Solutions to the implied matching problem
Loading...
Authors
Scally, Jonathan Richard
Issue Date
2013-05
Type
Electronic thesis
Thesis
Thesis
Language
ENG
Keywords
Cognitive science
Alternative Title
Abstract
When humans interact with complex environments they have a set of expectations which define their goals and their framework for reasoning. This stands in contrast to the essentially unbounded space of possible inputs which exists in the environment. While humans can effectively work in complex environments, this situation presents a serious challenge to the development of flexible and autonomous human-level intelligent agents. Many of the inputs available to the agent "in the world" are functionally equivalent or extremely similar to the agent's expectations, but require an expensive conversion or modification in order to match. We term this the Implied Matching Problem. Conversion of the inputs in real time produces either a combinatorial explosion of input terms, an arbitrary restriction of the potential range of the input, or significant amounts of bookkeeping. None of these approaches is neurally plausible and all of them prohibit scaling to real world applications. A flexible conversion approach was implemented for the Polyscheme Cognitive Architecture which demonstrates flexible matching across functionally equivalent category and "world" terms. This implementation suggests that performing implied matching across the dimensions of cognition most critical for human reasoning (termed the Cognitive Substrate) can enable robust inference in complex domains beyond what current systems can achieve.
Description
May 2013
School of Humanities, Arts, and Social Sciences
School of Humanities, Arts, and Social Sciences
Full Citation
Publisher
Rensselaer Polytechnic Institute, Troy, NY