RESEARCH ARTICLE
Observing and Intervening: Rational and Heuristic Models of Causal Decision Making
Björn Meder*, 1, Tobias Gerstenberg1, 3, York Hagmayer2, Michael R. Waldmann2
Article Information
Identifiers and Pagination:
Year: 2010Volume: 3
First Page: 119
Last Page: 135
Publisher ID: TOPSYJ-3-119
DOI: 10.2174/1874350101003010119
Article History:
Received Date: 27/8/2009Revision Received Date: 11/12/2009
Acceptance Date: 7/1/2010
Electronic publication date: 13/7/2010
Collection year: 2010
open-access license: This is an open access article distributed under the terms of the Creative Commons Attribution 4.0 International Public License (CC-BY 4.0), a copy of which is available at: (https://creativecommons.org/licenses/by/4.0/legalcode). This license permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Abstract
Recently, a number of rational theories have been put forward which provide a coherent formal framework for modeling different types of causal inferences, such as prediction, diagnosis, and action planning. A hallmark of these theories is their capacity to simultaneously express probability distributions under observational and interventional scenarios, thereby rendering it possible to derive precise predictions about interventions (“doing”) from passive observations (“seeing”). In Part 1 of the paper we discuss different modeling approaches for formally representing interventions and review the empirical evidence on how humans draw causal inferences based on observations or interventions. We contrast deterministic interventions with imperfect actions yielding unreliable or unknown outcomes. In Part 2, we discuss alternative strategies for making interventional decisions when the causal structure is unknown to the agent. A Bayesian approach of rational causal inference, which aims to infer the structure and its parameters from the available data, provides the benchmark model. This account is contrasted with a heuristic approach which knows categories of causes and effects but neglects further structural information. The results of computer simulations show that despite its computational parsimony the heuristic approach achieves very good performance compared to the Bayesian model.