The autopilot never forgets how to fly. The pilot does.
On the night of June 1, 2009, an Airbus A330 was somewhere over the Atlantic, halfway from Rio to Paris. A small pressure probe on the outside of the aircraft iced over. The computer lost its airspeed reading and did the sensible thing: it switched off the autopilot and handed the airplane back to the two men in the cockpit. They had a working aircraft, full engines, clean systems. Four and a half minutes later it was in the ocean. All 228 people aboard died.
The pilots pulled back on the stick and held a perfectly recoverable airplane in a stall the whole way down. A synthetic voice called out "stall" more than seventy times. They did not believe it, or did not know what to do with it. Nothing was wrong with the airplane that a few hundred hours of hand-flying would not have caught in a second.
Twelve years earlier, an American Airlines training captain named Warren VanderBurgh had stood in a room in Dallas and described this exact accident before it happened. He had gone through the company's incidents and found that most of them traced back to the same thing. Pilots had learned to manage the automation instead of fly the plane. They followed the magenta course line the flight computer drew across the display, and the following became the skill. He called them children of the magenta line.
Autopilots made flying far safer. The real danger is quieter, and it sits underneath the safety. A tool built to lower your workload will, if you let it, lower your competence, and you will not notice, because everything works. The bill comes due in one specific moment: when the tool quits and hands the job back, and the job is now the hardest version of itself.
In 1983 the psychologist Lisanne Bainbridge wrote a short paper called "Ironies of Automation" that explained the crash before Airbus built the plane. Her argument runs in two steps and closes like a trap.
Step one. We automate the parts of a job a machine can do, and we leave the human the parts it cannot. So the human is left with the rare cases, the strange ones, the genuinely hard ones. Step two. A skill you do not practice fades. The more reliable the automation, the less the human practices, and the rustier they get.
Put the two together. The system calls the human in only when it fails, which is the worst possible moment, and by then the human is the least prepared they have ever been. The better the tool, the sharper this trap. William Langewiesche, writing after the crash, put the loop plainly: automation erodes skill, and eroded skill is used to argue for more automation.
The brain keeps its maps in a small seahorse-shaped structure called the hippocampus. That is where spatial memory lives. Now consider a London cab driver. To get licensed, they spend three to four years on a moped memorizing every one of roughly 25,000 streets inside a six-mile circle around Charing Cross, plus thousands of landmarks. It is called the Knowledge. About half wash out.
In 2000, Eleanor Maguire scanned London taxi drivers and a comparison group. The drivers' posterior hippocampus, the part that stores spatial maps, was measurably larger. The longer someone had driven, the larger it was. A later study followed 79 trainees over four years. The ones who passed the Knowledge grew that region. The ones who studied and failed did not. The map was building the brain that held it.
Now run it backwards. In 2020, Louisa Dahmani and Véronique Bohbot measured lifetime GPS use in fifty regular drivers, then tested how well they found their way with the arrow switched off. The people who had leaned on GPS the most did worst without it. Three years later they retested some of them. Heavier GPS use in between tracked a steeper decline. Same organ, opposite direction. The cabbie builds the map by hand and grows the muscle. The rest of us follow the arrow and let it soften.
The same trade runs through your ordinary day, one skill at a time. Every skill you hand to a tool, you also hand over the ability to tell when the tool is wrong, and this is where you find out which ones you have already given away.
Eight everyday skills. For each, be honest about where you stand. Not what you should do. What is true right now.
This is why some exercises in this course say TRY WITHOUT AI before they say anything else. Not because the tool is the enemy. Because the only way to keep the yardstick is to use it now and then, on purpose, while the stakes are a homework problem and not four minutes over the Atlantic.
The larger posterior hippocampus in taxi drivers came with a smaller anterior one, and they were measurably worse at some other kinds of spatial memory. Building one map that well cost them flexibility elsewhere. Even the person who keeps the skill pays for it. There is no free version of any of this.
In 1974 Harry Braverman wrote "Labor and Monopoly Capital", about what factory automation did to craft workers. A skilled machinist held knowledge in his hands and his judgment. Break the job into steps a machine can run, and the knowledge moves into the machine. The worker who remains does not need the skill and slowly does not have it. Braverman called it the degradation of work. The magenta line is the same story with a nicer screen.
Bainbridge noted one more turn. To keep a human able to take over, you have to let them practice on the real system, which means occasionally turning the reliable automation off and accepting the small risk that comes with a rusty human at the controls. Safety and skill-retention pull against each other. There is no setting that gives you both for free.
Pick one skill from your list that landed in "couldn't do it without the tool." Some time this week, do that one thing once with the tool switched off. Split the bill in your head. Find the address from memory. Write the paragraph before you ask the model. You are not proving anything to anyone. You are checking whether the yardstick is still there.