Overcoming Feature Space Bias in a Reactive Environment

Abstract

The NOME cognitive architecture drives an organism that acquires early language about a reactive environment from a teacher organism. Once its vocabulary is sufficiently developed, the pupil organism can be given simple commands, which cause it to create and execute verb-based plans in the environment. An example task is moving Tower-of-Hanoi disks across pegs. The NOME architecture isolates a set of primitive, undefined symbols that form the lowest level basis for cognitive activities. Limited to treating these dimensions as uninterpretable, the architecture undertakes to characterize their meanings by indirect methods. Dimensions are understood by learning about their interactions. Knowing how its basis dimensions interact allows the organism to selectively attend only the relevant features of the environment. Testing this interaction knowledge against the external environment allows the organism to evaluate the quality of match between the real and intended meanings for dimensions.