Discovering Causal Relations by Experimentation: Causal Trees

Eighth Midwest Artificial intelligence and Cognitive Science Conference

Volume

Eighth Midwest Artificial intelligence and Cognitive Science Conference

Contents

The Controlled Lesion Method (CLM) is a set of principles for inferring the causal structure of deterministic mechanisms by experimentation. CLM formulates an important part of the common-sense logic of causation and experimentation. Previous work showed that CLM could be used to discover the structure of deterministic chains of binary variables more accurately than statistical methods; however, as implemented, CLM was prone to error when applied to causal trees. A change of knowledge representation, repla~ing atomic symbols with predicate calculus expressions for representing events, makes it possible to refine the statement of one of the principles of CLM. As a result, CLM can now discover causal tree structures correctly. This suggests that a structured representation of events may be necessary for a causal discovery system.