Over such as motor cortex, there is still

Over the last few decades, the co-dependent disciplines of science and technology have come to
develop in such a way that the advancement of one can often stall in wait of a breakthrough in the other.
Sometimes it is science that awaits technology, seeking new methods and tools of study to move research
forward. However, most cases involve the opposite, with technology awaiting a scientific breakthrough to
drive innovation. It is this latter scenario that mediates the relationship between the advancement of brain-
machine interfaces (BMIs) and the organization of information processing in the brain. Today’s BMIs
function by capturing the brain’s neuronal signals, decoding them via an external computer, and then
translating them into actionable commands, most commonly through a neuroprosthetic device. Although
recent advances have shown BMI technology to be capable of carrying out basic command signals from
areas such as motor cortex, there is still a range of brain commands we aren’t capable of interpreting due
to a lack of understanding about the brain’s information processing systems. These systems are the center
of a long-running debate in cognitive science whose resolution is key to the advancement of BMIs, along
with many other technologies. That debate: modular vs. distributed information processing.

Until Rumelhart, Hinton and McClelland published their 1986 argument Parallel Distributed
Processing, the idea that the mind processed information in a modular system— with specific brain
regions corresponding to specific mental processes— had not yet been challenged. Their new theory,
however, proposed that the mind was instead organized as a parallel distributed processing mechanism, a
neural network composed of interacting units that collectively gave rise to our mental functions. And with
the addition of this new perspective, the debate began. It first took root in psychology, with scientists
using logical reasoning and behavioral studies to support their views. Then, in the late 1990s and early
2000s, neuroscientists studying activation patterns and neuronal connections started to apply their
findings to the debate, and thus a new set of tools were made available to study processing. Today,
scientists interested in processing use methods like fMRI, EEG, single neuron recording and lesion

We Will Write a Custom Essay Specifically
For You For Only $13.90/page!

order now

Page 1 of 6

experiments (think, Broca and Wernicke) to study how brain regions respond to stimuli and communicate
signals. Unfortunately, however, the limitations of these technologies make it significantly easier to run
studies that support modularity. For example, due to the nature of fMRI, a face processing study with
brain images implicating the FFA is easier to declare as empirical evidence in support of modularity than
a similarly run study on executive control suggesting multiple involved regions as evidence of
distribution. Nevertheless, the recent increase in academic papers studying distributed processing seems
to indicate a shift in belief among the scientific community— nowadays, most researches seem to either
support the theory of a whole-brain level of processing, or propose variations of a theory combining them

But there are still researchers whose work supports modular processing. In their study of human
visual cortex and body-related stimuli, Downing et al. found evidence of cortical regions in the brain that
respond selectively to images of the human body. Using fMRI to track activation, Downing and his team
presented a series of human body and human body-like stimuli to subjects, finding that each brain scan
showed similar activation in right lateral occipital cortex. This region, which they call the “extrastriate
body area” (EBA), showed greater activity for pictures of real human bodies as compared to partial
images, stick figures, etc., and had low activation for non-body stimuli. Downing et al. found no
anatomical overlap between the EBA and several other visual cortex regions, and partial overlap between
the EBA, MT and lateral occipital cortex— although there were voxels that solely activated in the EBA
and no other regions. From these findings, the paper concludes that there is an area of visual cortex
selective only for human bodies, making this study an often cited piece of evidence among scientists
arguing in favor of modularity.

And this study is one of many. Several experiments using fMRI to identify brain regions, like the
EBA and the fusiform face area, are often referenced as evidence of modularity in the debate’s reviews,
such as Husek et al.’s “Task-dependent evolution of modularity in neural networks” and Delis et al.’s “On
the Origins of Modularity in Motor Control”. These experiments, along with neuron recordings (for
example, Hart and Giszter’s study of modularity in motor control), all provide the same evidence: a single
area associated with a single function based on activity. There are a few innovative studies that also
support modularity, however, diversifying the range of evidence to draw from. For example, Lesicko et

Page 2 of 6

al. used neuron marker stains to identify a micro-network in rodent inferior colliculus that they then
studied further to identify modular regions of somatosensory processing. These marker stains highlighted
neurochemically distinct modules that correlated with distinct connectional modules (as inputs injected
into them terminated within instead of spreading to activate other areas). Thus, the authors argue that
these findings of modular regions that appear to have both functional and neurochemical distinctions
further support the theory of modular processing.

Lastly, in addition to conducting these traditional lab-based studies of modularity, neuroscientists
pursuing the debate have also recruited alternative methods of research, such as Berger et al.’s virtual
surgery study of motor learning. In this experiment, Berger and his colleagues used virtual surgery
technology to study the effects of two different surgeries that distinctly affect the neural motor synergies.
The authors claim that if the synergies aren’t modularly organized, then the adaptation rates between the
two surgeries would be similar as the adaptation could be achieved through similar motor learning
processes. However, if the synergies are modularly organized, then the neural mechanisms of motor
learning would be different causing the adaptation times to differ— as they found to be the case.

Looking to the other half of the debate, neuroscientists interested in the theory of distributed
processing often study higher level cognition while employing a variety of research strategies— some of
which are borrowed from traditional modular processing experiments. One such example is Valdez et al.’s
use of single neuron recordings while studying visual object encoding. In this experiment, Valdez and his
colleagues observed firing patterns of over 400 neurons in the hippocampus and amygdala while subjects
viewed objects in natural settings. While several past studies (showing subjects different objects in
artificial settings) have shown neurons encoding a single object, Valdez and his team found that in their
setting, the majority of neurons encoded more than one stimuli, and on average each neuron encoded 26%
of the objects in the experimental setting. Such results suggest that visual object processing in the brain
has distributed architecture.

Another study that uses methodology typically found in modular experiments to support
distribution is Schlegel et al’s fMRI study of motor control via a mental rotation task. In this experiment,
Schlegel and his colleagues used BOLD activation of multiple regions to show that the motor network of
the brain can recruit other regions to carry out certain functions. Participants in the task had to fixate on

Page 3 of 6

an object and either manually rotate it or mentally visualize its rotation along given axes. