Cognitive Biases: The Bugs in Judgment
A bias is a systematic tilt in how you process information. Most of the time it operates silently. But the moment a bias becomes a premise in an argument, it graduates from a psychological fact to a reasoning claim — and can be evaluated as one.
🎯 By the end you'll understand
Nine high-leverage biases — including confirmation bias, anchoring, availability, status quo, sunk cost, loss aversion, fundamental attribution error, and zero-sum bias — the crucial distinction between having a bias and arguing from one, and what debiasing techniques actually do and don't accomplish.
FRAMEWORKSystem 1 and System 2: why biases are sticky
Daniel Kahneman's Thinking, Fast and Slow (2011) gives the most useful frame for why biases persist: the mind runs two overlapping processes.
| Feature | System 1 (Fast) | System 2 (Slow) |
|---|---|---|
| Speed | Automatic, instant | Deliberate, effortful |
| Effort | Near zero | High — tires with use |
| Strengths | Pattern recognition, fast reactions | Complex reasoning, logic, abstract thought |
| Weaknesses | Jumps to conclusions, anchors, conflates familiarity with truth | Lazy — often rubberstamps System 1 output instead of checking it |
| Default? | Yes — runs constantly | Only when called |
Most cognitive biases are System 1 doing exactly what it evolved to do — and System 2 failing to override it. The fix almost always involves getting System 2 to actually engage. The obstacle is that System 2 is lazy, slow, and finite. You cannot run it all the time. You have to pick your battles.
Biases are not a sign of stupidity. They are the predictable side-effects of a fast, efficient heuristic system optimised for a different environment. A sportscar engine tuned for top speed is not broken because it gets poor mileage in city traffic.
CORE CONTENTNine high-leverage biases
Hundreds of named biases exist in the literature. Most cluster around a handful of root mechanisms. These nine cover the most consequential territory for everyday decisions and arguments.
1 · Confirmation bias
Definition: The tendency to search for, favour, and recall information that confirms existing beliefs — and to discount information that challenges them.
Confirmation bias is not primarily about ignoring contrary evidence outright. It is subtler: you assign higher scrutiny to evidence you dislike ("that study has a small sample") while accepting supportive evidence at face value. The prior belief survives encounters with data that should update it.
A manager who already dislikes a team member reads their performance report and notices every flaw while glossing over the positives — and does the reverse for a favourite employee. Both readings feel entirely objective from the inside.
2 · Motivated reasoning
Definition: Using reasoning as a tool to reach a predetermined conclusion rather than to discover truth. The goal is not "what is true?" but "how can I justify what I want to believe?"
Motivated reasoning differs from confirmation bias in being more active: the person constructs arguments and searches selectively for evidence to support a conclusion arrived at for non-epistemic reasons (ego, financial interest, group loyalty). They typically believe their reasoning is genuine.
A consultant recommends the strategy that maximises their future engagement — not because they're dishonest, but because their reasoning started at "this client needs our ongoing services" and worked backward through the evidence to justify it.
3 · Availability heuristic
Definition: Judging the probability of an event by how easily examples come to mind — not by actual frequency data.
Vivid, recent, or emotionally loaded events are more available in memory and feel more common than they are. The reverse is also true: slow, undramatic risks feel negligible even when statistically significant.
After a plane crash dominates the news for a week, many people overestimate flying danger while underestimating the far higher per-kilometre danger of driving. The availability of the story distorts the probability estimate.
4 · Anchoring bias
Definition: Over-reliance on the first piece of numerical information encountered (the anchor) when making estimates or decisions. Subsequent adjustments from the anchor are typically insufficient.
Anchoring is powerful and well-replicated: salary negotiation, price estimation, judicial sentencing, and estimates of completely unrelated quantities are all influenced by an initial anchor. The anchor need not be credible or relevant to exert its pull. You'll feel this yourself in the demo below.
5 · Status quo bias
Definition: A preference for the current state of affairs. Any deviation from the default is perceived as a loss, making change feel costlier than it rationally is.
Status quo bias interacts with loss aversion (below): the current situation becomes the reference point, and any departure from it is coded as a potential loss rather than a potential gain. It explains why default options in contracts, pension plans, and software settings have enormous power — most people never change them.
