Cognitive biases are systematic, predictable errors in human judgment that arise from mental shortcuts called heuristics. First mapped by psychologists Daniel Kahneman and Amos Tversky in the 1970s, researchers have since cataloged over 180 distinct biases affecting memory, perception, and decision-making. Understanding these biases is critical not only for personal choices but also for designing fairer AI systems, smarter trading strategies, and more resilient institutions.
What Are Cognitive Biases and Why Should You Care?
A cognitive bias is a repeatable, directional deviation from rational judgment. Unlike random mistakes, biases are systematic — they push entire populations toward the same wrong conclusions under the same conditions. That predictability is what makes them both dangerous and, paradoxically, useful: once you know the pattern, you can engineer around it.
The term gained mainstream traction after Amos Tversky and Daniel Kahneman published their landmark 1974 paper in Science, identifying three core heuristics — representativeness, availability, and anchoring — that produce reliable errors in probability estimation. Their research program, known as “heuristics and biases,” demonstrated that even trained statisticians fall into these traps when relying on intuition rather than formal analysis. Kahneman’s 2002 Nobel Prize in Economics cemented the field’s importance far beyond psychology.
Today the count of documented biases exceeds 180, though the exact number depends on taxonomy. Some biases — like loss aversion, central to prospect theory — reshape entire economic models. Others, like the IKEA effect (overvaluing things you built yourself), are narrower but equally replicable. The important question is never “how many biases exist?” but rather “which ones are costing me money, accuracy, or clarity right now?”
Where Do Cognitive Biases Come From? The Dual-Process Model
Kahneman’s Thinking, Fast and Slow (2011) popularized the framework that explains why biases exist at all. The human brain runs two parallel processing systems:
System 1 operates automatically, fast, with little effort and no sense of voluntary control. It handles face recognition, basic arithmetic (2+2), and emotional reactions. System 1 produces the heuristics that generate biases — it trades accuracy for speed because, in evolutionary terms, a fast approximate answer beats a slow precise one when a predator is approaching.
System 2 allocates attention to effortful mental activities, including complex computations. It is slow, deliberate, and energy-expensive. System 2 can override System 1, but it is lazy — it defaults to accepting System 1’s output unless something triggers active engagement.
Cognitive biases are the price we pay for System 1’s efficiency. They are not malfunctions. They are the predictable outputs of an optimization function that evolved to maximize survival, not statistical accuracy. The challenge in modern environments — financial markets, medical diagnosis, AI alignment — is that the survival heuristics misfire when the decision context is abstract, probabilistic, or high-dimensional.
The 12 Most Dangerous Cognitive Biases — A Practitioner’s Breakdown
Listing 180+ biases is easy; knowing which ones actually damage your decisions is harder. The following 12 are selected not for popularity but for measurable impact on financial decisions, AI system design, and everyday reasoning. Each includes the original research context, a real-world failure case, and a specific debiasing technique.
1. Anchoring Bias
Definition: The tendency to rely disproportionately on the first piece of information encountered (the “anchor”) when making subsequent judgments, even when the anchor is arbitrary.
Original evidence: In Tversky and Kahneman’s 1974 experiment, participants watched a roulette wheel land on either 10 or 65 before estimating the percentage of African nations in the United Nations. The group anchored to 10 guessed 25%; the group anchored to 65 guessed 45%. The random number — completely irrelevant — shifted estimates by 20 percentage points.
Why it matters in trading: Every price target, analyst forecast, and entry-level discussion is an anchor. When you read “Gold target: $2,800” before analyzing the chart, your trading decision is already contaminated. From my own CFD trading on Plus500, I’ve learned to analyze the chart structure — fair value gaps, liquidity sweeps, break of structure — before checking any external price targets. The sequence matters enormously.
Debiasing technique: Before making any quantitative estimate, generate your own number first. Write it down. Only then consult external sources. This simple “anchor-first” protocol reduces anchoring effects by up to 40% in laboratory settings.
