Saturday, October 25, 2008

Week 6: Computational models of cognition and affect (Part 1)

Wagar, B. M., & Thagard, P. (2004). Spiking Phineas Gage: A neurocomputational theory of cognitive-affective integration in decision making. Psychological Review, 111, 67-79.

In this article, Wagar & Thagard present a new computational model of cognitive-affective processing, called GAGE after the famous case of Phinneas Gage (whose personality changed dramatically after a tamping iron destroyed the left side of his brain, transforming him from a reliable, dependable and level-headed figure to an impulsive and profane individual). GAGE focus on the contribution of the ventromedial prefrontal cortex in gauging future consequences and behaving accordingly. Specifically, the VMPFC has been implicated in the ability to refrain from behavior leading to an immediate reward if that behavior has future negative consequences, or delaying immediate reward for a future, larger reward. In the GAGE model, Wagar and Thagard examine how the VMPFC and amygdala interact with the hippocampus to coordinate potential responses with bodily states associated with the current situation and contextual information about the situation. They were particularly interested in the mechanism by which context has a moderating effect on emotional reactions to stimuli.
Their model is an extension of A. Dimasio’s (1994) somatic marker hypothesis, whereby feelings or emotional states become associated with long-term outcomes of certain responses to a given situation. The VMPFC is thought to play an important role in generating somatic markers. In this hypothesis, sensory representations of a response to a current situation activate knowledge about previous emotional experiences in similar situations. These markers act as biases influencing higher cognitive processes that coordinate responses. Wagar & Thagard extend on this by suggesting four key brain structures involved in this process. First, the VMPFC responds in concert with the amygdala. However, Wagar & Thagard suggest the mechanism by which the amygdala response (or immediate emotional response) versus the VMPFC response (based on potential outcomes of responses) wins access to higher cognitive processing is through gating by the nucleus accumbens, which in turn gates information based upon contextual information received from the hippocampus. The process is hypothesized to unfold as such: 1) the VMPFC receives input from sensory cortices, representing behavioral options; 2) the VMPFC also receives input from limbic regions, providing information about internal bodily states, most notably from the amygdala; 3) the VMPFC records signals defining a given response by encoding representations of the stimuli and comparing it to the behavioral significance of somatic states that have been previously associated with the response; 4) the VMPFC generates a “memory trace” representing the action and expected consequences of that action; 5) through reciprocal connections to the amygdala, the VMPFC elicits a reenactment of bodily states associated with the specific action; 6) this covert emotional reaction is passed on to overt-decision making processes – however, the transmission of this information is gated by the NAcc, as controlled by the hippocampus, as 7) the hippocampus controls VMPFC and amygdala throughput by depolarizing the NAcc based upon context – the NAcc allows only activation signals from the VMPFC and amygdala through that are consistent with the current context allowing spike activity in the NAcc. As staed in the article, “The hippocampus influences the selection of a given response by facilitating within the NAcc only those responses that are congruent with the current context (p.70).”
This is where the article loses me a bit, because I am not entirely certain by what process the hippocampus is purported to match current contextual information with memory traces about past contexts, as generated by the VMPFC, in order to chose which potential response information to allow through. For example, given the tendency for individuals with anxiety and mood disorders, and particularly trauma, to misread the current contextual information, this would be an important part of this process to understand. Individuals who have experienced trauma, for example, show a tendency to disproportionally map past representations onto current contexts. If Wagar & Thagard are suggesting the hippocampus is matching current context to past memory traces, this would imply individuals with trauma have deficits in the ability of their hippocampus’s to accurately gauge the current context. (However, perhaps it is the result of affective influences on sensory processing, such that information about the current context is distorted, and thereby may match memory traces more closely.) In addition, while I can understand the mechanics of this proposed process, it leaves me with open questions about the hippocampus’s “motivation.”
Nevertheless, Wagar & Thagard go on to present evidence from two studies of the GAGE model. The goal of the first study was to see if GAGE could simulate the experimental results of the Iowa gambling task in Bechara et al., 1994. In this task, participants are given a choice of four decks and are asked to make series of card selections from each of the four decks. They are given $2,000 as a loan to start, and play the decks to try to capitalize on this loan. “Bad” decks give immediate rewards, but long-term net losses, whereas “good” decks give larger delayed rewards and overall net gain. The results from the initial experiment showed normal participants quickly adopted a strategy of pulling cards from the good decks, thereby demonstrating the ability to delay reward for greater ultimate gains. By contrast, participants with VMPFC lesions never learned this strategy, and continued to act upon immediate rewards without regard to future consequences. In the current study, Wagar & Thagard trained GAGE on this same task. When the VMPFC was taken out of the model, GAGE acted only upon immediate reward, whereas leaving the VMPFC in the model resulted in more selections from “good” decks and greater overall gains. In essence, without the influence of the VMPFC, the computer was acting upon emotional reactions elicited by the amygdala and reflecting the immediate situation. When the VMPFC was included, decisions were based upon potential outcomes, not immediate affective appraisals. VMPFC and amygdala responses were modulated by gating of the NAcc, which in turn was modulated by the hippocampus, in line with the proposed model above.
In a second study, the goal was to simulate the role of context in the integration of physiological affective arousal and cognition; specifically, the mechanism by which context moderates emotional reactions to stimuli. They had the machine gauge emotional reactions as positive or negative while in a positive versus a negative context. The results of this study showed that when the NAcc was presented with two possible VMPFC representations, the hippocampal-derived context drove GAGE’s behavior, such that positive contexts elicited positive responses and vice versa. The researchers go on to suggest that the NAcc stores associations between VMPFC and hippocampus to elicit representations based on the current context.
The two studies are summarized as such: Study 1 demonstrates that “the VMPFC and the amygdala interact to produce emotional signals indicating expected outcomes and that these expected outcomes compete with immediate outcomes for amygdala output…temporal coordination between the VMPFC and amygdala is a key component to eliciting emotional reactions to stimuli (p. 76).” Study 2 demonstrates that context exerts an effect on cognitive-affective integration, such that “For the signals from the VMPFC and the amygdala to access brain areas responsible for higher order reasoning, context information from the hippocampus must unlock the NAcc gate, allowing this information to pass through (p.76).” These conclusions lead to a few questions. First, does this suggest that the hippocampus overrides potential outcome decisions presented by the VMPFC? In other words, if the VMPFC is creating memory traces based on past experiences in similar situations, is it the case that the VMPFC is assuming one context that the hippocampus either rejects or confirms? Do the context appraisals formed by the hippocampus represent contextual memories or the actual current context? And how does the hippocampus form judgments about the current context, unless it is comparing it to prior encounters with the context? Isn’t contextual memory formation a key function of the hippocampus? There seems to be almost a memory loop going on here – the VMPFC is taking in sensory information and judging appropriate behavioral responses based upon the behavioral outcome of past encounters with the stimuli – which would imply some form of contextual representation. This information then is gated by the way in which the hippocampus judges the current context, and whether the information presented by the VMPFC is matching this judgment. The hippocampus encodes contextual memories about past encounters with the current context, which you would think influences the VMPFC’s initial representations. Is this process more dynamic than the GAGE model is implying?

No comments: