KahneBench
Ecological Validity

Ecological Domains

KahneBench tests biases across 5 real-world domains to ensure ecological validity and practical relevance.

Why Multiple Domains?

Cognitive biases don't manifest the same way in all contexts. A model might show anchoring bias in financial scenarios but resist it in medical contexts. Testing across multiple domains reveals:

  • Whether biases are context-specific or universal
  • How domain expertise affects bias susceptibility
  • Real-world implications for LLM deployment

The 5 Domains

Individual Decisions

individual

Personal finance, consumer choice, and lifestyle planning decisions

Typical Decisions

  • Investment allocation
  • Purchase decisions
  • Health and lifestyle choices
  • Career planning
  • Personal risk assessment

Example Actors

ConsumerInvestorPatientJob seeker

Professional Judgments

professional

Managerial strategy, medical diagnosis, and expert decision-making

Typical Decisions

  • Hiring decisions
  • Medical diagnoses
  • Legal case assessment
  • Business strategy
  • Performance evaluation

Example Actors

ManagerPhysicianLawyerAnalystConsultant

Social Interactions

social

Negotiation, persuasion, and collaborative problem-solving

Typical Decisions

  • Negotiation offers
  • Trust assessment
  • Team collaboration
  • Conflict resolution
  • Social influence

Example Actors

NegotiatorTeam leaderMediatorCollaborator

Temporal Decisions

temporal

Long-term planning, investment horizons, and delayed gratification

Typical Decisions

  • Retirement planning
  • Project timelines
  • Long-term investments
  • Strategic planning
  • Resource allocation over time

Example Actors

Financial plannerProject managerPolicy makerStrategic planner

Risk Assessment

risk

Evaluating uncertainty in policy, technology, and environmental domains

Typical Decisions

  • Safety protocols
  • Innovation adoption
  • Environmental policy
  • Insurance decisions
  • Technology investment

Example Actors

Risk analystPolicy makerSafety officerInsurance underwriter

Domain-Specific Testing

Each bias is tested with scenarios tailored to each domain. For example, the Anchoring Effect might be tested as:

  • Individual: Estimating the price of a car after seeing an initial figure
  • Professional: A physician estimating treatment dosage after an unrelated number
  • Social: Negotiating a salary after an initial offer
  • Temporal: Projecting retirement savings based on current balance
  • Risk: Assessing probability after an arbitrary percentage is mentioned