2. Confirmation Bias
Definition: The tendency to search for, interpret, and remember information in ways that confirm pre-existing beliefs, while dismissing or underweighting contradictory evidence.
Scale of the problem: Confirmation bias is arguably the most pervasive cognitive bias in human reasoning. It operates at three distinct stages: selective search (what information you look for), selective interpretation (how you read ambiguous evidence), and selective memory (what you remember afterward). Each stage independently pushes you toward your prior belief.
Real-world failure: The 2008 financial crisis was partly driven by confirmation bias at institutional scale. Rating agencies, banks, and regulators all had access to data suggesting that subprime mortgage defaults were rising — but each interpreted the data through the lens of “housing prices always go up,” a belief reinforced by decades of upward trends. Contradictory signals were systematically explained away.
Debiasing technique: Actively seek disconfirming evidence. Before committing to any thesis — investment, hiring decision, diagnosis — explicitly write down three reasons you could be wrong. The intelligence community calls this “Analysis of Competing Hypotheses” (ACH), and the CIA has used it formally since the 1990s.
3. Availability Heuristic
Definition: Estimating the probability of an event based on how easily examples come to mind, rather than on actual statistical frequency.
Original evidence: Kahneman and Tversky showed participants lists containing equal numbers of men and women. When the men were famous (and therefore more memorable), participants consistently overestimated the proportion of men on the list. Ease of recall was mistaken for frequency.
Modern example: After every plane crash, air travel anxiety spikes — despite aviation remaining statistically the safest form of transportation. Meanwhile, car accidents (far more deadly, far less newsworthy) generate minimal behavioral change. Media coverage acts as an availability amplifier, making rare dramatic events feel common and common undramatic risks feel negligible.
Debiasing technique: When estimating risk or frequency, always ask: “Am I recalling data or recalling headlines?” Replace anecdotes with base rates. If you’re assessing whether a particular machine learning model is reliable, look at benchmark datasets — not at the one viral failure case on social media.
4. Loss Aversion
Definition: Losses are psychologically approximately twice as painful as equivalent gains are pleasurable. A $100 loss hurts more than a $100 gain satisfies.
This bias sits at the core of prospect theory, Kahneman and Tversky’s most influential theoretical contribution. The asymmetry between gains and losses explains a wide range of observed behaviors: why people hold losing stocks too long (hoping to avoid realizing the loss), why insurance is systematically overpriced (people overpay to eliminate small risks), and why investors make irrational choices under market stress.
Trading reality check: In my CFD trading, loss aversion is the single hardest bias to manage. When a position moves against you by 3%, every fiber of your brain screams “it’ll come back” — even when the Smart Money Concepts structure (broken BOS, filled FVG) clearly says exit. The mathematical solution is simple: set stop-losses before entering. The psychological challenge is not moving them once the loss becomes real.
Debiasing technique: Pre-commit to exit criteria before entering any position. Write them down. Use automated stop-losses. The goal is to make the exit decision when you’re emotionally neutral, not when you’re watching P&L swing red.
5. Dunning-Kruger Effect
Definition: People with limited knowledge or competence in a domain tend to overestimate their own ability, while experts tend to underestimate theirs.
Nuance often missed: The Dunning-Kruger effect is frequently oversimplified into “stupid people don’t know they’re stupid.” The actual research (Kruger & Dunning, 1999) showed something more specific: poor performers lacked the metacognitive skills to recognize the gap between their performance and competence. Meanwhile, top performers assumed others found the task equally easy and thus underrated their relative standing.
Why it matters in AI: When non-technical executives evaluate AI capabilities, the Dunning-Kruger effect is rampant. They overestimate what a model can do (“just make it understand context”) and underestimate what it requires (clean data, rigorous evaluation, ongoing monitoring). This bias is a primary driver of failed AI projects — the confidence-competence gap produces unrealistic timelines and underresourced implementations.
6. Survivorship Bias
Definition: Concentrating on entities that passed a selection process and ignoring those that did not, leading to systematically optimistic conclusions.
