Cognitio 2011

Abstract: There are currently two general kinds of simulation tools for cognitive processes: neural networks and production systems. On one hand, cognitive simulations with neural networks (e.g., Eliasmith 2009) have biological plausibility; however, the opaque nature of neural processing makes it difficult to understand how psychological processes emerge from the neural dynamics, thus making it hard to extract an explanation out of a successful connectionist simulation. On the other hand, production systems (

e.g., Van Maanen 2009) use a symbolic approach, making it easier to interpret activity in the system; but the sequentiality inherent to fixed mechanisms for rule selection limits the simulation of cognitive processes. In this abstract, we introduce a hybrid alternative: the use of discrete-symbolic representations exhibiting dynamical change in time for simulations at a functional level.



The Complex Auto-Adaptive System (CAAS) (Camus 2008) is a massive multi-agent system that models the dynamics of knowledge activation and suppression as the system performs tasks or solves problems. CAAS has been mainly used for engineering purposes: control of robots or software entities in real-time. Organizations of agents realize the dynamic goal oriented behavior of the system. Involved in the goal-oriented/context-dependent behavior of the system, they share important traits with the Working Memory, Long Term Memory and Attention systems of the human brain. Our aim is to introduce CAAS as a simulation tool for cognitive scientists for simulations at the functional level. Temporal continuity is an important feature of cognitive representations (Spivey 2007), and we believe that CAAS’s dynamical properties are particularly suitable to cognitive simulations (more specifically cognitive disorders and bias). CAAS shares aspects with Dale and Spivey’s (2005) definition of ‘symbolic dynamics.’ In CAAS, ‘Symbolic dynamics’ is the dynamical, real-time, interaction of populations of minimal symbolic agents, out of which emerges a continuously changing geometrical representation of the environment. Symbolic and easily readable, the system reproduces various functional aspects of human cognition and thus provides a good platform to study human-like dynamic regulation. Furthermore, we think CAAS could be a good simulation tool either when we have an initially good understanding of the knowledge needed to perform the task and want to know how knowledge can organize itself into a performance of the task, or when we have no clear understanding of the knowledge needed and want to know which piece(s) of knowledge will be recruited to perform the task.



We present a new approach to cognitive simulations, one that correctly reproduces some of the mesoscopic dynamical activity of the brain, and use it to simulate traditional psychological tasks. Our hope is to provide new insights about the dynamicity of organization and disorganization (involved in some cognitive disorders) in the cognitive processes underlying human performance on cognitive tasks; we think that the observation of the processing in this artificial system (in normal and “disorganized” condition) could inform us on the underlying mechanisms of cognitive processing in humans. Testing the system with a Stroop task (Stroop 1935)  -  a test commonly used in psychology to study interference that occurs in control when two cognitive processes are in competition – we were able to simulate the same time effects of cognitive impairment (control and inhibition for context maintenance, attention and working memory) and adaptive cognitive behaviors than in “normal” and “low-span condition” in humans. We are currently working on a simulation of the Wason selection task (Wason 1966), a well-known task in the psychology of reasonning, where the properties of inhibition and control as used in the Stroop task simulation could be transposed to implement competition between systems 1 and 2 (see Evans 2008 for a survey) using emerging dynamics.