第 1 章 · 20 分钟

Substitution: how a question gets swapped for another

The load-bearing mechanism, and the standing of the 'two systems' phrase.

概念地图

Substitution

A hard question is swapped for an easy one and the answer is carried over.

深色为本章概念，浅色为其他章

陪练模式

不想只是读？让它带你走一遍。

导师把这一章拆成小步，每一步先问你，再讲；你用自己的话答，它先认下对的部分再纠偏。用你自己的 AI，对话只存在这台设备上。

对话只存在本机

Locus01

延伸阅读

Simon · 1950s · bounded rationality

Argued that computation and information are both limited, so people use rules that are good enough rather than optimal. He argued against the economic assumption of a costless calculating agent; the cost is that 'good enough' was never formalised, and later authors filled it in differently.

Foundational

Tversky and Kahneman · 1974 · Judgment under Uncertainty

Turned 'good enough rules' into three experimentally operable heuristics, each with a predicted direction of error. For the first time the departure could be bet on in advance; the cost is that the boundaries between the three are vague, and they are often used to label results after the fact.

Turn

The Gigerenzer line · from the 1990s

Holds that heuristics are ecologically rational in their natural environments, and that laboratory bias largely comes from how the task is presented (frequencies versus probabilities). The most persistent direct reply to the line above.

Counterexample

Kahneman · 2011 · Thinking, Fast and Slow

Gathers forty years of work into a popular book and introduces the System 1 / System 2 phrasing. This course treats the standing of that phrase separately: is it a narrative device or a mechanism?

Subject of this course

机制02

Question substitution

Asked whether a company is worth investing in over five years, people usually answer how they feel about the company now. The target question is hard, the heuristic question is easy, the answer is carried straight over, and the person does not notice the swap. Kahneman calls this step substitution. Most of the biases in the book are instances of it: which question was swapped in decides which way the error goes.

机制A hard question, through being swapped for an easier nearby one, produces a departure whose direction is predictable.

可迁移性测试

Move it to an architecture review: asked whether a design will hold for three years, the reviewer actually answers whether the document is clearly written. The isomorphism breaks in that document quality and design quality are at least slightly correlated, while 'how I feel now' and 'five-year return' are almost not.

机制03

The 'two systems' phrase

The book calls fast automatic processing System 1 and slow effortful processing System 2. What matters is the standing of the phrase: Kahneman states in the book that it is a personifying narrative device, not two entities. This course therefore files it under exposition rather than mechanism: it helps you remember when substitution occurs, it does not explain why substitution occurs, and it cannot be falsified on its own.

机制A label, through naming a set of phenomena, makes them easier to remember while adding no explanation.

可迁移性测试

Move it to medicine: calling a cluster of symptoms a syndrome helps communication and retrieval, but while the cause is unknown the name explains nothing. The isomorphism breaks in that a syndrome at least has diagnostic criteria, whereas System 1 has no criterion independent of the outcome.

Derivation04

From substitution to 'the direction of error is predictable'

What is being derived is the book's most valuable claim: the departure is not noise, it has a direction, and so it can be bet on in advance.

There is a target attribute T and a heuristic attribute H (read: the easy stand-in)
Chosen: it assumes an identifiable stand-in exists. Some judgements have none.
H is easier to evaluate than T
Chosen, and the precondition for substitution to occur at all.
The judge does not notice the swap
Chosen: it assumes reflection does not intercept it. Feedback and accountability weaken it, as chapter 3 shows.

推导 · 0 / 4

裂缝05

Cracks in this framework

争议地形06

本书主张

Judgement departs systematically under identifiable structures, and the departure comes from swapping a hard question for an easy one; these biases are stable, replicable and directionally predictable.

另一种看法

The ecological rationality line holds that heuristics perform well in the environments they evolved for, and that most laboratory bias is an artefact of task presentation. Calling the results 'errors' applies a normative standard detached from the environment.

分歧扎在

The disagreement is rooted in what counts as the correct answer. The book uses probability theory and expected utility as the benchmark; the other side says the benchmark should be performance in the natural environment. Neither disputes the data.

什么证据能裁决

What would settle it is the same subjects under two presentations: pose the same judgement as a single-event probability and as a natural frequency and compare the size of the bias. If the frequency version shows the same bias, the ecological account fails; if it shrinks markedly, part of the bias comes from presentation.

The evidence points to both being partly right: frequency framing does reduce some biases but does not remove all of them — anchoring is stable under both. What is missing is an ex ante criterion for which biases are presentation artefacts; current practice classifies them one effect at a time, after the fact.

Falsification07

Question substitution as a preregistrable test

The mechanism under test: a hard question, through being swapped for an easier nearby one, produces a departure whose direction is predictable.

Data source: newly collected judgement data, with items and stand-in attributes registered and published before collection
The whole point is naming the stand-in in advance, which an existing dataset cannot do.
Sample: a pre-computed sample size, all participants analysed, with no exclusion on response time or attention checks
Post hoc exclusion is the commonest degrees-of-freedom entry in this literature; sealed off in advance.
Window: a single session, within-subject, both versions completed in the same sitting
Within-subject removes individual differences; the same sitting keeps mood and time-of-day out.
Threshold: the correlation between judgement and the registered stand-in attribute is significantly higher than between judgement and the target attribute
This is the direct meaning of substitution: the judgement tracks the stand-in, not the target.
Failure condition: the two correlations do not differ significantly, or the judgement tracks the target more closely
Either counts as refutation.

推导 · 0 / 3

接口08

Which slot it hangs on

挂在哪个槽位The conclusion you already hold is probably 'people have cognitive biases, so think it over' — a reminder to yourself. This chapter goes at the part of it that can actually be executed.

Does this chapter (a) replace 'think it over', or (b) constrain it by asking which question you have substituted this time?

慢变量Register two slow variables: how often you write down the stand-in question on important judgements, and what share of your judgements ever get checked against an outcome. Look at the second first — a judgement with no feedback never corrects its own substitution.

小结09

本章小结

01Substitution is the load-bearing mechanism: a hard question is swapped for an easy one and the answer is carried over.

02'The error equals the difference between the two questions' is an identity; the finding is that such tasks can be built in advance.

03System 1 / System 2 is a narrative device: exposition, not a mechanism.

04Natural-frequency presentation and immediate error feedback both weaken substitution — its two failure conditions.

05Naming the stand-in attribute after the fact carries no empirical risk and is unfalsifiable.

提取练习 · 合上书，先自己答一遍。

?Is 'the departure from the target equals stand-in minus target' (a) an identity or (b) an empirical claim?

?Does 'experts are accurate in domains with immediate feedback' weaken (a) assumption two or (b) assumption three?

Forced choice10

A · Identify the mechanism

Which of these is a mechanism?

二选一

Forced choice11

C · Locate the crack

In which situation does this framework give a confident and wrong answer?

二选一

读到这里，把它记为已读。