As sensors and analysis have become cheaper (both monetarily and computationally), there has been an increased use of psycho-physiological measurements in behavioral experiments and human-in-the-loop simulations. These measurements provide extra context for the cognitive and behavioral process; psycho-physiological measurements give one an additional window into the processes, both unconscious and conscious, that would be otherwise difficult to measure through self-report or introspection. They also allow computational modelers to connect functional processing to physiological systems, to ground these processes in activity within systems, and communication between networks of physiological systems.

Below, I give a brief overview of three particular psycho-physiological measures that are useful for cognitive systems engineering in general, and can be used with cognitive architectures in particular.


Electroencephalography (EEG) is a functional technique that involves placing sensors over the surface of a study participant’s head to continuously record (i.e., with a temporal resolution in the millisecond range) electrical brain activity. This activity is associated with firing of synapses, causing an electrical signal that can be detected by the sensors. Though the temporal resolution is very useful for understanding responses to specific events (called event-related potentials, or ERPs), the temporal resolution can make it difficult to localize activity to some neural structures (and for others, it is virtually impossible.)

General EEG patterns are categorized to represent (on average) certain behavior in people; this activity is categorized as alpha (8-12 Hz), beta (18-30 Hz), gamma (30-70 Hz), theta (5-7 Hz), and delta (0.5-4 Hz) activity. Relaxed individuals typically show alpha activity or rhythmic waves of 8-12 Hz. Beta activity signifies alertness in a participant, whereas gamma rhythms mark an ability to integrate multiple stimuli into a coherent whole stimulus. It has been suggested that theta activity may signify either global inactivity (e.g., falling asleep) or an overlearned behavior process (resulting in a lower frequency EEG). Consequently, delta waves are associated with healthy sleep in humans.

Event-related potentials, or ERPs, are used to measure a particular response to stimuli. Areas that particularly respond to the stimuli (near the scalp surface) will have higher average signal amplitude after a stimulus response. ERP analysis is segmented at deflections (i.e., when the sign of a wave slope changes), and is named by whether the deflection is positive or negative, and roughly how long after the stimulus the deflection occurred; for example, a P300 ERP component is on a positive deflection and is roughly 300 ms after the stimulus.

It has been suggested that EEG can be useful for studying differences in time course of activation for specific systems in cognitive architectures; in the case of ACT-R, activity in specific modules and buffers may be correlated with specific EEG oscillation profiles (e.g., beta or theta) and also ERP data. van Vugt (2012) used an ACT-R model that completes an attentional blink task to correlate architecture behavior with activity in modules and buffers. They found location-specific correlations between theta activity, and both the imaginal and procedural modules. Correlations between delta waves and the declarative module, and between gamma activity and the visual module, were also found. Though these correlations can be useful for understanding some of the possible areas of the brain used across time and for generalizing ACT-R activity to other EEG data, a method with a higher temporal resolution is needed to more clearly localize the neural activity. The results found here and elsewhere (e.g., Cassenti, Kerick, & McDowell, 2011; van Vugt, 2013) would be even more useful with a complementarily increased understanding of potentially localized neural structures that are active that interact during predicted cognitive architecture system activity; that is, an increase in spatial resolution is also required.

< Prev   CONTENTS   Source   Next >