Classic case: During World War II, the U.S. military examined returning bombers to determine where to add armor. The initial analysis suggested reinforcing the areas with the most bullet holes. Statistician Abraham Wald pointed out the critical error: the planes that made it back are the ones that survived despite those hits. The missing data — planes that were shot down — was where the lethal vulnerabilities actually were. Armor should go where the returning planes had no holes.
In investing and trading: When someone says “Warren Buffett proves that value investing works,” they’re ignoring the thousands of value investors who used the same approach and underperformed the market. Mutual fund advertisements showing 10-year returns represent funds that survived — the underperformers were quietly closed and merged. Any backtest that doesn’t account for delisted or bankrupt assets is contaminated by survivorship bias.
7. Framing Effect
Definition: People react differently to the same information depending on how it is presented — as a gain or as a loss, as a percentage or absolute number.
Classic demonstration: Kahneman and Tversky’s “Asian Disease Problem” (1981) presented identical outcomes in two frames. When told “200 of 600 people will be saved,” 72% of participants chose this option. When the same outcome was framed as “400 of 600 people will die,” only 22% chose it. Same math, opposite preferences — the frame flipped the majority response by 50 percentage points.
In AI and policy: The EU AI Act frames AI regulation as “risk management” rather than “innovation restriction” — and this framing choice significantly affects public and industry perception. Similarly, presenting an AI model’s accuracy as “95% correct” versus “wrong 1 in 20 times” creates entirely different impressions despite being mathematically identical. When designing AI systems, the framing of outputs to end users is itself a bias introduction vector.
8. Hindsight Bias
Definition: After an event has occurred, perceiving it as having been predictable — the “I-knew-it-all-along” effect.
Why it’s dangerous: Hindsight bias corrupts learning. If every outcome feels predictable in retrospect, you never update your decision-making process — because you believe your process was right all along. In financial markets, this manifests as traders who confidently explain why the crash “was obvious” while ignoring that they didn’t act on that supposed knowledge before the crash.
Debiasing technique: Keep a decision journal. Before outcomes are known, write down your prediction, your confidence level (e.g., “70% sure gold breaks $2,700 by Friday”), and your reasoning. After the outcome, compare your recorded prediction to what you actually expected. The gap between the journal and your memory is the size of your hindsight bias.
9. Sunk Cost Fallacy
Definition: Continuing a behavior or endeavor because of previously invested resources (time, money, effort) rather than future expected value.
This isn’t purely irrational: While economists model sunk costs as irrelevant to forward-looking decisions, psychological research suggests the effect is partly explained by loss aversion (abandoning a project = realizing the loss) and partly by identity commitment (“I’m the kind of person who finishes what I start”). Both mechanisms are understandable but lead to demonstrably worse outcomes when the project’s expected value has genuinely turned negative.
In software and AI projects: The sunk cost fallacy explains why organizations continue pouring resources into failing AI implementations. “We’ve already spent $2 million on this RAG pipeline” is not a reason to spend $500K more if the architecture is fundamentally wrong. The rational question is always: “Given what I know now, would I start this project from scratch?” If no — kill it.
10. Bandwagon Effect
Definition: The tendency to adopt beliefs, ideas, or behaviors proportionally to how many other people already hold them, regardless of underlying evidence.
In markets: Every bubble in financial history — tulips, dot-com, crypto — was amplified by the bandwagon effect. Social proof overrides individual analysis. When “everyone is buying,” the psychological cost of not buying (FOMO) exceeds the analytical cost of a bad investment. This is why Smart Money Concepts in trading explicitly track where retail liquidity accumulates — because the bandwagon creates the liquidity pools that institutional players exploit.
In AI: The bandwagon effect drives technology adoption waves. Companies implement large language models not because rigorous analysis shows positive ROI for their specific use case, but because competitors are doing it and board presentations now require an “AI strategy” slide. The bias produces premature adoption, misaligned implementations, and eventually the disillusionment phase of the hype cycle.
11. Overconfidence Bias
Definition: Systematic overestimation of the accuracy of one’s own judgments, predictions, or abilities.
