Broad - Spectrum Mitigation and the Cognitive Neurobiological Interface : considering biological rhythms in Augmented Cognition

Brain-computer interface design relies upon mitigating human performance. Recent attempts have focused mainly on real-time cognitive processing. This paper will consider biological rhythms as a longer-term and more dynamic factor for determining human performance in such systems. To characterize the contribution of such neurobiological phenomena to human performance, biological rhythms will be proposed as the input for both “sub-cognitive” state gauges and a predictive model based on complexity theory that modulate the input/output properties of basic cognitive state gauges. The relationship of biological rhythms to cognitive processing exists at multiple temporal scales, and provides an addition-al level of complexity to currently held conceptual models. Most notably, a concept called broad-spectrum mitigation will be considered as a way to improve human performance and more effectively augment cognitive processing. 1.0 Introduction: Modeling Broad-Spectrum Mitigation In the design of direct brain interfaces, an instance of which is the DARPA Augmented Cognition (AC) project, models of mitigating sub-optimal human performance has focused mainly on real-time cognitive processing (see Rapaport et al, 2005). This paper will consider biological rhythms as a longer-term and more dynamic factor for determining human performance in such systems. The relation-ship of biological rhythms to cognitive processing will then be considered as a means of providing an additional level of complexity to the AC conceptual model. Three types of rhythms may be relevant both specifically to attentional states and to human performance in general. Ultradian rhythms may affect daily fluctuations in arousal (Dunlap et al, 2004), while circadian and circannual fluctuations may influence even longer-term trends in human performance and indirectly inform temporally complex mitigation strategies. Specifically, it will be argued that biological rhythms set the underlying context for performance mitigations at different temporal scales. Thus, understanding their effect on arousal and task performance will lead to predictive models that serve as a type of “broad spectrum mitigation”. Broad spectrum mitigation, as a general concept, uses the complexity of long-term trends in cognitive neurobiological systems to build computational models that fill in gaps inherent in real-time system mitigation. 1.1 Broad-Spectrum Mitigation as a Control System The idea of broad-spectrum mitigation, similar to many current brain interface and Augmented Cognition systems, can be thought of as a semi-closed loop control system. Inputs from subcognitive factors such as biological rhythms combine with the more direct contributions from cognitive state gauges to provide performance mitigation to the user that is sensitive to changes in user state at multiple time scales. Figure 1 shows the major contribution of sub-cognitive components is that of an additional input. It is envisioned that inputs from the cognitive and broad-spectrum gauges operate at different time scales, so that while cognitive state gauges are updated continually, broad-spectrum in-puts provide meaningful information to the system on the scale of minutes to hours. Van Dongen et al (1999) have used the Lomb-Scargle periodogram to observe that informative biological rhythms are 2 both aperiodic and exist at multiple time scales. This means that not only do sub-cognitive inputs provide secondary and sporadic but critical information to the closed-loop system, they also require novel machine learning models for producing real-time mitigations. Figure 1: the image at the left (A) shows the architecture of a standard Augmented Cognition system. The image at right (B) shows the addition of a sub-cognitive component. 1.2 Iterated Maps as a Predictive Model The other aspect of employing broad-spectrum mitigation is that of developing an adaptive filter that integrates the contributions of both levels of information at specific time steps. Even though the Yerkes-Dodson curve is reliable for predicting when mitigations should be employed, using more ma-thematically rigorous models such as iterated maps (see Strogatz, 2001) may lend insight into events leading up to breakdowns in arousal and self-correct performance shortcomings of real-time statistical classification and pattern recognition methods. An iterated or logistic map can be used to trans-form the closed-loop control system into a dynamical system. The logistic map is sensitive to the behavior of chaotic time series, and can be defined by the following equation xn+1 = λ [xn(1-xn)] [1] where λ is the sum of the cognitive and sub-cognitive gauge readings at a particular time step, and xn is the degree of mitigation applied by the system at a particular time step. The value for lambda is typically positive, while the value of x is generally between 0 and 1. The purpose of this model is to predict the level of mitigation needed in the next moment given current inputs from both the cognitive and subcognitive gauges. Across time, critical values of lambda are reach-ed; during very short intervals, the system undergoes a phase transition and becomes stable at multiple values of xn . This temporary instability should be “bursty” across time scales; such phenomena imply that broad-spectrum mitigation will be enforced in brief and semi-periodic intervals usually related to some isolated event. Burstiness is a common property of data streaming on the internet and other communication networks (see Di Cairano-Gilfedder and Clegg, 2005), and may also act as a trigger for more complex mitigations. It is predicted that changes in the sub-cognitive state gauge over particular periods of time will drive these bifurcations, which render the system temporarily unstable. The