IST researchers examine interplay of cognition, physiology, behavior

UNIVERSITY PARK, Pa. -- Researchers at Penn State’s College of Information Sciences and Technology are using cognitive architecture – a broad theory of human cognition based on a wide selection of human experimental data and implemented as a running computer simulation program – to understand how cognitive processes interact with biological systems and influence social behavior.

This work, being conducted by doctoral students Christopher Dancy and Changkun Zho, and Frank Ritter, professor of information sciences and technology, has broad implications in areas including educational software design, military operations, video games and the formation of social networks.

“This work shows two important directions for simulating and predicting human behavior,” Ritter said. Dancy is investigating and modeling connections between physiology, emotion, and cognition. Zhao’s work focuses on modeling social processes.

Dancy’s research paper, “Towards Adding a Physiological Substrate to ACT-R,” co-authored by Ritter and Keith Berry, a U.S. Army flight surgeon and doctoral student in the bioengineering department at Penn State, was recently honored as best paper at the 21st annual Conference on Behavior Representation in Modeling Simulation (BRIMS) on Amelia Island Plantation in Amelia Island, Fla.

In this paper, Dancy explained how the researchers connected a physiological simulation system to a cognitive architecture to imitate how the sympathetic nervous  system -- one  the three parts of the autonomic nervous system, which mobilizes the body's nervous system fight-or-flight response -- affects human cognitive processes, such as memory.

“This is an opportunity for evolution of the cognitive architecture and a new way of looking at it,” Dancy said.

Cognitive architectures, many of which are based on an analogy of the mind resembling a computer, have been in use since the 1980s, he said. Modules in a cognitive architecture represent functional centers for different aspects of human behavior and cognition.

“The work by Dancy and Berry shows that we can start to take account of how our physical body influences, and is influenced, by cognition,” Ritter said.

ACT-R is a cognitive architecture mainly developed by John Anderson at Carnegie Mellon University. ACT-R specifies how the brain itself is organized in a way that enables individual processing modules to produce cognition. A “model” (i.e. program) in ACT-R refers to a combination of a cognitive architecture and a set of knowledge that is needed to perform a particular task. The researcher determines the specific knowledge incorporated into a particular model.

Running a model produces a step-by-step simulation of human behavior that specifies individual cognitive operation (i.e., memory encoding and retrieval, visual and auditory encoding, motor programming and execution and mental imagery manipulation).

In the article, Dancy, Ritter and Berry present ACT-R Phi, an extension of ACT-R that provides an additional representation of human physiology. According to the authors, the connection of a physiological model to a cognitive architecture provides an opportunity to simulate a wide range of human behavior.

“Now, the simulated brain has to worry about the body,” Ritter said.

The extended version of ACT-R, according to Dancy, Ritter and Berry, allows a user to computationally realize theories involving cognition, physiology and their interaction. ACT-R Phi utilizes a new module to send information to, and receive information from, HumMod, a system that allows one to continuously simulate physiology over time.

To establish a connection between ACT-R and HumMod, the researchers represent the HumMod system as a substrate, or substratum, running in conjunction with the ACT-R architecture. To realize this substrate within the ACT-R architecture, the researchers created a module that represents a physiological substrate and applies its effects to other modules.

According to the authors of the article, basic bodily motivations, such as hunger and thirst, ultimately can affect memory recall and attention. HumMod provides a representation of the effect of osmoreceptors, which measure changes in extracellular body fluid and are known to be responsible for thirst. The model also includes representations of stress hormones, such as cortisol and norepinephrine.

The cognitive model described in the paper provides a representation of a scheduled sound that causes a form of a startle reflex. This reflex, according to the authors, causes sympathetic nervous system activation in HumMod to affect the modules in ACT-R Phi that represent declarative (factual) knowledge.

“How does that loud sound, over time, affect retrieval of those memories?” Ritter said.

According to the researchers, by affording more representations to simulate environmental effects on human behavior and cognition through a physiological connection, the architecture becomes capable of developing more human-like agents for simulation environments like those in the military or gaming.

“From a military perspective, one can now also begin to simulate more behaviors related to fear and anxiety and cognition (e.g., PTSD),” they wrote in the article.

The next step in the research, Ritter said, is to examine how other physiological factors influence behavior.

Zhao’s research takes a rather unusual approach in that it employs ACT-R, which is primarily associated with cognitive science, to model social processes. “Socio-cognitive Networks: Modeling the Effects of Space and Memory on Generative Social Structures,” written by Zhao, Ryan Kaulakis, Jonathan H. Morgan, Jeremiah W. Hiam, Joseph P. Sanford and Ritter, all from the College of IST, and Geoffrey P. Morgan from the School of Computer Science at Carnegie Mellon University, received an honorable mention at the BRIMS conference.

In the paper, the researchers examine how environmental factors contribute to the consolidation of social ties in memory. According to Zhao, Dunbar’s number, which was first proposed by British anthropologist Robin Dunbar, holds that a person can retain an average of 150 social ties in his or her memory. In keeping with Dunbar’s theory, Zhao and his colleagues hypothesized that larger populations acting over longer time periods in fully connected environments will result in the most connected network structures. They also expect that less connected environments will result in more fragmented networks.

Using the ACT-R architecture, Zhao and his colleagues constructed VIPER, an agent-based simulation that enables them to model human social behavior. Within the environment provided by the server, agents or human subjects are situated on maps of interconnected rooms. The agents can see and communicate within each room.

The objective of the simulation model, Zhao said, is to demonstrate how people form and maintain social networks, and under what conditions they decrease or decay. Using VIPER as a platform, the researchers model the effects of memory, physical layout, and time. To replicate human navigation behavior, they implemented two navigation strategies: random-walk and fixed-path. The random-walk strategy replicates navigation patterns without a specific goal, while the fixed-path strategy follows a set path in a small area.

“This strategy simulates routine navigation behavior, such as going to work or shopping,” the authors wrote.

From the simulation results, the researchers found that “shared public locations have higher interaction densities, in a matter similar to the water-cooler effect.” The results also indicated that smaller groups have stronger ties, that a “fixed path” results in strong localized ties, and that population size has the highest influence on the density of a social network.

By analyzing human social patterns, researchers can gain a better understanding of terrorist networks, cohorts in the military, and social cohesion in all groups.

“The work by Zhang et al. shows that human memory can be an important influence on social interactions, both short-term and long-term,” Ritter said. “We will now look at how other factors interact with memory for social networks, and improve our understanding of how networks form and maintain themselves.”

Contacts: 
Last Updated August 06, 2012