Quantified: In calibration studies, when people say they are “90% confident” in an answer, they are typically correct only 70–80% of the time. This 10–20 percentage point gap between confidence and accuracy is remarkably stable across domains, cultures, and expertise levels. Even experts are overconfident — they’re just overconfident about harder questions.
In trading: Overconfidence is the number-one predictor of excessive trading frequency. Barber and Odean’s (2000) study of 66,465 brokerage accounts showed that the most active traders earned annual returns 6.5 percentage points lower than the least active. They traded more because they believed — incorrectly — that they could time the market. More trades = more opportunities for overconfidence to destroy capital.
Debiasing technique: Practice probabilistic calibration. Instead of saying “I think gold will go up,” say “I assign 65% probability to gold rising above $2,750 by Friday.” Track these predictions over time. Your calibration score — the correlation between your confidence and actual outcomes — reveals how accurately you understand your own uncertainty.
12. Status Quo Bias
Definition: Preference for the current state of affairs, treating any change as a loss — even when the alternative is objectively superior.
Mechanism: Status quo bias combines loss aversion (change = potential loss) with the endowment effect (overvaluing what you already have) and omission bias (inaction feels less risky than action, even when the expected values are identical). Together, these create powerful inertia against rational updating.
In AI adoption: Organizations resist replacing legacy systems even when modern alternatives are demonstrably better. The status quo — a 15-year-old data pipeline — is “known.” The alternative — a modern vector database architecture — is “unknown.” The bias systematically overweights the risks of change and underweights the ongoing costs of stagnation.
The 4 Categories of Cognitive Biases: A Taxonomy
Buster Benson’s widely referenced cognitive bias codex organizes over 180 biases into four categories based on the underlying problem each bias attempts to solve. This taxonomy is more useful than alphabetical lists because it reveals why each bias exists — which directly informs how to counter it.
| Category | The Problem | What the Brain Does | Key Biases | The Cost |
|---|---|---|---|---|
| Too Much Information | We’re bombarded with data and need to filter | Notices things that are salient, repeated, or anchored | Availability, anchoring, attentional bias, frequency illusion | We miss important data that doesn’t “pop” |
| Not Enough Meaning | The world is ambiguous and we need stories | Fills gaps with stereotypes, assumptions, and patterns | Confirmation bias, bandwagon, stereotyping, halo effect | We construct narratives that feel true but aren’t |
| Need to Act Fast | We’re time-constrained and must decide now | Favors immediate, low-effort options | Status quo, sunk cost, loss aversion, hyperbolic discounting | We choose quick over optimal |
| What Should We Remember? | Memory is limited; we can’t store everything | Saves generalizations, discards specifics | Hindsight bias, false memory, peak-end rule, leveling/sharpening | Memories are reconstructions, not recordings |
Cognitive Biases in AI: When Machines Inherit Human Errors
A common misconception is that AI eliminates cognitive biases. The reality is more nuanced: AI systems can inherit, amplify, and even create new forms of bias that have no direct human equivalent.
How Human Biases Enter AI Systems
AI bias enters the pipeline at multiple stages. Training data bias occurs when the dataset reflects historical human decisions — including all their embedded biases. A hiring algorithm trained on a decade of resume screening data will learn the biases of the human reviewers who did the screening. Label bias occurs when the humans who annotate training data bring their own cognitive biases to the labeling process. Selection bias occurs when the data collected doesn’t represent the population the model will serve.
The EU AI Act, which entered its full enforcement phase in 2025, explicitly requires bias auditing for high-risk AI systems across healthcare, finance, law enforcement, and employment. Organizations deploying AI in these domains must now document their bias mitigation strategies — a regulatory acknowledgment that cognitive biases don’t disappear when you digitize them.
LLMs and Cognitive Bias: A Mirror Problem
Large language models present a unique challenge: because they’re trained on vast corpora of human text, they absorb the statistical fingerprints of every human cognitive bias embedded in that text. A 2025 study published in Frontiers in Big Data identified that LLMs exhibit measurable versions of anchoring, framing effects, and confirmation-like patterns in their outputs. The models don’t “have” biases in the human psychological sense — they reproduce the distributional patterns of biased text.
