Better Anyway

The arithmetic under the story

In 1886 Francis Galton published a paper on the heights of a few hundred English families. Tall parents had tall children, but on average not quite as tall as themselves. Short parents had short children, but not quite as short. The children landed about two thirds as far from the average as their parents. Galton called it regression towards mediocrity, and he thought he had found a law of heredity.

He had found something bigger, and it has nothing to do with heredity. Any result you can measure is made of two parts: the part that stays and the part that does not. Skill, health and the real quality of a thing stay. Luck, weather and the mood of the day do not. When you pick out the most extreme results, you are not picking out only the most extreme people. You are picking out the ones who had both a real edge and a good roll that happened to point the same way. Measure them again and the edge is still there. The roll is not. They come back toward the middle, not because anything was done to them, but because part of what made them extreme was never theirs to keep.

That is the whole principle. The less of a result belongs to the thing being measured, the further it falls back. It is true by arithmetic, so it is true of everything: heights, golf scores, a quarter’s sales, blood pressure, a restaurant’s reviews, the way you feel on a Wednesday night.

The instructor who was right

In the 1960s a young psychologist named Daniel Kahneman was teaching Israeli Air Force flight instructors that praise works better than punishment. A senior instructor told him he had it backwards. When he praised a cadet for a clean maneuver, the next one was usually worse. When he screamed at a cadet for a bad one, the next one was usually better.

The instructor’s observation was correct. His conclusion was wrong. Praise came after the best attempts, which were partly luck, so the next attempt fell back. Shouting came after the worst attempts, which were partly bad luck, so the next one rose. Kahneman had the instructors toss two coins at a target behind their backs, with no feedback at all. Those who did best the first time mostly did worse the second, and the worst mostly did better. Nobody had said a word to anyone.

The conclusion he drew is a cruel one. Because we reward people after they do well and punish them after they do badly, the world quietly punishes us for rewarding and rewards us for punishing.

Now notice when you act. You see the doctor on the worst day. You fire the coach after the worst season. You change the strategy after the worst quarter, start the diet after the worst weigh-in, and call the repairman when the noise is loudest. Every one of those moments is an extreme, and every extreme is followed, on average, by something better. Whatever you did gets the credit. A large share of what people believe works is really the timing of when they tried it.

What AI changes

Regression is nearly invisible to one person, for a structural reason. You live one story, and you only ever see the version where you acted. The fix people found is the control group: others at the same extreme who did nothing, so you can see how far they came back on their own. Medicine built its method around that comparison, largely to beat this one illusion. But nobody carries a control group through daily life. No one could keep the books.

A personal AI delegate can. It holds your record: every bad week, every recovery, every remedy, and how long the last six bad weeks took to fix themselves. It can hold the ordinary course for people like you in moments like this. For the first time, the comparison that science needed a laboratory for can travel with an ordinary person, quietly, at the moment of decision.

Here is the part most people will miss. The first thing regression will flatter is the AI itself. We call in help at our worst moments: the inbox at its most flooded, the business in its worst month, the body on its worst day. Whatever we call in will be followed by improvement, and we will credit it. Every new tool, every new hire and every new delegate arrives at an extreme and rides the recovery. So the test of a delegate worth keeping is not whether things got better after you brought it in. Things were going to get better. The test is whether it tells you so, including about the part it would love to take credit for.

The delegate worth keeping is the one that says: you were going to get better anyway.

Five predictions

1. Before-and-after stops being evidence.

   “It worked for me” will be met, politely, by your AI asking: compared to what? Testimonials, before photos and turnaround stories lose their force, because your delegate knows they were nearly all recorded at the bottom.

2. Delegates will refuse credit.

   The ones people trust longest will report their own results net of the recovery that was coming anyway. That phrase, or something like it, becomes a normal line in a monthly report.

3. Fewer changes at the bottom.

   People with delegates will stop rewriting strategy, letting people go and switching providers on the worst week. Judgment moves from single extremes to runs. An employee’s terrible month gets read as a terrible month.

4. Praise comes back.

   The old belief that harshness works was built on regression. Once parents, coaches and managers can see in their own records that the praised did not get worse because of the praise, that belief loses its evidence.

5. Your bad day becomes a warning light.

   Offers that find you at your worst moment get flagged, because that is exactly when any remedy will look like it worked. Sellers have always known to find you at the bottom. Your AI will know it too.

Living it

I publish here nearly every day, and I have handed most of the choosing to my AI. Some pieces land and some do not. The day after one does unusually well, the next will almost always do less, and the easy story is that the AI lost its touch and I should take the wheel back. That story is regression wearing a costume. I teach young people too, and the same rule holds there: a student who has a brilliant first month will almost certainly have a quieter second one, and the worst thing I could do is tell them they slipped. I have written about where the bottleneck goes when work gets cheap in The Last Line Is You, about why a measure lies once you lean on it in The Wager Inside Every Measure, and about a sense that reports the wrong thing in The Leans. This is the same family. The extremes are not where the truth is. They are where the noise is loudest.

The claim ledger