The Sigma Cognitive Architecture and System: Towards Functionally Elegant Grand Unification

Abstract

Sigma (Σ) is a cognitive architecture and system whose development is driven by a combination of four desiderata:grand unification,generic cognition,functional elegance, andsufficient efficiency. Work towards these desiderata is guided by thegraphical architecture hypothesis, that key to progress on them is combining what has been learned from over three decades’ worth of separate work oncognitive architecturesandgraphical models. In this article, these four desiderata are motivated and explained, and then combined with the graphical architecture hypothesis to yield a rationale for the development of Sigma. The current state of the cognitive architecture is then introduced in detail, along with the graphical architecture that sits below it and implements it. Progress in extending Sigma beyond these architectures and towards a full cognitive system is then detailed in terms of both a systematic set of higher levelcognitive idiomsthat have been developed and severalvirtual humansthat are built from combinations of these idioms. Sigma as a whole is then analyzed in terms of how well the progress to date satisfies the desiderata. This article thus provides the first full motivation, presentation and analysis of Sigma, along with a diversity of more specific results that have been generated during its development.

Continuous phone recognition in the Sigma cognitive architecture

“…In contrast, Sigma's mixed (symbolic + probabilistic) and hybrid (discrete + continuous) corebased on graphical models (Koller & Friedman, 2009) has already been shown capable of a uniform non-modular integration of symbolic decision making with probabilistic perception in the form of conditional random fields (CRFs) and simultaneous localization and mapping (SLAM) (Chen, et al, 2011). More broadly, a variety of forms of memory, learning and decision-making have been demonstrated in Sigma, along with forms of perception and mental imagery (Rosenbloom, Demski, & Ustun, 2016). The hypothesis underlying this workand more broadly, the speech processing work in Sigmais that Sigma is sufficiently capable of implementing speech and providing a tight coupling of it and language processing with cognition in a non-modular fashion on top of the architecture without adding any language or speech specific capabilities to the architecture.…”

“…Such virtual humans require a real-time combination of a broad set of human-level capabilities, from central cognitive (thought) processes to peripheral sub-cognitive (perceptual and motor) processes (Campbell, et al, 2011). Sigma predicates its ability to support Virtual Humans on the fact that its development is guided by a quartet of desiderata (Rosenbloom, Demski, & Ustun, 2016). Two of the four desiderata derive directly from this requirement of real-time combination of broad capabilities: (1) grand unification, combining both traditional cognitive capabilities and sub-cognitive capabilities such as motor control, vision, and speech; and (2) sufficient efficiency, executing quickly enough for anticipated applications.…”

“…Sigma achieves phone segmentation and recognition by leveraging the graphical architecture hypothesis -that the key to making progress on Sigma's desiderata entails a blending of lessons learnt from the history of work on both cognitive architectures and graphical models (Rosenbloom, Demski, & Ustun, 2016). Although not directly biologically inspired, graphical models share many of the attributes of neural networks and a number of neural network algorithms map directly onto them (Jordan & Sejnowski, 2001).…”

“…Although not directly biologically inspired, graphical models share many of the attributes of neural networks and a number of neural network algorithms map directly onto them (Jordan & Sejnowski, 2001). Graphical models have the added advantage of providing a simple path for implementing symbolic processing and integrating it with sub-symbolic processing (Rosenbloom, 2012b;Rosenbloom, Demski, & Ustun, 2016). Sigma uses factor graphs as its graphical model of choice, with a variant of the summary product algorithm as its solution method (Kschischang, Frey, & Loeliger, 2001).…”

Clarifying metacognition through computational modelling

“…Importantly, Newell (1972) set down a production system at the heart of the architecture,"I laid out production systems as detailed models of the human control structure." Following Newell's lead, researchers began to build cognitive architectures to model human cognition, such as Clarion (Sun, 1997), Epic (Meyer & Kieras, 1997), Sigma (Rosenbloom, Demski, & Ustun, 2016), and ACT-R (Anderson & Lebiere, 1998). While these researchers would disagree on much, they would also discover much common ground.…”