Who Needs Emotions? The Brain Meets the Robot

asic principles for emotional processing 105
lines but not for long, as this book attests (Fellous, Armony, & LeDoux,
2003). As we have discussed above, it may be fruitful for computational
models to approach the problem of emotion by considering one emotion at
a time and to focus on how the emotion is operationalized without losing
the “big picture” of how feelings might emerge.
This approach has led to the discovery of basic principles that may apply
to other emotions as well as fear:
Emotions involve primitive circuits. These primitive circuits are
basic, robust processing units that are conserved across evolution.
In some circumstances, cognitive (i.e., nonemotional) circuits can
function independently from emotions.
Emotional memories are somewhat different from other kinds
of memory. They may last longer and be more vivid (reassociate
rigidly and effectively with other memory items). Some types
of nonemotional memory (e.g., working memory) help extinguish
emotional memory (e.g., fear).
There are two parallel routes of emotional processing of a stimulus.
One is fast (thalamic–amygdala pathway); the other is slower
(cortical–amygdala pathway) and presumably modulates the fast
route. (Compare the dual routes analyzed in Chapter 5, Rolls.)
There are two physically separate inputs to an emotional (evaluation)
system. The first is reserved for simple stimuli such as a
tone (LA→CE in the fear circuit); the second is reserved for more
complex stimuli, such as context, and includes more processing
stages (hippocampus→B/AB→CE in the fear circuit).
Emotional expressions are triggered by a central signal (CE activation),
but the specifics of the expressions are determined locally
(lateral hypothalamus, blood pressure; periaqueductal gray,
freezing; bed nucleus, stress hormones, etc., in the fear circuit),
according to the current state of the animal (current heart rate,
environmental conditions, actual levels of hormones).
These basic principles might serve as a starting point in the design of
computational models of emotions.
The future of emotion research will be bright if we keep in mind the importance
of focusing on a physiologically well-defined aspect of emotion, using
an experimental approach that simplifies the problem in such a way as to
make it tractable, circumventing vague and poorly defined aspects of emotion,
and removing subjective experience as a roadblock to experimentation. This is
not to suggest that the problems of feelings should not be explored, but, instead,
that they should be explored in a way that builds on a firm understanding
of the neural mechanisms that subserve the underlying behaviors.