A Data-Acquisition Model for Learning and Cognitive Development and Its Implications for Autism

A data-driven model of learning is proposed, where a network of nodes and links is constructed that represents what has been heard and observed. Autism is viewed as the consequence of a disorder in the data-acquisition component of the model---essentially, it is the result of getting an ``inappropriate'' distribution of data. The inappropriate data distribution leads to problems in data segmentation, which, in turn leads to a poor network representation. It is shown how the model, given inappropriate data distributions, can reproduce the main cognitive deficits associated with autism, including weak central coherence, impaired theory of mind, and executive dysfunction. In addition, it is shown how the model itself can explain the inappropriate data distribution as the result of an inappropriate initial network. Finally, we discuss the relationships between our model and existing neurological models of autism, and the possible implications of our model for treatment.Halpern's work supported in part by NSF under Grant IIS-0090145, a Guggenheim Fellowship, a Fulbright Fellowship, and a grant from the NWO. Sabbatical support from CWI and the Hebrew University of Jerusalem is also gratefully acknowledged