New York, 1854. A crowd has gathered in a large exhibition hall. Above them, a man stands on a platform suspended from a single rope, very high up.
He orders the rope to be cut.
The crowd holds its breath. The platform drops a few centimetres... then stops dead. The man takes a bow and declares: "All is safe, gentlemen, all is safe."
His name is Elisha Otis. He has just presented the safety brake for lifts.
The brake that made cities rise
Lifts existed before Otis. But nobody wanted to ride in them: if the rope snapped, you fell. Buildings therefore stayed low, because you could not ask people to climb twenty floors on foot.
Otis's brake changed all that. Once reassured, people agreed to ride. Lifts spread. And a few decades later, the first skyscrapers rose from the ground.
The brake did not hold buildings back. It made them rise.
What gets forgotten in the AI debate
For a month now, the AI debate has looked like a match. On one side, those calling for a slowdown. On the other, those replying that slowing down means losing the race. As if safety and speed were necessarily enemies.
This weekend, however, Microsoft's boss proposed something else: an emergency brake for AIs, which humans could pull mid-task. Not to stop the race, but to be able to keep driving without fearing the crash.
In the mid-nineteenth century, to brake a train, workers ran along the roofs of the carriages, at full speed, to tighten the brake on each carriage by hand. A highly dangerous job, and trains that were inevitably short and slow. In 1869, the American George Westinghouse invented the compressed-air brake, which allowed the driver to brake the whole train with a single action. The result: longer, heavier and faster trains. Once again, it was a better brake that made it possible to go faster.
Why a good brake speeds things up
The reason is simple. Without a reliable brake, the only way to avoid a crash is to go slowly, everywhere and all the time. With a reliable brake, you can accelerate where possible, because you know you will be able to stop where necessary.
Racing drivers know this well: the fastest cars on a circuit are those that brake latest and hardest before the corners. The brake is not the enemy of speed. It is its precondition.
But not all brakes are equal
We must stay clear-eyed. A brake can also become an alibi: announced to reassure, without ever checking that it works. Or a pretext: imposed to hinder a competitor rather than to protect the public.
A real brake can be recognised by three things. Someone has the right to pull it. It has been tested, and it works. And it is not controlled by what it is supposed to stop: a brake that the machine can disable itself is not a brake.
What we take away
The real debate about AI may not be "should we slow down or speed up?". It is rather: what brakes need to be fitted so we can go far without crashing?
Lifts did not conquer the sky because they went fast. They conquered it because everyone knew they could stop.