Cognitive Architecture

The goal of this effort is to develop a sufficiently efficient, functionally elegant, generically cognitive, grand unified, cognitive architecture in support of virtual humans (and hopefully intelligent agents/robots – and even a new form of unified theory of human cognition – as well).

A cognitive architecture is a hypothesis about the fixed structures that provide a mind, whether in natural or artificial systems, and how they work together – in conjunction with knowledge and skills embodied within the architecture – to yield intelligent behavior in a diversity of complex environments.

A grand unified architecture integrates across (nominally symbolic) higher-level thought processes plus any other (nominally subsymbolic) aspects critical for successful behavior in human-like environments, such as perception, motor control, and emotions. A generically cognitive architecture spans both the creation of artificial intelligence and the modeling of natural intelligence, at a suitable level of abstraction. A functionally elegant architecture yields a broad range of capabilities from the interactions among a small general set of mechanisms – essentially what can be thought of as a set of cognitive Newton’s laws. A sufficiently efficient architecture executes quickly enough for its anticipated applications; for example, taking no more than 50 msec per cognitive cycle for real-time virtual humans.

Our focus is on the development of the Sigma (∑) architecture, which explores the graphical architecture hypothesis that progress at this point depends on blending what has been learned from over three decades worth of independent development of cognitive architectures and graphical models, a broadly applicable state-of-the-art formalism for constructing intelligent mechanisms.  The result is a hybrid (discrete+continuous) mixed (symbolic+probabilistic) approach that has yielded initial results across memory and learning, problem solving and decision making, mental imagery and perception, speech and natural language, and emotion and attention.

A range of short descriptions of Sigma can be found in the various the papers about it that are listed on the Recent Publications page, but the best short overview is the the 2013 AISB paper.  A much longer, more detailed, and up to date paper is now available at 2016 JAGI paper.

A public release of Sigma is available under a BSD 2-clause open-source license.  Information on the release and how to set it up can be found at Sigma Release page.  Also included is a tutorial that covers some of the basics of using Sigma.