GMU BICA: An Integrated Self-Aware Cognitive Architecture

This project introduces a general cognitive architecture allowing for a computational implementation of the key features of human cognition, including examples like basic human memory systems (working, semantic, episodic, procedural), the autonomous cognitive growth ability, social, emotional and communicational capabilities, and self-awareness. According to our view, the key element enabling these features in an intelligent agent is the Self of the agent, understood as neither the hardware nor the software, but an idealized abstraction represented in the agent's mind. Our approach to implementation of this Self is based on four building blocks: (1) neuromorphic cognitive map, (2) a set of self axioms, (3) mental state lattice, (4) the framework of schemas that generalize productions, operators and chunks used in popular cognitive architectures like Soar, Act-R and Epic. In addition, we suggest a mapping of our cognitive architecture onto brain structures. This mapping helps us to borrow computational and design principles from cognitive neuroscience.

 

Example: Necker Cube Perception

Our first implementaiton of the architecture was done in Matlab and in Python during this Spring semester. It was limited to perception of a stationary object: a cube.  

The animation (Figure 2) shows as the schema of a face of a cube is mapped onto a wire diagram of the cube (presented to the system via the IO buffer).

At the next step (not shown here) a cube schema is used, which interprets one of the two upright faces as the front face. The mapping of this schema is ambiguous: there are two possibilities for the pe rceived cube configuration. One of them is selected spontaneously. As the time flows, a new mental state is activated, and the entire process of perception repeats in that mental state (possibly, with a different outcome). Then the new mental state takes the place of I-Now, thereby updating the perceived content (I-Now represents consciousness in this framework). This process repeats. As a result, t he perceived cube configuration spontaneously flips, in agreement with psychological experiments.

                   


Figure 2. (try mouse over): A model of cube perception based on our architecture. IO: input/output buffer. SM: semantic memory (the cube schema is not included).

 

Figure 3. A recent demo (flash video):
Cognitive mapping of an indoor environment

 

Figure 4. A more recent example:
A bootstrapped learning scenario

Recent presentations of our progress in this project include:

Event   Location   Date   Presenter   Presentation type
Don Perlis seminar   College Park, MD   20 March 2006   Samsonovich   invited talk
ARCH Lab BB   Fairfax, VA   22 March 2006   Samsonovich   invited talk
The C. Conference   Tucson, AZ   7 April 2006   Samsonovich   conference talk
DARPA meeting   Boston, MA   9 May 2006   De Jong   conference talk
AGIRI Workshop   North Bethesda, MD   21 May 2006   Samsonovich   invited conference talk
ICDL 5   Bloomington, IN   2 June 2006   Samsonovich   conference poster
AAAI Workshop   Boston, MA   16 July 2006   Samsonovich   conference talk
WCCI   Vancouver, BC   17 July 2006   De Jong   conference talk
AAAI Workshop   Boston, MA   17 July 2006   Samsonovich   conference talk
WCCI   Vancouver, BC   17 July 2006   De Jong   conference panel discussion
DARPA meeting   San Francisco, CA   9 August 2006   De Jong   conference talk
SFN   Atlanta, GA   October 2006   Ascoli   conference poster
Krasnow BB   Fairfax, VA   November 2006   Samsonovich   talk
Krasnow Retreat   Fairfax, VA   January 2007   Samsonovich   talk: annual group report