Dieter Jaeger and Phillip Wolff, How to Build Bridges between Computational Neuroscience and Cognitive Psychology
In the first CMBC Faculty Lunch Discussion of the Spring
2015 semester, titled ‘How to Build Bridges between Computational Neuroscience
and Cognitive Psychology?’ Dr. Phillip Wolff (Department of Psychology, Emory
University) and Dr. Dieter Jaeger (Department of Biology, Emory University) conversed
about the ways in which cognitive psychology and neuroscience can develop more
meaningful cross-disciplinary collaborations. The two fields address
overlapping research questions and use some of the same modeling tools, but,
historically, have been relatively isolated from one another. Drs. Wolf and
Jaeger emphasized the benefits that can come from greater collaboration between
the two disciplines, and acknowledged the challenges that make collaboration
inherently difficult. The lunch discussion itself provided a starting-point for
bringing faculty and researchers from both disciplines together, encouraging
dialogue about the advantages of building a stronger alliance between the two
disciplines.
Collaboration between cognitive psychology and computational
neuroscience has the potential to be mutually beneficial, as research findings
from one discipline help constrain and guide research in the other. Cognitive
psychology research delineates how human minds represent information and
perform computations in different contexts. Through subtle manipulations of
task design, cognitive psychologists can identify whether information can be
represented in different ways and which representation is most likely, given
the particular context, learning-related changes, and individual cognitive biases.
For example, cognitive psychology research has found that the spatial relationship
between objects can be represented in multiple ways by the human mind, and that
different cultures have biases toward particular representations. A ball
sitting on the ground in proximity to a chair could be described as being to
the north of the chair, in front of the chair, or to the right of the chair.
The first description captures allocentric properties of the scene, the second reflects
the position of the objects relative to each other, and the third corresponds
to an egocentric perspective. By asking participants to describe the scenes, cognitive
psychologists can discern which type of spatial representation participants are
using. These varied representations likely map on to different neural
processes, so it would be important for a neuroscientist to identify which
representation is being used in a particular task, in order to test a more constrained
hypothesis about neural pathways of activity. In this way, research findings
from cognitive psychology provide top-down guidance to neuroscientists, allowing
them to make more precise predictions about the location and type of neural
activity they expect to see.
Neuroscientists examine neural processes at varying spatial
levels, from subcellular processes taking place in individual neurons, at one
end of the continuum, to interactions between functional brain networks at the
other end of the continuum. Just as cognitive psychology findings can be used
to provide top-down guidance in constraining the hypotheses of neuroscientists,
the corpus of neuroscience research can provide bottom-up information about the
flow of activity in neural pathways and help psychologists make more refined
predictions about cognition, perception, and behavior. For example, lesion and
neuroimaging research has provided evidence that there are distinct learning
and memory systems in the brain—a hippocampal-based system and a striatal-based
one--and that these systems competitively and cooperatively interact, depending
on the context. Neuroscience research can help cognitive psychologists refine
predictions and build more accurate models to capture how people learn new
skills and remember information.
Despite the advantages to be had from greater communication between
the disciplines, there are a number of challenges that make collaboration
difficult. Some attendees raised the point that between the cellular processes that
computational neuroscientists study, and the higher-level cognitive phenomena that
cognitive psychologists research, exist multiple layers of neurobiological
complexity that are challenging to traverse. A single neuron itself is very
complex, receiving inputs from thousands of other neurons and responding to diffuse
neuromodulators in a variety of different ways. Thousands of neurons organize into well-structured networks,
which themselves give rise to larger-scale maps, such as the somatotopic map in
the somatosensory cortex. At a larger spatial scale, brain regions with
well-studied functions come into view, and these regions interact with one
another in coordinated ways, defining systems (e.g., the limbic system). Because
there are many intermediary steps between the level of neurons and the level of
functional systems, and each level is itself defined by profoundly complex
interactions, it is inherently challenging to model how activity in individual
neurons, or even neuronal networks, map on to higher-level phenomena like
language and perception. Despite these challenges, neuroimaging is a promising tool
for bridging the gap between nervous system activity at different spatial levels
and human cognition. Neuroimaging detects functional activity at the systems
level in the brain as humans perform cognitive and perceptual tasks. Thanks to
research being carried out by investigators like Shella Keilholz in the Wallace
H. Coulter Department of Biomedical Engineering at Emory University and Georgia
Institute of Technology, we are gaining insight into how systems-level activity
detected by neuroimaging corresponds to the coordinated neural activity occurring
on a much finer spatial scale. Hence, neuroimaging provides an important link
between how activity at smaller spatial scales in the brain generates
higher-level cognitive processes.
Another challenge that has proven to be problematic
historically for the two disciplines is the dearth of models and theories that
make clear and testable predictions. In the early 1990s, the two disciplines temporarily
came together to experiment with a new generation of connectionist network
models. However, neuroscientists quickly soured on these models because they
failed to make clear predictions about neural activity, and psychologists found
that they modeled cognition and behavior in ways that were very different from
how human minds appear to work. As a result, the two disciplines drifted apart
and made less effort to coordinate conferences and meetings together. Recently,
new and exciting machine learning algorithms have emerged that may encourage
the disciplines to work together more closely again, by providing the means to
design and test more promising models. In addition, efforts made by centers
like CMBC and other cross-disciplinary institutes and national meetings can facilitate
greater communication between the researchers in these fields. As a case in
point, the CMBC Faculty Lunch Discussion was successful in bringing together Emory
researchers from computational neuroscience and cognitive psychology, some of
whom had never met before, and creating a space for the two fields to consider
how they can benefit from greater collaboration.
