Daniel Kahneman on Cutting Through the Noise
Daniel Kahneman — Nobel laureate and author of Thinking, Fast and Slow — joins Tyler Cowen in Episode 56 to range across happiness, memory, bias, noise in human judgement, superforecasting, artificial intelligence, and the collaboration with Amos Tversky that reshaped how we understand the mind.
Key ideas
- Noise is the underestimated twin of bias. Bias — a systematic tilt in one direction — gets most of the attention in behavioural research. But noise (random, inconsistent variation in judgements that should be identical) is at least as large a problem. When Kahneman studied insurance underwriters, two specialists assessing the same case differed by 50% on average — a figure that shocked the firm’s own executives, who expected 10%.
- Memory, not experience, is what we optimise for. Happiness in the moment and satisfaction with how one’s life has gone are distinct things, and people demonstrably optimise the second. Memories persist; the experience itself is gone instantly. This is why peak moments and endings dominate how we evaluate events, even when their duration was trivial.
- Delay intuition; break decisions into independent dimensions. For problems outside a domain of genuine expertise, forming an overall impression too early corrupts the use of information. The better approach: assess each relevant dimension separately, then aggregate — preventing any single vivid datum from colonising the whole judgement.
- Optimism is capitalism’s engine, not a flaw to correct. Overconfidence and optimism are not the same thing. Optimism — exaggerating one’s odds of success — makes people more appealing, secures resources, and induces risk-taking. At the individual level the expected value may be negative; at the societal level it is what drives economic progress.
- Expert intuition requires regularity, feedback, and experience — conditions most professionals lack. Chess masters and firefighters develop reliable intuition because their environments are regular and feedback is rapid and unambiguous. CEOs rarely meet those conditions, so their intuitions are far less trustworthy than they feel.
Content
Memory, experience, and the happiness trap
Kahneman opens by distinguishing experienced happiness — how good a moment feels as it is lived — from evaluative satisfaction: the retrospective sense of how one’s life has gone. The two point in different directions. Asked about his own life, he answers drily: ‘Neither.’ He suspects most people optimise for memories rather than moments because memories are all we keep; the experience vanishes in real time. This explains why endings and peak intensity dominate evaluations far more than duration does — intensity is evolutionarily informative (it signals severity of threat or reward); duration is not.
On vacations as investments in memory-formation, he observes that people tend to place the peak near the end, implicitly aware of the peak-end rule. But he doubts most happiness research translates cleanly into behaviour: interventions that tell people how to become happier are hard to evaluate because subjects who know they are in a study cannot answer honestly, and their friends — the only valid raters — are never asked.
Bias: what it is and what it is not
Kahneman resists Tyler Cowen’s attempt to frame everything as bias. Sports partisanship is not a bias — it is identification, an emotion. Music as cinematic backdrop is not biasing — it completes the experience. Loss aversion shrinks with expertise and routine: seasoned traders shed the endowment effect that novices show, confirming John List’s research on expert behaviour.
Where bias does operate, he is sceptical that insight alone removes it. Knowing intellectually that ‘two hours late’ and ‘one minute late’ produce the same outcome does not abolish the emotional sting of the near-miss; it would ‘take a lot of work’ to override it, not a single resolved decision.
Noise: the forgotten source of error
This is where Kahneman is most animated in the conversation. Noise — the technical term for random variability in judgements that should be identical — is distinct from bias, which is a systematic tilt in one direction. An archer whose shots cluster consistently to the left has bias; one whose shots scatter randomly has noise. Both reduce accuracy; the procedures for correcting them differ entirely.
The insurance-underwriter study crystallised the problem. Fifty specialists each priced the same routine case. Executives, when asked to forecast how much two underwriters would differ on average, said about 10%. The actual figure was 50%. That means half the signal in any single underwriter’s premium was pure noise — entirely wasted analytical effort. He finds comparable variability in claims assessors, judges, and essay graders, and suspects it is near-universal wherever humans exercise judgement.
Within individuals, noise is equally striking: the same person is measurably more lenient before lunch than after, and more lenient on a cool day than a hot one. He draws the practical implication clearly: reducing noise and reducing bias are independent levers, and most organisations only work the second.
Delay intuition: advice for CEOs
Asked what he would tell a CEO facing non-routine decisions — where no panel of twelve equals can be averaged — Kahneman recommends ‘divide and conquer.’ Assess each dimension of the problem separately and independently before forming an overall view. Intuition forms fast and colonises subsequent reasoning; the discipline is to deny it that foothold until the information is complete.