Organ donation rates differ dramatically between countries that use opt-in vs. opt-out defaults — not because populations have different values, but because most people stick with whatever the default is. The content of the default, not the values, drives the outcome.
6 · Sunk-cost fallacy
Definition: Continuing a course of action because of previously invested resources — time, money, effort — that cannot be recovered, rather than evaluating future costs and benefits on their own merits.
In rational decision theory, sunk costs are irrelevant: the money is gone regardless of what you do next. The rational question is always: given where I am now, what is the best option going forward?
If I hadn't invested anything yet, would I choose to start this today given what I now know? If the answer is no, prior investment is not a good reason to continue.
7 · Loss aversion
Definition: Losses feel roughly twice as painful as equivalent gains feel pleasurable. People will work harder to avoid a $100 loss than to acquire a $100 gain of the same size.
Loss aversion is one of the most replicated findings in behavioural economics. It is not irrational in all contexts — avoiding losses was genuinely more critical than acquiring gains in ancestral environments where losses could be catastrophic. But it systematically biases modern financial and strategic decisions toward excessive risk-aversion about downside, and insufficient attention to foregone gains.
An investor holds a losing stock long past the point where a rational analysis says to exit, because selling "makes the loss real." The paper loss already exists; the reluctance to realise it is loss aversion, not strategy.
8 · Fundamental attribution error
Definition: Overemphasising dispositional (character-based) explanations for others' behaviour while underemphasising situational (context-based) explanations. For our own behaviour, we tend to do the reverse.
When someone cuts you off in traffic, you think "they're a reckless person." When you cut someone off, you think "I was running late and the lane was merging." Same behaviour, different attribution depending on whose behaviour is being explained.
The fundamental attribution error makes us systematically harsh in judging others' failures (blaming character) and sympathetic about our own (citing circumstances). It undermines fair evaluation, management, and policy — because changing character is hard; changing situations is often tractable.
9 · Zero-sum bias
Definition: Incorrectly assuming that one person's gain must be matched by another's loss even when the situation is non-zero-sum — when no such constraint exists (Meegan, 2010, Frontiers in Psychology).
The bias applies correct reasoning from genuinely scarce, fixed-pool resources (food, territory, mating partners in ancestral environments) to modern contexts where value can be created rather than just divided: knowledge, trade, investment, praise, and innovation. Zero-sum bias causes unnecessary competition, hoarding, resentment, and missed cooperation.
P2 — "Whenever someone profits from investment, another person loses" — is a direct expression of zero-sum bias applied to financial markets. It treats investment as if the total wealth available is fixed, when in fact productive investment can grow the total.
INTERACTIVEFeel the anchor pull
Anchoring is easy to describe but hard to feel. Answer both questions quickly — don't overthink. Then see what happened.
KEY DISTINCTIONHaving a bias vs. arguing from one
This distinction is essential and often missed. It draws a clean line between psychology and logic.
A person can have a bias — feel it, act on it unconsciously, let it shape their preferences — without ever making an explicit argument. A loss-averse investor who quietly avoids certain assets is expressing a psychological tendency. That tendency might be suboptimal, but it is not an argument and so cannot be evaluated as reasoning.
The moment that same person says "Because my gain on this investment must be someone else's loss, I shouldn't invest" — the bias has become a premise. It is now making an explicit claim that can be evaluated, challenged, and potentially refuted:
- Is the claim "my gain must be someone else's loss" true as a universal? (No — we saw this in lesson 1.6.)
- Is the claim probabilistically well-supported? (Depends on the type of investment — not as a blanket statement.)
- Does it follow that something generating occasional harm is therefore always wrong? (Not without further premises about proportionality, consent, alternatives.)
Bias as psychology: a tendency, a feeling, a disposition. Visible in behaviour. Not yet a claim. Not yet evaluable as reasoning.
Bias as premise: an explicit empirical or causal claim that expresses the bias's logic — e.g., "because gain requires loss." Now it is a claim. It can be tested against evidence. It can be contradicted by a counterexample. It can be shown to be too universal. The tools of Stages 1 and 2 apply directly.
This matters practically because it tells you when it is appropriate to challenge a bias as a reasoning error. Telling someone "you're being loss-averse" when they haven't made an argument is a psychological observation, not a logical refutation. Showing that their explicit premise — "gains always require equivalent losses" — is false as a universal claim is a logical refutation, and a legitimate one.
