IST lab receives software designed to mirror how humans think and act

UNIVERSITY PARK, Pa. -- Cognitive modeling -- an area of computer science that deals with simulating human problem solving and mental task processes in a computerized model -- has been applied in a variety of areas such as military simulations, computer game and user interface design, and artificial intelligence (AI) applications. Such models, which often act as agents (self-contained computational systems) in synthetic environments, can be used to simulate or predict human behavior or performance on tasks similar to the ones modeled. The Applied Cognitive Science Lab at Penn State’s College of Information Sciences and Technology (IST) recently was gifted copies of a commercial agent development software product that is designed to create more realistic models that reflect how emotion and physical factors affect human decision making and behavior.

“If an opponent in a simulated environment always does the right thing and always does it immediately, it can be hard to learn how to respond to that opponent’s real-life counterpart,” said Frank Ritter, a professor at the College of IST and co-director of the Applied Cognitive Science Lab. “It’s useful in simulation to have a model that more accurately reflects how people really behave.”

Agent Oriented Software (AOS) has given the Applied Cognitive Science Lab 20 copies of the commercial JACK (Java Agent Construction Kit V5.6) and CoJACK (Cognitive JACK) agent development software. AOS supplies products for building, running and integrating commercial-grade multi-agent systems, built on a logical foundation: Beliefs, Desires, Intentions (BDI).

Agents are defined in terms of their beliefs (what they know and what they know how to do), their desires (what goals they would like to achieve), and their intentions (the goals they are currently committed to achieving).  AOS technology is employed across a wide range of industries and applications, including oil production decision support, space vehicle decision support systems, and military simulation and synthetic environments. For the latter application, AOS supplies CoJACK, a novel cognitive architecture that uses psychological principles to predict how humans will behave in a given situation. CoJACK is used in simulation systems to create virtual actors, and models the structural properties of the human cognitive system. It also predicts how moderating factors such as fatigue and emotion affect the cognitive architecture and thus behavior.

“You can think of it as a big JAVA (a popular computer programming language and platform) library for writing complex algorithms that behave either like AI programs or like humans,” Ritter said.

Cognitive architectures, many of which are based on an analogy of the mind resembling a computer, have been in use since the 1980s. Modules in a cognitive architecture represent functional centers for different aspects of human behavior and cognition. A cognitive architecture is a blueprint for intelligent agents. It proposes (artificial) computational processes that act like certain cognitive systems, most often like a person, or acts intelligent under some definition.

According to “CoJACK: A Moderated  BDI Cognitive Architecture for Realistic Virtual Actors,” a report prepared by the AOS Group, in the past decade the computer games industry has developed photorealistic 3D virtual environments that are in widespread use in homes all around the world and have also been adapted to military applications. Typically, whether for games or military applications, simulated human entities are an essential part of the scenario. However, in contrast to hardware such as aircraft, tanks and weapons systems, even highly-trained humans can vary significantly in their response to a given situation.

“Although the inherent variability of humans has been widely recognized, virtual environments have tended to neglect this phenomenon because it is very difficult to model the depth and breadth of human behavior,” the report states.

CoJACK is intended to provide improved behavioral realism in synthetic human agents for CGF and computer gaming applications. It accomplishes that goal by simulating the structural properties of the human cognitive system—the information processing mechanisms that are fixed across tasks.

“It is generally accepted that humans share a common cognitive architecture,” the AOS report states. “CoJACK embodies these commonalities and, in part, expresses the individual variation as differences in the values of the agent’s internal (cognitive) parameters.”

CoJACK is explained in detail in “CoJACK: Achieving Principled Behavior Variation in a Moderated Cognitive Architecture,” which was written by Rick Evertsz of the AOS Group; Ritter; Paolo Busetta and Matteo Pedrotti of the AOS Group; and Jennifer Bittner, a research scientist now with the Air Force, who holds a doctorate in cognitive psychology from Penn State.

CoJACK is an augmentation of JACK, a system for creating intelligent agents including teams. JACK applications consist of a collection of autonomous agents that take input from the environment and communicate with other agents. JACK, Ritter said, which is “written for AI programmers by AI programmers,” is used to create intelligent agents in a wide range of simulations. However, a JACK agent will execute its plans without error and as fast as the hardware platform allows.  Human cognition and performance, on the other hand, makes mistakes and varies.

“Human behavior can be modified (moderated) by a range of factors, including temporal, environmental, physiological and internal factors,” the researchers wrote in the article.

Those moderators, Ritter said, include the effects of various types of stress and substances such as caffeine. CoJACK, Ritter said, was designed to model those “non-rational aspects of cognition.” One of the major limitations of most AI-based models, according to the researchers, is that they fail to represent the time taken to think and act. CoJACK was designed for the purpose of modelling the variation in human behavior. CoJACK alters the performance of a JACK agent so that it more closely resembles how a human would perform the task at hand.

“Co-JACK does not think as fast or as accurately as JACK,” Ritter said.

The Applied Cognitive Science Lab, which is also co-directed by David Reitter, an assistant professor at the College of IST, focuses on models that can explain human behavior for testing human-computer interfaces, to understand network formation, how moderators such as caffeine and stress influence behavior, and to serve as colleagues and opponents in simulations. The lab’s cognitive simulations show how the human mind interacts with networked teams and communities and how these evolved properties give rise to emergent group cognition. The lab members support this work with software to build models more easily, record behavior, and to build tutors, and through courses related to cognitive science and IST in general.

Ritter and the AOS researchers created CoJACK, Ritter said, by looking at other cognitive architectures, such as ACT-R, and incorporating useful elements into JACK. ACT-R is a cognitive architecture with sub-symbolic parameters that predict variation in cognition. CoJACK has over 30 cognitive parameters taken from ACT-R and adapted for a BDI architecture.

“We think (CoJACK) is easier to use than the cognitive architectures we took it from,” Ritter said.

Ritter and his fellow researchers have used CoJACK in a number of tasks, including caffeine and task appraisal studies, in studies exploring fear and group cohesion, and as agents in synthetic environments.  They have also used CoJACK to explore the effects of rules of engagement (ROE) on performance. They have found that applying ROEs slow down behavior and would be influenced by factors that influence cognition. In addition, the researchers have used CoJACK to model how fear and morale vary and lead to action and inaction in an improvised explosive device scenario.

Copies of the CoJACK software product are available to be used at Penn State at no cost. Individuals who are interested in obtaining a copy for Unix, Mac or Windows should contact Ritter at frank.ritter@psu.edu

Contacts: 
Last Updated March 18, 2014