This has practical implications for anyone using AI in decision-making. If you ask an LLM for investment advice, its response reflects the aggregate biases of its training corpus — which is overwhelmingly bullish (because published financial commentary skews optimistic) and recency-biased (because recent events are overrepresented). Understanding cognitive biases in the human context is now a prerequisite for understanding AI limitations, not an alternative to it.
Human cognitive biases arise from evolutionary heuristics. AI biases arise from training data distributions. The mechanisms are completely different, but the observable outputs can look similar. Debiasing AI requires statistical techniques (reweighting, adversarial training, fairness constraints). Debiasing humans requires cognitive techniques (pre-commitment, calibration, structured analysis). Using the wrong type of intervention on the wrong type of system doesn’t work.
Cognitive Biases in Trading and Finance: Where the Stakes Are Measured in Dollars
Behavioral finance exists as a field precisely because cognitive biases make markets deviate from the “efficient market” predictions of classical economics. Here’s how the most damaging biases manifest in real trading scenarios:
Anchoring + Recency bias = traders fixate on recent highs/lows as reference points, ignoring broader structural levels. When gold drops from $2,850 to $2,720, the “$2,850” anchor makes $2,720 feel cheap — even if structural analysis shows support at $2,650.
Confirmation bias + Overconfidence = traders seek information confirming their existing position. If you’re long, you read bullish analyses. If you’re short, you read bearish ones. This creates an echo chamber around your position that reinforces it regardless of incoming evidence.
Loss aversion + Sunk cost = the classic “adding to a loser” pattern. The position is underwater, the loss hurts, so rather than cutting the loss, the trader adds capital — hoping the average entry price moves closer to current price. This turns a manageable loss into a catastrophic one.
Survivorship bias + Bandwagon = new traders follow “guru” accounts who show winning streaks (survivorship) and build followings based on recent performance (bandwagon). The invisible graveyard of failed traders — who used the same strategies — never appears in the feed.
After 2+ years of CFD trading on Plus500 using Smart Money Concepts, I’ve found exactly three debiasing techniques that actually survive real-money pressure: (1) pre-written trade plans with entry, stop-loss, and target defined before market open; (2) a trade journal reviewed weekly to catch recurring bias patterns; (3) a 24-hour rule — never enter a position within 24 hours of a large loss. Everything else is theory until you test it with capital at risk.
How to Actually Reduce Cognitive Biases: 7 Evidence-Based Techniques
Research consistently shows that merely knowing about biases does not significantly reduce them. The meta-finding across debiasing studies is clear: changing the decision environment works better than trying to change the decision-maker. Here are the seven techniques with the strongest empirical support:
1. Pre-commitment protocols. Decide your criteria before you encounter the data. Set stop-losses before entering trades. Define hiring criteria before reading resumes. Write evaluation rubrics before grading essays. Pre-commitment removes the opportunity for biases to influence judgment at the point of decision.
2. Consider-the-opposite. Before finalizing any conclusion, explicitly argue for the opposite position. Research by Mussweiler et al. (2000) showed this simple instruction reduces anchoring effects significantly. The key is genuine engagement — actually constructing the counter-argument, not just acknowledging it exists.
3. Base rate inclusion. For any probability estimate, start with the base rate (how common is this thing in general?) before adjusting for specific evidence. Most probability estimation errors stem from ignoring base rates entirely and relying solely on case-specific information.
4. Red team / pre-mortem. Before launching a project, imagine it has already failed spectacularly. Then work backward to identify what caused the failure. This “prospective hindsight” technique (Klein, 2007) increases the identification of potential problems by 30% compared to simply asking “what could go wrong?”
5. Structured decision-making. Replace unstructured deliberation with checklists, scoring matrices, or decision trees. Commercial aviation’s safety revolution was built not on making pilots smarter but on giving them checklists that bypass common cognitive errors. The same principle applies to any high-stakes domain.
