class 12

Defining High-Stakes Decision Environments

• Characteristics of High-Stakes Contexts: These environments are defined by decisions where the consequences are enormous, often involving the survival of individuals, the fate of fortunes, or the well-being of entire communities.

• Representative Roles: Examples include emergency room doctors diagnosing collapsing patients, jurors in capital murder cases, and pilots managing mid-flight equipment failures.

• Cognitive Battlegrounds: High-stakes scenarios represent a test for human thought due to high uncertainty and the consumption of cognitive resources by psychological pressure (ticking clocks, public scrutiny, or moral weight).

• The Problem of "Choking Under Pressure": Performance can drop precisely when needed most. Research by Sosnowski and Brosnan ($2023$) highlights this drop in mental performance under stress.

• Keinan's ($1987$) Findings: Study participants under pressure made hurried choices and overlooked obvious information merely because the stakes felt high.

• Core Challenging Elements:

  • High Uncertainty: Such as a surgeon attempting an experimental procedure or an intelligence analyst assessing a threat with incomplete data.

  • Group Dynamics: Collective judgments in crisis committees or emergency teams can be distorted by social dynamics.

  • Expertise Paradox: While experts are relied upon for their experience, they may be more susceptible to certain biases than novices (e.g., when working outside their narrow domain) and often wrongly believe they are immune to cognitive bias.

Stress, Emotion, and the Heuristic Trap

• Physiological Response: High pressure causes adrenaline floods, increased heart rate, and narrowed attentional focus.

• The System 1 vs. System 2 Shift: Based on Kahneman ($2011$), stress functionally impairs the prefrontal cortex (System 2: analytical/logical) and gives influence to the limbic system (System 1: intuitive/automatic).

• Stress-Induced Deliberation-to-Intuition (SIDI) Model: Proposed by Rongjun Yu ($2016$), this model suggests stress undermines System 2, leaving the individual dependent on automatic, intuitive impulses.

• Satisficing: Rather than optimizing, individuals under stress select the first workable option. Keinan ($1987$) found that even without strict time limits, the internal sense of urgency created by high stakes led participants to skip systematic consideration.

Specific Biases Amplified by Pressure

• Framing Effects: Stress makes individuals highly sensitive to how information is presented. A panicked official may prefer an "$85\%$ chance to save $100$ lives" over a "$15\%$ chance of $100$ deaths," despite the propositions being identical. Stress specifically amplifies loss aversion (Yu, $2016$).

• Overconfidence Paradox: Performance degrades while confidence remains high or rises. In financial trading, Barber and Odean ($2000$) found that time pressure led to more trades with higher confidence but worse outcomes.

• Confirmation Bias and Tunnel Vision: In forensic contexts, investigators may fixate on a suspect and interpret all subsequent evidence through that lens.

  • Brandon Mayfield Case ($2004$): FBI fingerprint examiners wrongly matched a print from the Madrid bombings to Mayfield. Contextual bias (his religious conversion) caused multiple experts to "see" a match that wasn't there. When evidence for the real culprit surfaced, examiners initially rejected it (belief perseverance).

  • Dror et al. ($2006$) Experiment: Fingerprint analysts were given prints they had previously identified correctly but told they were from the Mayfield case; most subsequently reversed their correct judgments due to contextual suggestion.

• Emotional Biases:

  • Fear: Leads to risk aversion and pessimism.

  • Anger: Creates a false sense of control and optimism (Lerner and Keltner, $2001$).

  • Cognitive Load: Emotions consume working memory, creating a bottleneck for complex reasoning.

Structural Solutions to Reclaim System 2 Thinking

• Sequential Unmasking (Linear Sequential Unmasking - LSU): Controlling information flow to shield analysts from biasing context until independent judgments are made.

  • In forensics, examiners analyze crime-scene evidence first without knowing suspect details (Dror and Kukucka, $2021$).

  • Studies by Taylor et al. ($2019$) show LSU results in more consistent matching decisions.

• Checklists: Offload memory requirements and enforce process discipline.

  • Atul Gawande: Introduced surgical-safety checklists globally, significantly reducing complications and deaths (Haynes et al., $2009$).

  • Psychological Function: They free mental bandwidth by automating recall of basic steps (e.g., "Administer antibiotics") and force pauses to consider alternative diagnoses.

• Metacognitive 'Pause' Mechanisms: Thinking about your thinking through "decision time-outs."

  • Yale et al. ($2022$): A study in a pediatric hospital found that diagnostic time-outs prompted alternative considerations in more than half of cases, potentially averting errors.

  • Sterile Cockpit Rule: Aviation rule prohibiting non-essential conversation during takeoff and landing to protect cognitive space.

