Gell-Mann Amnesia
Treat the error you caught in your own field as a sample, not an exception
- Difficulty
- Easy
- Time to result
- ~ongoing to results
- Steps
- 5
- Confidence
- 82%
Gell-Mann amnesia is the failure to generalise from an error you already detected. You read a piece on a subject you know intimately and see immediately that it is wrong, the facts garbled, the causation inverted, the emphasis misplaced. Then you turn the page to a subject you do not know, and your trust quietly resets to full. The mechanism is that deep expertise gives you an error detector for exactly one topic, and you never ask what the same detector would have found on all the other pages. Using it properly means treating any error you catch inside your own domain as a sample of the source's process, estimating a rate, and pricing down the unverifiable remainder accordingly. The second move matters just as much: check whether the error cuts against the author's thesis or accidentally proves it.
Origin
The term comes from Michael Crichton, who named it after the physicist Murray Gell-Mann. Naval Ravikant invokes it live on this episode while reading Matt Ridley's How Innovation Works, catching Ridley on Silicon Valley's founding myths and then noticing the correction proves Ridley's own argument.
Core principles
- 01Expertise gives you an error detector for exactly one topic.
- 02The errors you can see are a sample of the errors you cannot.
- 03Trust resets silently when you turn the page. That reset is the bug.
- 04Error rate is a property of the source's process, not of the topic you happened to know.
- 05Catching an error is not the same as discrediting the argument; check which way the error cuts.
How to run it
- 1
Read the source on the one topic you truly know
Deliberately seek out the section of a book, report or article that overlaps with your own deep expertise. That section is the only place your error detector actually works.
Pro tip Do this before you read the rest, so your judgement of the source is not already anchored by agreement.
- 2
Log the errors rather than wincing at them
Write the specific mistakes down: wrong causation, missing prior actor, misattributed credit, garbled numbers. An unwritten error gets forgotten by the time you turn the page.
Watch out Distinguish genuine errors from framing you merely dislike. Only the first kind is evidence about the process.
- 3
Convert the errors into a rate
Turn the list into a rough density, for example two substantive errors in this chapter. A rate is transferable to the rest of the source in a way that a verdict of wrong is not.
Pro tip Rates keep you honest in both directions: a low rate means you should raise your trust, not just lower it.
- 4
Propagate the rate to everything you cannot check
Assume the same process produced the chapters outside your expertise. This is the step everyone skips and it is the whole point of the model.
Watch out Propagate confidence, not conclusions. A source with a moderate error rate is still worth reading, just not worth quoting unchecked.
- 5
Check which way the error cuts
Ask whether the mistake undermines the author's thesis or accidentally supports it. Naval's catch, that Friendster and MySpace preceded Facebook, actually reinforced Ridley's argument that invention is distributed and the winner absorbs the credit.
Pro tip An error that confirms the thesis is a strong signal the thesis is real, because the author found it despite their own sloppiness.
In the wild
Ridley's book presents Facebook and Airbnb as innovators and names Mark Zuckerberg and Brian Chesky as the founders. Naval, who has been in Silicon Valley for decades, immediately sees the omission: MySpace preceded Facebook and Friendster preceded MySpace, while Airbnb was preceded by VRBO, HomeAway, Couchsurfing and Craigslist. He calls it a mild case of Gell-Mann amnesia, then makes the second move, noting the correction proves Ridley's own point about over-lionising a few named inventors when invention is a team and distributed process. Ridley accepts the correction on both counts.
→ Instead of either dismissing the book or letting the error slide, Naval extracted a calibration on the source and a strengthened version of the author's thesis.
An operator reads a market report covering six sectors, one of which is their own. In their sector the report misstates the pricing model and credits the wrong incumbent with a category shift. Rather than discounting the report entirely, they log two substantive errors in eight pages, apply the same density to the other five sectors, and downgrade every number they were about to put in a board deck from fact to estimate needing a second source.
→ The report still informs the strategy, but no unverified figure from it survives into a decision document.
Common mistakes
Turning the page and resetting trust
The amnesia is the default behaviour, not an unusual lapse. Unless you deliberately carry the error rate forward, your brain treats the next section as a fresh, credible source.
Using it to discard the whole source
The model is a calibration tool, not a licence to dismiss. Ridley was wrong on the Silicon Valley details and still right on the underlying thesis, which is exactly the outcome the model should let you see.
Only running it on sources you already distrust
Applied selectively, the model just launders existing bias. The test is worth most on the writers and outlets you are inclined to believe.
Is it for you?
Best for
Anyone forming views from books, journalism or research they cannot independently verify.
Not ideal for
Sources you can verify directly line by line, where you should just check rather than estimate.
From the transcript
“I had a mild case of Gell-Mann amnesia reading your book, if you remember that framework.”
“when it gets to a topic that you're intimately familiar with, you realize it's nonsense or it doesn't quite apply. Yet, you continue believing everything…”
“there was a dissonance where you proved your point by showing that we tend to over-lionize and remember a few inventors as being the creators…”
From the episode
Matt Ridley: How Innovation Works, Part 2
Matt Ridley