Generalized Temporal Induction with Temporal Concepts in a Non-axiomatic Reasoning System

The introduction of Temporal Concepts into a Syllogistic based reasoning system such as NARS (Non-Axiomatic Reasoning System) provides a generalized temporal induction capability and extends the meaning of semantic relationship to include temporality.

  • PDF / 67,482 Bytes
  • 4 Pages / 439.37 x 666.142 pts Page_size
  • 57 Downloads / 228 Views

DOWNLOAD

REPORT


2

Evolving Solutions Ltd., Newbury, UK [email protected] Institute for Software Technology, Graz University of Technology, Inffeldgasse 16b/II, Graz, Austria [email protected]

Abstract. The introduction of Temporal Concepts into a Syllogistic based reasoning system such as NARS (Non-Axiomatic Reasoning System) provides a generalized temporal induction capability and extends the meaning of semantic relationship to include temporality.

1 Introduction For the purpose of this paper, NARS [6] can be considered as a reasoning system which takes two premises, a task and a belief, and carries out an inference process, using defined logic rules. The task and belief are selected according to a control system and are required to have a common component in order for the logic rules to apply. Due to this constraint, arbitrary premises such as two sequentially occurring events (with no common component) cannot be selected for inference [3]. This constraint presents a problem for temporal reasoning, where it is desired to form sequences of arbitrary events, for sequence learning. NAL (Non-Axiomatic Logic) [6] is the logic used by NARS and includes logic rules for temporal induction but these rules require special handling and do not sit comfortably within the unified principle of cognition that applies to semantically related logic rules. The introduction of Temporal Concepts addresses this shortfall [7] and allows the temporal aspect of premises to be considered as a semantic relation between premises, thereby allowing sequences of arbitrary events, instead of relying on an event-chainer: applying inference between succeeding events, as was the case in OpenNARS 1.7.

2 Temporal Concurrency Temporal concurrency can occur on vastly different timescales, for example two sub-atomic interactions versus two birthdays. For the purpose of this discussion concurrency, in NARS, is defined as two events occurring within a temporal window, called DURATION, where DURATION is defined as a number of system cycles. © Springer International Publishing Switzerland 2016 B. Steunebrink et al. (Eds.): AGI 2016, LNAI 9782, pp. 254–257, 2016. DOI: 10.1007/978-3-319-41649-6_25

Generalized Temporal Induction with Temporal Concepts

255

The justification for this approach is based on research in cognitive science, whereby, humans discern events as being concurrent when experienced within a temporal window of roughly 80 ms [1]. The primary role of this implicit form of temporal concurrency is to allow the formation of perception sequences. When events span longer time windows, an explicit representation can be used (expressed in Narsese) [5]. Different NARS systems can have different values for DURATION, where a system perceiving extremely fast perception streams, such as monitoring chemical interactions, would have a short DURATION time in the order of nanoseconds or microseconds, whilst on the other hand a system monitoring whale migration data could have a much longer DURATION.

3 Implementation NARS contains two types of concept: gen