• Blinded Second Opinions: In forensics, requiring a second expert to examine evidence independently without knowledge of the first expert's conclusion.

Collective Intelligence as a Debiasing Strategy

• Aggregation of Independent Judgments: Combining individual assessments can outperform even the best expert.

• Tangen et al. ($2020$) Fingerprint Study: Professional examiners analyzed prints individually, and their decisions were pooled via majority rule. False positives dropped by up to $8\%$ and false negatives by up to $12\%$ compared to the average individual examiner.

• Consensus Panels: Diverse experts discuss complex cases after independent review to cancel out individual random errors and biases.

Forensic and Investigative Decision-Making

• Systemic Myopia: Findley and Scott ($2006$) describe tunnel vision as a natural tendency worsened by organizational pressure.

• Steven Avery Case: Illustrates how investigators may focus on one suspect while ignoring other leads, leading to controversial convictions.

• Devil's Advocate Panels: Separate groups of investigators review major cases as if they were the defense to find holes in the primary theory.

• Fresh Eyes: Research on cold cases by Rossmo ($2009$) suggests new investigators are less wedded to original theories, yielding more breakthroughs.

• AI-Based Digital Triage: Machine-learning scans vast data (surveillance, chat logs) consistently.

  • Marturana and Tacconi ($2013$): AI correctly classified $99\%$ of seized devices.

  • Hitchcock et al. ($2016$): Implementing such systems cut forensic lab exhibits by $75\%$.

Medical High-Stakes Decisions

• Diagnostic Decision Biases:

  • Anchoring: Latching onto a first diagnosis (e.g., "heart attack") while neglecting others (e.g., "pulmonary embolism").

  • Diagnostic Momentum: Once a diagnosis is made, it passes unchallenged through the clinical chain (Smith, $2021$).

  • Availability Bias: Diagnosing based on what is top-of-mind, such as common flu during a season or a memorable rare case.

• AI in Diagnosis: McKinney et al. ($2020$) showed an AI system identifying breast cancer in mammograms with lower false-negative and false-positive rates than radiologists; the human-AI combination was the most effective.

• Structured Risk Communication: Moving from relative risk to absolute risk helps patients understand choices.

  • Relative vs. Absolute: "This drug cuts risk in half" (Relative) is less clear than "Risk reduced from $4\%$ to $2\%$ over $5$ years" (Absolute).

  • Natural Frequencies: Using formats like "$2$ out of $100$ people" (Gigerenzer and Edwards, $2003$).

• Cognitive Forcing Strategies: Tactics developed by Pat Croskerry to disrupt bias, such as forcing oneself to name at least three alternative diagnoses for every patient.

Case Study: USS Vincennes and Iran Air Flight 655

• Incident: On $3$ July $1988$, the US Navy cruiser USS Vincennes shot down Iran Air Flight 655 (civilian airliner), killing all $290$ people.

• Cognitive Factors: The crew operated under extreme stress and "scenario fulfillment" (expecting an attack and therefore seeing it). They misinterpreted data as a descending F-14 fighter jet when the Airbus A300 was actually climbing (Fogarty, $1988$).

• Automation Bias: Misinterpretation of complex data from the Aegis combat system coupled with confirmation bias led to the tragedy.

Corporate and Policy Complexity

• Normalisation of Deviance: Sociologist Diane Vaughan ($1996$) used this term to describe the Space Shuttle Challenger disaster, where small rule-bendings and ignored O-ring warnings became accepted until failure occurred.

• Groupthink: Irving Janis ($1972$) identified this in the Bay of Pigs fiasco. Features include illusions of unanimity and suppression of dissent.

• Debiasing Methods:

  • Dialectical Inquiry: Creating two teams to develop competing plans and debate their merits.

  • Pre-Mortem: Developed by Gary Klein ($2007$), where a team imagines a project has failed and works backward to identify causes. It surfaces specific risks better than simple brainstorming (Mitchell et al., $1989$).

  • Staged Decision Gates: A stage-gate process (Cooper et al., $2001$) uses independent reviewers at fixed intervals to determine if a project should continue, countering the sunk-cost fallacy and team enthusiasm.

AI as a Cognitive Partner

• Reducing Anchoring: AI can control the sequence of evidence presentation to judges or doctors.

• Bias in AI: The COMPAS algorithm used in US courts displayed racial bias, flagging Black defendants as high-risk at higher rates (Dressel and Farid, $2018$).

• Strategic Dissent: Organizations like Amazon use "Disagree and Commit" to encourage push-back during the decision phase. Red-teaming is used in military/intelligence contexts to simulate adversary thinking after the $2003$ Iraq WMD failures.