Usually when we talk about the intersection of Driving and in-car technology, safety is at the center of the discussion. We see that mostly in sensors and cameras that help drivers to back up, stay in lane, and stay aware of what’s in their blind spots, but we also hear about safety when manufacturers talk about voice-controlled in-car infotainment systems.

The discussion always ends the same way. In the end, Autonomous Driving arrives as the ultimate safety feature, removing the most unsafe part of driving — the often distracted driver. Fully autonomous testing thus far has indicated that they should prove far safer than human drivers, but the AI, infrastructure, and regulatory framework to make autonomous driving a reality still aren’t in place. That hasn’t stopped car companies from slowly adding bits and pieces of the automotive future into the present — something that a AAA study released today indicates is causing problems.

The AAA Foundation for Traffic Safety published a study run by University of Utah researchers on how distracting in-car infotainment systems are. While the study doesn’t exactly reflect how manufacturers intend for these features to be used in the car, it does uncover some surprises about features that are generally considered safe to use for drivers — most conspicuously including voice control.

The study used 30 different 2017 models from a number of manufacturers, with 24 different drivers trying out each one. The study had each driver of each car drive unimpeded, perform a task while keeping their eyes on the road, and perform a task that required visual distraction. On top of that, the study individually tested interactions with controls on the center console (sometimes including physical controls like knobs to control the infotainment system), the center stack (the touchscreen itself), and voice commands, each of which were applied to navigation, calling, texting, and controlling music and entertainment. Using a number of different measurements, the researchers used cameras and lights to determine how distracted drivers became during each task.

Those results weren’t great. You can pore over the full report here, but the short of it is that while there was variance from car to car, no car and no control system was rated as having low cognitive demand — all 30 cars were rated as having either moderate, high, or very high demand. The real eye-opener was that voice controls, thought to be the safe way to interact with the connected car, were found to be the second-most distracting control method, far more so than the center stack and its touchscreen. That seems counter-intuitive, but the results are partially explained by voice exchanges taking the longest of any control method. Navigation proved to be the most distracting task, with texting in second and calling and entertainment controls about even as the least distracting of the tasks.

Nothing in the study should come as a surprise, save for voice controls being more distracting than touchscreen controls. Human brains can only really focus on one thing at a time — if you introduce a secondary task to driving, no matter what it is, it’s going to make driving less safe.

What’s really interesting here is this middle ground we’ve found ourselves in. If autonomous driving is the goal, then it follows that cars should become more and more similar to computers — that includes all of these navigation, communication, and entertainment features. If we don’t have to drive, transit time is freed up to do things more productive or fun! The problem is that we’ve seen automotive manufacturers include these forward-looking features in today’s cars while humans are still driving them. That’s proving to be dangerous.

To an extent, the industry is aware. It’s anecdotal evidence, but I’ve never talked to anyone in the automotive industry who has ever suggested that touchscreens were safe to use while driving — they’ve all said those controls should be used while the car is stopped or by a passenger (admittedly, most have suggested voice controls as the safe alternative, something that has now been cast into doubt). Thing is, automakers can’t control how drivers use their vehicles. We’ve seen with smartphones and now infotainment systems that many drivers, when given the opportunity to drive while distracted, do so regularly!

It’s not just a problem with infotainment systems. The most famous case of a car company going too far, too fast would be Tesla. Just last month they were deemed partially responsible for the death of a Tesla Model S driver who used Tesla’s semi-autonomous drive system, Autopilot, beyond what it was designed for. While the National Transportation Safety Board agreed that the driver acted irresponsibly by treating Autopilot as a fully autonomous system and not keeping his hands on the wheel, they concluded that Tesla had provided no real checks to prevent drivers from misusing their technology.

Tesla Autopilot in its past iteration (the current version of Autopilot shuts off when it detects that the driver has removed their hands from the wheel) was the perfect example of the pitfalls of the transition between human driving and autonomous driving. By introducing technologies of the future alongside the imperfections of human drivers, automakers run the risk of creating a less safe driving environment today. Even if that’s in the service of developing the safest possible drive systems, that’s an ethically shaky place to be.

Some companies have been more careful about how they manage this transition. Nissan and Volvo (as two examples) have both introduced similar semi-autonomous systems, but their systems amount to advanced cruise control — they can maintain speed, slow down in traffic, stay in lane, and not much else. These systems still use the same kind of sensor-and-algorithm combinations that will eventually power fully autonomous driving, but they’re being implemented in a way that prevents drivers from mistaking them for anything resembling fully autonomous driving. They even make it clear in the names — Nissan’s ProPILOT Assist and Volvo’s Pilot Assist II (ironically, both companies have entries on the list of cars with infotainment systems that cause very high levels of distraction).

But, when you zoom out, the whole industry process starts to look a bit like the iterate-fix-iterate-fix cycle that Silicon Valley has found so endearing for years. We can debate the merits of moving fast and breaking things in the world of consumer electronics, but applied to the life-and-death business of cars, it looks like an untenable position.