APPLIEDBack to our argument: zero-sum bias as premise
The running example illustrates the transition from bias to premise precisely:
P1: "It is wrong to profit when another person loses."
P2: "Whenever someone profits from investment, another person loses."
C: "Therefore, investing is wrong."
P2 is zero-sum bias stated as a premise. The person making this argument may genuinely feel that investment is a form of taking — an emotional intuition driven by zero-sum thinking. That intuition is a psychological state. But by articulating it as P2, they have converted it into a testable empirical claim.
And as a testable claim, it fails. As established in lesson 1.6: the universal form is defeated by a single counterexample (productive investment that creates new value); the narrower question of whether investment is often harmful is an empirical question requiring systematic evidence; and the jump from "some harm occurs" to "investing is wrong" requires additional premises that the argument does not supply.
The bias did not make the argument wrong by being a bias. It made the argument wrong by generating a false empirical premise. That is the analysis to make — not "you're biased" but "this specific claim, which reflects a zero-sum framing, is false as a universal and unsupported as a statistical generalization."
PRACTICALDebiasing: what actually helps
Not all debiasing works equally well. Here are the approaches with the strongest evidence base.
Consider the opposite
Before finalising a judgment, deliberately generate reasons it might be wrong. Research on confirmation bias shows that the simple instruction — "list three reasons your view might be incorrect" — reduces bias measurably. It forces System 2 to generate counter-evidence that System 1 would normally suppress.
When you've reached a conclusion, ask: "What would have to be true for me to be wrong about this?" Then actually answer it — not as a formality, but as a genuine search.
Use base rates (the outside view)
The availability heuristic and overconfidence both improve substantially when people consult frequency data rather than relying on memory. Before estimating a probability, ask: "What is the historical base rate for this class of event?" Kahneman and Tversky called this the "outside view" — stepping outside your specific situation and asking how situations like yours usually go.
A startup founder estimating their chances benefits from knowing that roughly 90% of startups fail within ten years — not because that kills ambition, but because it correctly frames the prior before you start counting reasons yours will be different.
Premortems
Imagine it is one year from now and the project has failed. What went wrong? By assuming failure rather than asking "will this fail?", the exercise bypasses motivated reasoning. Developed by Gary Klein, championed by Kahneman as one of the most practical organisational debiasing tools.
Have each team member independently write likely reasons for failure, then share. The process surfaces concerns that social pressure would otherwise silence.
Separate the decision from the decision-maker
Structured processes — checklists, decision trees, blind evaluation where possible — reduce the influence of identity and motivated reasoning. When a hiring committee evaluates candidates on explicit criteria before discussing impressions, confirmation bias and halo effects are reduced (not eliminated).
HONEST ASSESSMENTThe limits of debiasing
Here is the uncomfortable truth: the evidence for debiasing is much weaker than the evidence for the biases themselves.
- Knowing about a bias does not make you immune. People who score high on "need for cognition" tests — people who enjoy reasoning and know about cognitive biases — show roughly as much anchoring and confirmation bias as those who don't.
- Training effects fade. Debiasing training shows short-term effects. Long-term transfer to real decisions is modest at best.
- Motivation matters enormously. Debiasing techniques work better when people genuinely want to be accurate. If you are being paid to reach a particular conclusion, no amount of "consider the opposite" will overcome that incentive.
- The techniques require effort. They all demand System 2 engagement. In conditions of stress, time pressure, or fatigue — conditions under which the most consequential decisions are often made — System 2 is the first thing to go.
The goal of understanding biases is not to become bias-free — that state is not reachable. The goal is to identify high-stakes decisions where bias is most costly, apply debiasing tools deliberately in those situations, and build structures (checklists, diverse teams, explicit criteria) that reduce the system-level impact of individual cognitive errors.
PRACTICE🔨 Weekly exercise
Keep a brief daily log. Each day, record one judgment you made — a belief, a decision, a reaction — and then ask: "What evidence could change my mind about this?"
The exercise has two effects. First, it makes you notice when you can't actually answer the question — when a belief you hold has no falsification condition, which is a warning sign. Second, it builds the habit of treating your own judgments as revisable hypotheses rather than fixed truths.
