Skip to Main Content
Summary form only given. Artificial intelligence applications, particularly those involving natural language understanding, are actually less ambitious than they, were decades ago. Statistical and machine learning techniques have had great success on tasks that can be treated without understanding, but there are many important areas that require semantic systems. The key to building more powerful AI applications is to model the world knowledge and the linguistic and other basic abilities that people bring to bear. We now know that these abilities can not be fully expressed in abstract formalisms, but require models that map onto human biology and behavior. Cognitive science is the field that is best placed to unite the theory and applications of intelligence. However, as with other fields, there is a tendency to specialize in a narrow sub-domain. This talk will make the case that a unified cognitive science is now possible, based on rapid advances in all the relevant disciplines, including computational modeling, and that there are already applications of this more cognitive approach to AI.