Ask the automakers and tech companies trying to build cars that drive themselves to defend their work, and they turn to two key arguments: Autonomous cars will save lives, and, by eliminating the need for a human driver, they’ll open the car to new uses and users.

Less frequently invoked, but equally integral to that vision of a safer, more comfortable world, is efficiency—which is to say, less environmental impact that your old, dumb, gasoline-powered ride. But a report the Center for American Progress released today undermines that assumption. “It could go either way,” says Myriam Alexander-Kearns, one of the authors.

Alexander-Kearns and her co-authors say no one can know what robot cars’ environmental impact will be without first answering three questions: How will automation affect total vehicle miles traveled, how will it impact congestion, and how efficient will the vehicles be?

The problem is that nobody knows how these vehicles will be used, especially in the early years. You could assume that because the tech will likely debut via ride-sharing services, people will need fewer cars, and so cars will drive fewer overall miles. Or, because riding in a car gets way nicer when you don’t have to do anything, vehicle miles might soar.

Smarter cars could drive more closely together, but that will require a critical mass of autonomous cars on the road—a threshold that’s decades away. Shared cars could spend lots of time “deadheading”—driving around empty between pickups, but that might actually improve emissions, since they’d be running constantly, and warm engines are cleaner than cold ones.

“Existing research does not draw clear or consistent conclusions,” says Alexander-Kearns. Kinda the opposite: Some studies show great potential, others predict doom. When the EPA looked at the problem, Christopher Trundler, head of the agency’s Office of Transportation and Air Quality, said the technology could create “utopia or dystopia for the environment.”

The authors of the CAP report call for more rigorous research, more data collection, more sophisticated computer modeling tools that could reveal more about how these cars would interact with each other and their human-driven counterparts in a variety of situations, including different degrees of penetration.

“We shouldn’t be waiting until autonomous vehicles are everywhere to start asking the really important questions,” Alexander-Kearns says.

One part of this equation, however, should be easy to solve: How much each autonomous vehicle pollutes. Provided you’re starting with the right renewable energy sources, if the cars are electric, none of this matters quite as much. These days, just about every future-facing concept combines autonomy with electric power.

But if policy makers don’t maintain an emphasis on fuel efficiency and electrification, if they don’t encourage the infrastructure needed to make both of these technologies feasible, then automakers will run their self-driving cars on dino juice.

“We really do want to emphasize the incredible opportunity this is to work on electrification,” says Miranda Peterson, who co-authored the report. Otherwise, the future may not feel quite as future-y.