Aim for specificity: not "I should be more open-minded" but "I believe X because of Y — here is the evidence that would shift me: Z." One entry per day for seven days.
WEAKNESSCommon mistakes to avoid
This is itself a bias — the bias blind spot. People systematically rate themselves as less biased than average. Awareness of biases in the abstract does not grant immunity to them in practice.
Labelling someone's argument as a cognitive bias is not a counter-argument. The bias might be present and the conclusion might still be correct. Deal with the evidence on its merits — which is exactly what lesson 1.6 showed you how to do.
Sunk costs are literally irrelevant to future expected value. The feeling that they matter is real; the inference that they should change your forward-looking decision is wrong.
GLOSSARYKey terms introduced
- System 1
- Fast, automatic, effortless thinking; runs constantly; source of most heuristics and biases.
- System 2
- Slow, deliberate, effortful thinking; limited capacity; often fails to override System 1.
- Confirmation bias
- Seeking and favouring information that confirms existing beliefs.
- Motivated reasoning
- Using reasoning to justify a predetermined conclusion rather than to find truth.
- Availability heuristic
- Judging probability by ease of recall rather than actual frequency.
- Anchoring bias
- Over-relying on the first number encountered when making estimates.
- Status quo bias
- Preferring the current state; perceiving any change as a loss.
- Sunk-cost fallacy
- Continuing a course of action because of past investment that cannot be recovered.
- Loss aversion
- Losses loom roughly twice as large as equivalent gains.
- Fundamental attribution error
- Over-attributing others' behaviour to character; under-attributing to situation.
- Zero-sum bias
- Incorrectly assuming one person's gain requires another's equivalent loss (Meegan, 2010).
- Bias blind spot
- The tendency to see oneself as less biased than other people.
- Premortem
- A structured exercise assuming a plan has already failed and asking why, to surface hidden risks.
ACTIVE RECALLCheck your understanding
Answer in your own words first, then reveal.
1What structural feature of System 1 vs. System 2 explains why knowing about a bias doesn't make you immune to it?›
System 1 is fast, automatic, and runs constantly — it generates quick judgments that feel correct. System 2 is slow, effortful, and limited — it would be capable of catching System 1's errors but defaults to laziness and often simply rubberstamps System 1's output. Knowing about a bias is a piece of knowledge stored in System 2's domain; it doesn't change the automatic output of System 1. You would need System 2 to actively engage and override — which requires deliberate effort and a specific prompt, not just general awareness.
2What is the critical difference between having a bias and arguing from one? Use the running example to illustrate.›
Having a bias is a psychological state — a tendency, feeling, or disposition that may shape behaviour unconsciously. Arguing from a bias means the bias's logic has been converted into an explicit premise. In the running example: a person may feel that investment is taking from others (zero-sum bias as psychology). The moment they state P2 — "Whenever someone profits from investment, another person loses" — the bias has become an empirical claim. That claim is now testable, can be contradicted by counterexample (productive investment creates new value), and can be shown to be an unsupported universal. The tools of argument analysis now apply directly.
3Distinguish loss aversion from the sunk-cost fallacy. Are they the same bias?›
They are related but distinct. Loss aversion is a general asymmetry: losses feel roughly twice as painful as equivalent gains feel pleasurable. This shapes risk preferences — people are more motivated to avoid losing $100 than to acquire $100. The sunk-cost fallacy is a specific downstream consequence: because losses loom large, the "loss" of having invested time or money in something makes people reluctant to abandon it, even when future expected value says they should. Sunk-cost fallacy is partly driven by loss aversion — selling a losing position "makes the loss real" — but also by ego protection and escalation dynamics. Loss aversion is the broader mechanism; sunk-cost fallacy is one of its more consequential manifestations.
4Name two debiasing techniques with genuine evidence behind them, and state one important limit of debiasing in general.›
Two evidence-backed techniques: consider the opposite (deliberately generate reasons your belief is wrong) and base rates / outside view (consult historical frequencies rather than reasoning from the specific case). One important limit: knowing about biases does not make you immune — the bias blind spot means people rate themselves as less biased than others regardless of training, and debiasing effects fade over time without structural supports. Techniques also require System 2 engagement, which degrades under stress and fatigue — exactly the conditions under which consequential decisions are made.