A Cognitively Inspired Heuristic for Two-armed Bandit Problems: The Loosely Symmetric (LS) Model

https://doi.org/10.1016/j.procs.2013.10.043Get rights and content
Under a Creative Commons license
open access

Abstract

We examine a model of human causal cognition, which generally deviates from normative systems such as classical logic and probability theory. For two-armed bandit problems, we demonstrate the efficacy of our loosely symmetric model (LS) and its implementation of two cognitive biases peculiar to humans: symmetry and mutual exclusivity. Specifically, we use LS as a simple value function within the framework of reinforcement learning. The resulting cognitively biased valuations precisely describe human causal intuitions. We further show that operating LS under the simplest greedy policy yields superior reliability and robustness, even managing to overcome the usual speed-accuracy trade-off, and effectively removing the need for parameter tuning.

Keywords

reinforcement learning
causal induction
biconditional reading
symmetry
mutual exclusivity
n-armed bandit problem
exploration– exploitation dilemma
speed-accuracy tradeoff

Cited by (0)

Selection and peer-review under responsibility of the Program Committee of IES2013.