A Wholistic Cognitive System (Seer-2) for Integrated Perception, Action and Thought
University of Wisconsin-Madison Department of Computer Sciences
MetadataShow full item record
This paper describes a computer-programmed model, SEER-2, that can perform a variety of different types of actions that span the human intellectual processes, including naming objects, describing scenes, answering simple questions, deducing solutions to simple problems, pointing to objects, and manipulating objects. Recognition, description, memory search, and deduction are all handled in the same semantic cognitive network, using the same basic type of general transformation. Thus they are all intermingled, and can conveniently interact and call upon one another, as needed. SEER-2's acts, both external and internal, are a function of a variety of influences, including the presses of internal needs and of external objects, as well as verbal commands (that is, the understood import of verbal utterances). SEER-2 is thus a first step toward an integrated wholistic cognitive system, one that is as simple and general as possible yet begins to handle a representative variety of cognitive processes. The SEER system uses a set of transforms that compose its memory network, and contain the specific information needed to handle the types of problems it will encounter - much as a parser must be given a set of rewrite rules that contain the needed information about the vocabulary and grammar of the languages to be parsed. In future systems we will attempt to have these memory transforms learned through experience. But for now they must be given the system. Examples of SEER-2's behavior when given a small memory are presented.