Notes on Computational Rationality

Computational Rationality

  • Definition: Computational rationality refers to the study of how agents make optimal decisions under constraints of limited computational resources.

Hierarchical Structure

  • Models: Three primary types of models are highlighted:
    • Computational Models: Mathematical models using computational methods for simulating systems.
    • Computational Cognitive Models: Designed to simulate human cognitive processes.
    • Computational Rationality Models: Focus on predicting optimal decision-making strategies with limited resources.

Computational Models

  • Core Concept: General models utilizing computational methods applicable across various fields, such as:

    • Cognitive science
    • Biology
    • Physics
    • Economics

  • Characteristics:

    • Not restricted to cognitive science; can simulate weather or financial models.
    • Employ multiple methods such as differential equations, reinforcement learning, and machine learning.
    • Aim to simulate system behavior rather than optimizing computational strategies.

  • Examples:

    • Weather models use numerical methods for predicting changes in weather patterns.

Computational Cognitive Models

  • Core Concept: Focus on simulating human cognitive processes, including perception, memory, reasoning, learning, and decision-making.
  • Characteristics:
    • Aim to understand cognition rather than optimizing computational processes.
    • Include symbolic models (e.g., ACT-R), neural networks, and Bayesian cognitive models.
    • Address psychological mechanisms in language comprehension and problem-solving.
  • Examples:
    • ACT-R: Models how humans remember and make decisions.

Computational Rationality Models

  • Core Concept: These models predict how agents will make decisions when faced with limited computational resources, often utilizing:
    • Bayesian reasoning
    • Reinforcement learning
    • Partially Observable Markov Decision Processes (POMDPs)
  • Characteristics:
    • Assess the impact of cognitive costs (attention, time, load) on decision-making.
    • Focus on optimizing the computational process of decision-making rather than just behavior.
  • Example: Choosing a coffee shop involves evaluating costs (time, money) against benefits (quality, enjoyment).

Choosing the Right Model

  • There are specific model choices based on research focus:
    1. Attention management falls under Computational Rationality Models.
    2. Simulating human attention without optimization is Computational Cognitive Models.
    3. Optimizing computational resources belongs to Computational Models.

Decision-Making Under Constraints

  • Traditional vs. Computational Rationality:
    • Traditional rationality focuses on optimal choices given unlimited resources.
    • Computational rationality emphasizes decision-making with limited resources.
  • Balancing Accuracy and Efficiency:
    • There's always a trade-off between decision accuracy and resource efficiency.
    • Examples include navigation systems that prioritize speed and efficiency over the best route.

Heuristics and Approximations

  • Heuristics: Computational rationality often involves mental shortcuts to simplify decisions.
  • Example: A medical AI uses common symptom patterns instead of analyzing every condition exhaustively.

Real-Time Adaptation

  • Rapid decisions are necessary in dynamic situations (e.g., autonomous vehicles). Systems must react under uncertainty.
  • Example: An autonomous vehicle makes fast decisions based on sensor data, optimizing given limited processing power.

Practical Applications

  • Fields: Computational rationality applies to AI, robotics, and HCI. It helps systems function effectively in real-world scenarios.
  • Example: A smart assistant prioritizes alerts based on context without overwhelming the user.

Model Architectures

  • Elements of model architecture include:
    • Control module: Decides movement based on beliefs about target location.
    • Motor module: Executes decisions, influenced by human-like noise.
    • Perception module: Localizes target with foveated vision, impacted by fixation.
    • Memory module: Integrates location estimates into a belief representation.

Building Computational Rational Models

  1. Define Goals: Establish clear objectives for the decision-making process.
  2. Define Environment: Simulate relevant conditions impacting decision-making.
  3. Cognitive Limitations: Consider how decision-making abilities are bounded by various factors (e.g., noise).
  4. Optimal Behavior: Determine the best actions under given constraints.
  5. Model Validity: Ensure that models align with real-world observations and predictions.

Assignment: Finding Examples of Computational Rationality in Daily Life

  • Analyze a daily decision-making problem using computational rationality principles.
    • Describe the problem, consider factors, and apply principles to evaluate decisions.
    • Report must include sections on problem description, analysis, application, and conclusion.