He is careful to distinguish this from paralysis by analysis: ‘It’s not so much a matter of time… it’s a matter of planning how you’re going to make the decision, and making it in stages.’ The underlying principle is that independent assessments of sub-components contain more usable signal than a single holistic impression.
When to trust intuition: the Klein conditions
Gary Klein — Kahneman names him as a friend sitting in the audience and ‘violently opposed’ to the delay-intuition view — provides the necessary corrective. Expert intuition, of the kind firefighters or chess masters demonstrate, is genuine and reliable. The conditions that produce it are: a regular environment (one where patterns recur), sufficient experience, and rapid, unambiguous feedback. Spouses recognising the emotion in each other’s voice over a telephone meet all three. Most corporate decisions do not.
Kahneman and Klein spent six years mapping this boundary and published a joint synthesis. His summary: ‘If those conditions do not develop, I don’t think we can trust people who say that they’re experts.‘
Superforecasting and the wisdom of aggregation
On Philip Tetlock’s superforecasting research, Kahneman is admiring. The basic finding — that numerate, open-minded, information-hungry individuals, selected by results over a year, outperform CIA analysts on medium-term geopolitical questions — he calls ‘proved beyond a shadow of a doubt.’ The mechanism is not exotic: they treat each question as an instance of a reference class, then switch to inside-view details, adopt multiple perspectives, and aggregate across them. No averaging per se — but disciplined multi-angle attention.
He contrasts this with the wisdom of crowds, which works only when individual errors are independent. When a shared bias is present, pooling amplifies it: everyone’s confidence grows, and the bias wins. Aggregation fixes noise; it cannot fix systematic error unless the error-free participants have a way to put more weight on their own correct assessment.
AI, automation, and the foreclosure of jobs
Kahneman is blunt on AI and labour. He cites chess as the template: Kasparov fell to a program in 1997, and for a while human-computer teams outperformed either alone. That phase is over — ‘the programs do not need the grand masters.’ Dermatology diagnosis is already better done by algorithms than by physicians. He expects the same trajectory across most judgement-intensive professions, including essay grading: ‘There is so much noise in essay grading that it’s quite easy to imagine a program that would look at various indices and that would do better than hurried and tired professors.’
On human-override of algorithmic decisions, he offers a clear rule of thumb: override is justified only when genuinely new, out-of-sample information arrives that the model has never seen — the banker who discovers that the loan applicant just been charged with fraud. Otherwise, overriding formulas systematically degrades decisions.
The replication crisis and System 1 / System 2
Asked about results in Thinking, Fast and Slow that failed to replicate, Kahneman is disarmingly candid: ‘There were whole sets of results that I published in Thinking, Fast and Slow that I wish I hadn’t published because they’re not reliable.’ He is unbothered, however, about the core distinction between automatic, associative processing (System 1 — the kind that gives you 2+2 without effort) and deliberate, effortful reasoning (System 2 — the kind needed for 17×24). That phenomenological contrast is not a claim subject to replication; it is an observable feature of cognition.
On psychology’s broader recovery from the crisis: ‘Within a decade, psychology has changed… it clearly is a better science than it was 10 years ago because of the replication crisis.‘
The Tversky collaboration and the Noise book
On what is undervalued about Amos Tversky — beyond what Michael Lewis captured in The Undoing Project — Kahneman singles out ‘the mental energy, just the joy of thinking.’ And Tversky’s humour mattered substantively: their habit of laughing at the stupidities they discovered in their own thinking gave their work its characteristic ironic tone and made the subject matter ‘virtually impossible in other fields’ to treat with the same lightness.
The forthcoming Noise (co-authored with Cass Sunstein and a former McKinsey director) will argue that noise is underestimated and that the right corrective is to induce greater uniformity of process — structuring decisions to reduce vulnerability to ‘all sorts of irrelevant influences.’ He frames the deeper theoretical claim as the interplay between statistical thinking (which noise demands) and causal thinking (which bias invites), two modes he sees as fundamentally in tension.
Related
- Daniel Kahneman — speaker; psychologist and Nobel laureate
- Tyler Cowen — host
- Deciding Under Uncertainty — theme; noise and bias as the two independent sources of error in judgement
- Rationality — concept; Kahneman’s argument that consistency alone is an insufficient normative principle for a finite mind