STSTEER

STEER

Strategic and behavioral decision-making

STEER asks whether a model reasons the way an economically rational agent would. It starts from the foundations and builds up to settings with other agents in them, and to decisions made on behalf of a group. It is described in the STEER paper (ICML 2024), and a fixed sample of its questions is published as a dataset.

Taxonomy

Browse all elements →
Setting 1Foundations

Core mathematical and cognitive skills that underlie economic reasoning: arithmetic, optimization, probability, logic, and theory of mind.

  • 1.1Arithmetic
  • 1.2Optimization
  • 1.3Probability
  • 1.4Logic
  • 1.5Theory of Mind
Setting 2Decisions in Single-Agent Environments

Preference formation and decision making when an agent has a set of alternative choices, each of which leads either to a single, deterministic outcome or to a draw from a probability distribution over outcomes.

  • 2.1Axioms of Utility in Deterministic Environments
  • 2.2Avoidance of Cognitive Biases in Deterministic Environments
  • 2.3Axioms of Utility in Stochastic Environments
  • 2.4Risk Neutral Expected Utility Computations
  • 2.5Avoidance of Cognitive Biases in Stochastic Environments
Setting 3Decisions in Multi-Agent Environments

Enriches the single-agent setting, requiring the agent to make decisions when the outcomes depend on interactions with other economic agents with their own preferences and beliefs.

  • 3.1Normal Form Games
  • 3.2Extensive Form Games
  • 3.3Imperfect Information in Extensive Form Games
  • 3.4Infinitely Repeated Games
  • 3.5Bayesian Games
Setting 4Decisions on Behalf of Other Agents

Asks the agent to aggregate the preferences of other agents to achieve good outcomes for all.

  • 4.1Axioms of Social Choice
  • 4.2Social Choice
  • 4.3Desirable Properties in Mechanism Design
  • 4.4Mechanism Design