6. Probabilistic calibration training. Practice making explicit probability estimates and track your accuracy over time. Superforecasters — the top performers in Philip Tetlock’s prediction tournaments — aren’t smarter than average. They’re better calibrated: their stated confidence levels more accurately reflect actual probabilities.
7. Environmental restructuring. Design the decision environment to make biased choices harder. Nudge theory applies behavioral insights to architecture: default options, choice ordering, information presentation. This works because it targets the structure rather than the individual, and it’s one of the core insights of the behavioral economics movement pioneered by behavioral economics.
Cognitive Biases vs. Logical Fallacies: What’s the Difference?
These terms are frequently confused but refer to fundamentally different phenomena. A cognitive bias is a systematic pattern of deviation from norm or rationality in judgment — it’s a psychological phenomenon rooted in how the brain processes information. A logical fallacy is an error in the structure of an argument — it’s a formal or informal reasoning mistake that weakens a conclusion.
Confirmation bias (cognitive bias) makes you selectively seek evidence for your existing belief. An appeal to authority (logical fallacy) is citing an expert’s opinion as proof when the expert’s expertise is irrelevant to the claim. Both can lead to wrong conclusions, but through different mechanisms: one is a perceptual/memory error, the other is a structural argument error.
In practice, they often compound. A trader influenced by loss aversion (bias) might construct an ad hoc rationalization (fallacy) for why a losing position will recover. The bias creates the motivation; the fallacy provides the intellectual cover. Recognizing both — and their interaction — is essential for any serious decision-maker.
The Future of Debiasing: AI as Both Problem and Solution
The most interesting frontier in cognitive bias research is the feedback loop between human biases and AI systems. Humans with cognitive biases build AI systems. Those AI systems, deployed at scale, amplify the biases into societal outcomes. Those societal outcomes become the training data for the next generation of AI systems.
Breaking this loop requires intervention at multiple points: pre-processing (debiasing training data before it enters the model), in-processing (embedding fairness constraints directly in the training objective), and post-processing (adjusting model outputs to meet fairness criteria). The technical toolkit is expanding rapidly — adversarial debiasing, counterfactual fairness, causal inference methods — but the fundamental challenge remains: you can’t define “fair” without making value judgments that are themselves subject to cognitive biases.
For practitioners — traders, developers, analysts — the actionable takeaway is this: cognitive biases are not a knowledge problem. They’re an environment design problem. You will not think your way out of biases by trying harder. You will structure your way out of them by changing the conditions under which you make decisions. Checklists beat willpower. Automation beats discipline. And understanding the mechanism beats memorizing the name.
FAQ
What is the most common cognitive bias?
Confirmation bias is widely considered the most pervasive cognitive bias. It operates at three stages — search, interpretation, and memory — making it exceptionally difficult to detect in your own thinking. Research suggests virtually every person exhibits confirmation bias to some degree, regardless of intelligence or expertise.
How many cognitive biases are there?
Researchers have documented over 180 distinct cognitive biases, though the exact count depends on taxonomy. Some catalogs list 150+, others exceed 200. The Cognitive Bias Codex, a widely referenced visual taxonomy, organizes them into four categories: too much information, not enough meaning, need to act fast, and memory limitations.
Can cognitive biases be eliminated?
No. Cognitive biases are inherent products of how the human brain processes information under constraints. They cannot be fully eliminated, but they can be systematically reduced through environmental restructuring, pre-commitment protocols, calibration training, and structured decision-making frameworks. Research shows that changing the decision environment is more effective than trying to change the decision-maker.
What is the difference between a cognitive bias and a heuristic?
A heuristic is a mental shortcut or rule of thumb that the brain uses to make quick judgments. A cognitive bias is the systematic error that results when a heuristic misfires in specific contexts. For example, the availability heuristic (judging frequency by ease of recall) is the shortcut; the resulting overestimation of dramatic but rare events (like plane crashes) is the bias.
Do AI systems have cognitive biases?
