Theory-Induced Blindness

What it is

The coinage is Kahneman's own, introduced in Thinking, Fast and Slow with the words "I call it", so this is not a case where a famous name has absorbed someone else's work.

His two worked cases are from the prehistory of his own field and their power is in the timescales. Bernoulli's utility model, the account of decision-making under risk that economics ran on for centuries, had no reference point in it: outcomes were treated as states of wealth rather than as changes from where a person was standing. The omission is not subtle once seen, and it went unnoticed for more than two hundred and fifty years, by people who used the model daily and were not fools. Then there is Markowitz, who published a version of the missing idea about a quarter of a century before prospect theory and was largely ignored. The correction was not merely unthought. It was in print, and the field could not use it, because there was no felt problem for it to solve.

Kahneman then turns the concept on himself and on prospect theory, which is the move that makes it credible. He gives several pages to his own theory's blind spots: it has no regret in it, no disappointment, and it assumes options are evaluated independently, which he says does not hold. He closes by noting that the theory was accepted not because it is true, and that he and Tversky were lucky.

In effect

The pattern is the one this publication keeps catching. A trade book endorses a finding. The finding becomes a tool for explaining things. The tool works well enough that the anomalies get the benefit of the doubt. Then somebody runs a preregistered replication and the tool turns out to have been holding nothing up. Ego depletion, social priming and the willpower literature are each that story.

Everything the vault does about evidence status is machinery built against this, and the countermeasures are all versions of refusing the benefit of the doubt: an evidence status field that forces a verdict on sourcing to be written down rather than assumed, provenance markers separating a full text read from an abstract only, a reliability audit on every book that asks what has failed rather than what the book says, a two-sides rule that keeps a live dispute live instead of resolving it in favour of the tool currently in use, and the discipline of crediting the first author rather than the famous one.

The clearest instance in the library outside Kahneman's own examples sits in The Elephant in the Brain, where the authors, having argued that laughter is a play signal, note that the account also explains the occasions when nobody laughs and offer that as a strength. The anomaly is seen, is handled, and the handling is presented as depth. The book also raises a warning about generalising from brain-damaged subjects on the page before it generalises from them.

The practical rule is that when a theory is working well is exactly when its anomalies are least visible, so an audit has to be scheduled rather than triggered. Waiting until something feels wrong means waiting for a signal the mechanism suppresses.

What it does not say

It does not describe a measured effect, and anyone who writes that studies show theory-induced blindness has misunderstood what they are quoting. There is no experiment, no sample and no effect size, and asking whether it replicates is a category error.

It is not the same thing as an unfalsifiable frame, and the two should not be merged. This concerns what a person holding a theory can see, where the anomalies are noticed and the doubt is deferred. That one concerns what a theory is built to admit, where the anomaly is not deferred but counted as confirmation. One is a fact about readers, the other a property of an argument.

It does not excuse anyone. Kahneman applied it to his own theory in the chapter that introduced it, which is the standard the concept sets.

---