the latest major business to mull incorporating driverless cars in their business model. But while technology enthusiasts cheer on the rise of autonomous vehicles, billions in costs and safety pitfalls are being swept under the rug. The infrastructural changes necessary to make driverless cars a reality will set taxpayers back to the tune of billions, while opening the door to unforeseen safety consequences.

Permissionless innovation is, in general, a welcome approach to new technologies. For too long, bureaucrats have erred on the side of risk-aversion, resulting in life-saving technologies never reaching customers. But technological changes that necessitate infrastructural changes are a bit less straightforward, owing to current governmental control of roads and bridges. If new types of vehicles come along that require these alterations, lawmakers and bureaucrats must decide whether or not taxpayers will be on the line to facilitate such innovations.

And to facilitate the rise of driverless cars, taxpayers would inevitably be on the hook for billions in new spending. Just one example: The color of traffic lights is not detectable to vehicle cameras when the sun is immediately behind the light. Governments' interface for alerting cars to proceed cautiously in construction zones is designed for humans, and would need to be revamped in order to convey the same information to robots. Local revenue collection would likely take a huge hit, as municipalities will no longer be able to rely on reckless drivers to fill their coffers.

These adaptation costs would likely add up quickly, even if we were assured that driverless cars would eventually be able to improve sufficiently to deal with these problems.

And such accommodations may be worthwhile for a truly game-changing technology. But the "driverless cars" frequently touted by pundits are simply not there yet. For starters, there is no real evidence that the vehicles being tested are even a safety improvement over the current fleet. To gauge the safety of these vehicles, it helps to first differentiate the levels of automation currently in play. Cutting-edge vehicle technologies like the Tesla Autopilot are "level 2," meaning that the human driver retains controls even as some features are automated. "Level 3" vehicles, the variety currently being tested by Google, are largely autonomous but require human intervention at critical junctures.

Since level 2 vehicles have the most real-world experience and provide a bridge to the autonomous world, examining their safety is key. Tesla received a public relations boost earlier this year when the National Highway Transportation Safety Administration found that the Autopilot feature lowered crashes by nearly forty percent.

While such a large finding should have established the credibility of quasi-autonomous vehicle features, flaws in the study cast doubt on government claims. The report failed to take into account the introduction of automatic braking several months before the introduction of autopilot. And a 2016 study from the Insurance Institute for Highway Safety examining the impact of automatic brake features found that a 39 percent reduction of rear-end crash rates based on that feature alone. If IIHS findings are to be believed, nearly the entire decline in Tesla crash rates can be attributed to automatic braking, instead of the automatic steering touted as a prelude to full autonomy.

It's all too easy for policymakers and business leaders to fall for the allure of new technologies. The introduction of self-checkout machines and autonomous underwater vehicles had stakeholders expecting the full replacement of man for machine. But as Massachusetts Institute of Technology professor David Mindell points out, practitioners of autonomy recognize the need for human involvement as new technologies are rolled out. And if, as polling suggests, humans are uncomfortable with the idea of an automated system calling the shots, they may be overtly inclined to override the system. Mistakes in the other direction are just as likely. If drivers are too comfortable with their car's operating system and become inattentive to conditions on the road, they won't be able to take the helm when their vehicle demands human input.

Even if driverless cars make roads safer in some ways, these potential hazards make it difficult to see the net benefit of widespread adaption. Companies like Domino's are right in trying to add value to their operations, and make delivery a smoother process overall. But unforeseen costs and pitfalls must be addressed in order to ensure that taxpayers aren't bilked for overhyped technologies.

Ross Marchand is a policy analyst for the Taxpayers Protection Alliance.

If you would like to write an op-ed for the Washington Examiner, please read our guidelines on submissions.