AI systems do not experience biases psychologically, but they can exhibit bias-like patterns in their outputs because they learn from human-generated data that contains embedded biases. LLMs trained on internet text reproduce the distributional patterns of biased content. Debiasing AI requires technical interventions (adversarial training, fairness constraints, data reweighting) rather than the cognitive techniques used for human debiasing.
How do cognitive biases affect trading and investing?
Cognitive biases are a primary source of systematic trading errors. Loss aversion causes traders to hold losing positions too long. Overconfidence leads to excessive trading frequency. Anchoring bias distorts price target assessments. Survivorship bias makes successful strategies appear more reliable than they are. Behavioral finance research shows that bias-aware traders who use pre-commitment protocols and decision journals significantly outperform those who trade on intuition alone.
Who discovered cognitive biases?
The modern study of cognitive biases was pioneered by psychologists Daniel Kahneman and Amos Tversky through their “heuristics and biases” research program, beginning in the late 1960s. Their seminal 1974 paper in Science identified three foundational heuristics — representativeness, availability, and anchoring — that produce systematic errors. Kahneman received the Nobel Prize in Economics in 2002 for this work. However, earlier researchers including Herbert Simon (bounded rationality, 1950s) laid the theoretical groundwork.
Bibliography
- Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases. Science, 185(4157), 1124–1131. https://doi.org/10.1126/science.185.4157.1124
- Kahneman, D. (2011). Thinking, Fast and Slow. Farrar, Straus and Giroux.
- Tversky, A., & Kahneman, D. (1981). The framing of decisions and the psychology of choice. Science, 211(4481), 453–458. https://doi.org/10.1126/science.7455683
- Kruger, J., & Dunning, D. (1999). Unskilled and unaware of it: How difficulties in recognizing one’s own incompetence lead to inflated self-assessments. Journal of Personality and Social Psychology, 77(6), 1121–1134. https://doi.org/10.1037/0022-3514.77.6.1121
- Barber, B. M., & Odean, T. (2000). Trading is hazardous to your wealth: The common stock investment performance of individual investors. The Journal of Finance, 55(2), 773–806. https://doi.org/10.1111/0022-1082.00226
- Klein, G. (2007). Performing a project premortem. Harvard Business Review, 85(9), 18–19.
- Tetlock, P. E., & Gardner, D. (2015). Superforecasting: The Art and Science of Prediction. Crown.
- Ahmad, A., Vallès, M., & Idaghdour, Y. (2026). Bias in AI systems: Integrating formal and socio-technical approaches. Frontiers in Big Data, 8, 1686452. https://doi.org/10.3389/fdata.2025.1686452
- Kumar, A., Dhanka, S., Sharma, A., et al. (2025). A comprehensive review of bias in AI, ML, and DL models: Methods, impacts, and future directions. Archives of Computational Methods in Engineering. https://doi.org/10.1007/s11831-025-10483-6
- Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis of decision under risk. Econometrica, 47(2), 263–291. https://doi.org/10.2307/1914185
- European Parliament and Council. (2024). Regulation (EU) 2024/1689 laying down harmonised rules on artificial intelligence (AI Act). Official Journal of the European Union. https://eur-lex.europa.eu/eli/reg/2024/1689
- Simon, H. A. (1955). A behavioral model of rational choice. The Quarterly Journal of Economics, 69(1), 99–118. https://doi.org/10.2307/1884852
[…] foundation underneath all of this is the same body of research we cover in our pieces on cognitive biases and behavioral finance: humans are not the perfectly rational homo economicus of textbook models. […]
[…] age groups, experts, and, as of 2025, large language models. If you’ve read our explainer on cognitive biases, anchoring is the textbook example of a System 1 heuristic in Kahneman’s dual-process […]
[…] For wider context see the disposition effect, anchoring, market equilibrium and cognitive biases. […]
[…] 2021 exposed how thinly capitalised the firm was relative to clearing demands. The UI’s cognitive-bias profile — confetti, push notifications, free-stock lottery — is gamification-by-design, and the […]