The Cars That Talk to Each Other

Self-driving cars aren’t failing because of bad sensors. They’re failing because they don’t talk.

TL;DR: Every autonomous car today is a lone intelligence trying to reconstruct the world from its own cameras and radar. The better idea — a shared protocol, like HTTP for roads, where every car broadcasts its position and intent to every other car — would make sensors cheaper, crashes rarer, and the whole system more robust. The obstacle isn’t the technology. It’s the politics of agreeing on a common language.

The intersection is a four-way with no signal, just painted stop lines that have faded to grey. A white sedan rolls up from the north. A delivery van from the east. A cyclist from the west. In the old world, this is the moment where someone flashes their high-beams, or inches forward, or waves a hand through a windshield. In the new world nothing happens, visibly. The van goes first. The sedan tucks in behind. The cyclist is given a berth wider than any human driver would have offered, because her helmet is broadcasting too, and the cars know exactly where she is. The man in the sedan is watching a cooking video on his phone. He does not look up.

This isn’t 2026. This is somewhere further on. And the thing that changed was not the car.

Today’s self-driving cars are lonely. That’s the part nobody talks about. Each one is a small, anxious mind trying to reconstruct the whole world from its own five senses. Tesla uses eight cameras and a neural net trained on video, and when it works it works beautifully, and when it doesn’t, it sometimes rolls through a red light because a shadow fell the wrong way. Rivian is adding lidar — radar’s better-looking cousin — which helps in snow and at night, but adds thousands of dollars of rotating glass to every hood. Waymo does the whole thing at once, with a sensor stack on the roof that looks like a small satellite, and then charges you by the mile and makes you schedule ahead and won’t let you leave your sunglasses in the glovebox, because it isn’t your car.

Three companies. Three theories. All of them solitary.

The idea I keep coming back to is that we’ve been trying to solve the wrong problem. We are teaching each car, individually, to understand a world it shares with every other car. We’re building a thousand private intelligences where we could be building one public one. The cars don’t need to see each other better. They need to talk to each other.

There is a version of this future where governments do something they almost never do, which is the sensible thing. They sit down — Brussels, Washington, Delhi, Beijing — and agree on a protocol. Not a product, not a platform. A protocol, the way we agreed on HTTP for the web, or GSM for phones. Every autonomous car, regardless of who made it, speaks the same language. Every car broadcasts its position, its intent, its next small decision. Every car listens. The whole road becomes a mesh, end-to-end encrypted, humming below human hearing.

The car behind you wants to change lanes. It doesn’t flash, or nose in, or pray. It says so, in a half-packet of information, and the cars around you make room before you’ve noticed. A truck two blocks ahead hits black ice; the knowledge propagates backward through forty vehicles in under a second, and by the time you reach the spot, your car has already slowed. An ambulance is coming; the whole street parts for it like a zipper opening.

The economics change too. You don’t need lidar on every car, because the car doesn’t need to see the one in front of it — it’s been told where the one in front of it is. The sensor becomes cheap. The car becomes cheap. The thing you own, the one with your gym bag on the passenger seat and your kid’s crayon scribbles on the door panel, can drive itself for the price of a mid-range phone’s worth of silicon.

This is the part I’d like to believe. I am less sure about the rest.

Protocols are slow. We are thirty years into the internet and still arguing about email spam. Any common language for cars will be written by committees that include people who sell lidar, and people whose grandfathers built combustion engines, and people in ministries who don’t want to be blamed for the first spectacular crash. The rollout will be uneven. For a long time, the new cars will have to keep one eye open for the old ones — the human driver in a 2019 sedan who still has to be guessed at, because she isn’t on the network. And someone, somewhere, will figure out how to spoof a packet, and for a week or a month the news will be about a city where the cars went mad. They’ll fix it. They always do. But people will remember.

And then there’s the quieter obstacle. The world is full of internal combustion cars right now, and behind them an industry that has spent a century perfecting them — shaving grams off a piston, tuning an exhaust to pass a new emissions rule. That investment has gravity. It doesn’t disappear because a better idea arrives; it bends the better idea around itself for a few decades.

Still. The kid in the back of the white sedan, eight years old in some year that isn’t this one, will not know any of this. She will not know that her parents grew up in a world where cars didn’t speak to each other, where every intersection was a small negotiation conducted in glances. She will know only that the car stops when it should, and moves when it should, and that the road outside her window is quiet in a way that roads used to not be. She will take it for granted, the way we take electricity for granted, the way her great-grandparents took running water.

That is what a working future looks like. Not a revolution. Just a thing you stop noticing.