Zollman Bandit Model (2008 Variation)
A bandit model by Kevin Zollman exploring how network structure affects scientific communities' ability to reach accurate consensus. This is the 2008 variation focusing on division of labor and cognitive diversity. View the full paper here.
Abstract
There is growing interest in understanding and eliciting division of labor within groups of scientists. This paper illustrates the need for this division of labor through a historical example, and a formal model is presented to better analyze situations of this type. Analysis of this model reveals that a division of labor can be maintained in two different ways: by limiting information or by endowing the scientists with extreme beliefs. If both features are present however, cognitive diversity is maintained indefinitely, and as a result agents fail to converge to the truth. Beyond the mechanisms for creating diversity suggested here, this shows that the real epistemic goal is not diversity but transient diversity.
Model Implementation (2008)
This variation implements a multi-armed bandit problem where scientists:
- Choose between two methodologies (A or B) based on expected values
- Update their beliefs using Beta distributions after observing outcomes
- Share information within their network structure
Key Features
- Learning Mechanism: Bayesian updating with Beta-Binomial conjugate priors
- Decision Rule: Choose methodology with higher expected value
- Information Sharing: Agents observe neighbors' experimental outcomes
- Network Effects: Different structures (complete, cycle, wheel) affect information flow
Model Parameters
num_nodes: Size of network (default: 10)a_objective_2008: True success probability of methodology A (default: 0.19)b_objective_2008: True success probability of methodology B (default: 0.71)num_trials_per_step_2008: Number of trials per experiment (default: 5)max_prior_value_2008: Maximum value for initial Beta distribution parameters (default: 4.0)graph_type: Network structure - "complete", "cycle", or "wheel"
Agent Attributes
Each scientist maintains:
a_alpha,a_beta: Beta distribution parameters for methodology Ab_alpha,b_beta: Beta distribution parameters for methodology Ba_expectation,b_expectation: Expected success rates for each methodology