Abstract

O’Loughlin and Thagard (2000) present a specific computational implementation of the idea that the problems encountered by a child with autism in classic False Belief tasks derive from a failure to maintain coherence among multiple propositions. They argue that this failure can be explained as a structural feature of a connectionist network attempting to maintain coherence. The current paper criticizes this implementation because it falsely predicts that the same children will have a parallel problem with the False Photographs task. The fact that the content of representations makes a difference while the structure remains constant casts doubt upon their claim.