First principles · The pratfall effect
After the spill
A blunder makes a capable person more likeable, and only a capable one. AI is about to make everyone look capable and to make blunders cheap to fake. The spill stops proving anything. What you do in the next ten seconds starts proving everything.
A flaw only warms a performance that was already trusted.
The same spill, on two different people, moves the room in opposite directions.
- Capable · no spillAdmiredImpressive, and a little far away.
- Capable · spillLiked mostThe polish now has a person inside it.
- Ordinary · no spillForgettableNothing to admire, nothing to forgive.
- Ordinary · spillLiked leastThe flaw confirms what you already suspected.
The vendor’s deck is flawless. Every slide lands, every number ties out, every answer arrives before you finish the question. You are impressed, and you are not going to remember any of it.
Then the projector dies. The founder looks at the dark screen, laughs, says “well, there goes my best slide,” picks up a marker and draws the whole thing on the whiteboard, badly, from memory. That is the part you tell your partner about at dinner. That is when you decided.
The rule
In 1961 the Bay of Pigs invasion collapsed, and President Kennedy took the blame for it in public. His approval went up. A young social psychologist named Elliot Aronson found that strange enough to test. In 1966, with two colleagues, he had students listen to a tape of a young man trying out for a college quiz team. In one version the candidate was brilliant. In another he was ordinary. And in half the tapes, near the end, there was a clatter and the candidate said, “Oh my goodness, I’ve spilled coffee all over my new suit.”
The brilliant candidate who spilled was liked best of all. The ordinary candidate who spilled was liked least. Same spill, same clatter, same suit. Opposite effect.
Here is why, in plain terms. A flawless performance is admirable and slightly inhuman. You cannot quite locate the person behind it, so you keep your distance. A blunder from someone you already trust is a small surprise against a high expectation, and the surprise tells you something the polish could not: there is a body in there, someone who can drop things, someone like you. The same blunder from someone you did not trust is no surprise at all. It just confirms the low expectation and lowers it further.
The spill was never charming. It was evidence that someone was there.
That is the first principle hiding inside the experiment. A flaw is information, and information is measured against what you expected. It warms only when it lands on top of trust that was already earned. You cannot spill your way into being liked. You can only spill your way into being real, once you are already good.
What AI changes
AI changes two inputs at once. First, it makes competence the default. The email is clear, the proposal is clean, the deck ties out, the code compiles, for nearly everyone, on the first try. In Aronson’s terms, we are about to put every candidate in the brilliant column.
The obvious reading is that this makes the pratfall more powerful than ever: if everyone is polished, then every flaw should humanize. That reading misses the second input. The same machines that make polish free make imperfection free too. A writing tool can add a casual typo, a little hesitation, a warm aside, on request. Once a flaw can be manufactured, it stops being evidence of anything. A typo used to mean a person was typing. Soon it will mean nothing at all.
So the spill dies as a signal. But the reason it worked does not die, because the reason was never the coffee. The reason was surprise: an unplanned moment that showed what was underneath the polish. And there is one part of an unplanned moment that cannot be prepared in advance, which is what happens next. You can stage the spill. You cannot stage your response to the spill you did not stage. The signal moves from the flaw to the recovery: the next ten seconds, live, with something at stake and no script.
I have written before about how the machines at the center of modern AI are built to remove surprise, in The Machine That Eats Surprise, and about what happens when surprise is manufactured, in Counterfeit Surprise. The pratfall sits exactly between those two. A perfectly predicted performance gives you nothing to feel, and you cannot feel a person who never surprises you. A faked surprise gives you something to feel and nothing to trust. Only the real one does both.
You have two proposals for the warehouse job. Both are flawless. Both were plainly made with the same kind of tool, and you could not tell them apart if the logos were removed.
Then a second email arrives from one of them. “I had your square footage wrong in yesterday’s quote. Here is the corrected number, here is how I got it wrong, and I have taken the difference off.” You read it twice. You sign with that one before lunch.
What I expect
- Manufactured imperfection becomes a feature, then becomes worthless. Tools will offer to make writing sound more human by roughening it. For a short while it will work. Then readers will learn that roughness is a setting, and they will discount it entirely, the way we already discount a stock photo of a smiling team.
- A flawless record starts to read as unverified. A portfolio, a résumé or a page of testimonials with no visible miss will be read as smoothed, not as excellent. The question people learn to ask will be: show me one that went wrong, and what you did.
- The correction becomes the credential. Owning a specific mistake in public, named, dated, fixed and paid for, will become one of the strongest trust signals a person or a business can send. Some firms will keep a public log of their errors and corrections, and it will win them work.
- Live and unedited rooms gain value. The phone call, the site visit, the unscripted question at the end of the meeting: anywhere the next ten seconds cannot be prepared will matter more, not less, because it is where the recovery can be watched.
- The scripted vulnerability post dies. “I failed, and here is what I learned” is already a genre. Machines will produce it at scale, and it will stop moving anyone. What will move people is a failure that was not chosen for the story, handled while they watched.
Where I am standing
I publish every day, with AI in the room for most of the work, so I am the brilliant candidate on the tape whether I like it or not. The polish is no longer mine to claim. What is still mine is the moment I get something wrong in front of you, and what I do next: whether I name it, fix it and say so plainly, or quietly edit and hope. That moment is the only part of the work no machine can do on my behalf, because it is a response to something nobody planned. It is also the part I have come to think matters most. It sits beside what I wrote in Face Value about the human face as the place where liability lives, and in Trust Keeps Its Own Clock about trust accruing at its own pace.
The claim, and how it fails
- Already true
- Polished writing, decks and proposals already cost almost nothing to produce, tools can already add casual imperfection on request, and flawless prose already draws the suspicion that a machine helped.
- What has to happen
- Competence has to become the default expectation for nearly everything people write or show, so that a flaw no longer tells you a person is present and only the response to a real, unplanned flaw still does.
- Where I am probably wrong
- Aronson’s result came from one small experiment, and later studies found the effect depends on who is watching. If it is weaker than its fame, the whole frame is thinner than I have drawn it. And if people simply stop caring whether a person is present at all, the signal will not move to the recovery. It will just disappear. If by the end of 2029 a spotless record wins work more reliably than a mistake owned in public, I was wrong.
Back to Thursday
The projector is still dark. The founder is still at the whiteboard, drawing a chart that is worse than the one on the dead slide. Nobody in the room is looking at the chart. Everyone is watching her, because for the first time all afternoon something has happened that she did not plan, and they want to see what she does with it.
That was the whole pitch. It always was.
Polish is free now. What you do after the spill is not.
Background
- Elliot Aronson, B. Willerman and J. Floyd, “The effect of a pratfall on increasing interpersonal attractiveness,” Psychonomic Science (1966).
- Elliot Aronson, The Social Animal.
- John Rector, The Machine That Eats Surprise.
- John Rector, Counterfeit Surprise.
- John Rector, Face Value.
Related: The Light Was Never the Point · Trust Keeps Its Own Clock · The Machine That Eats Surprise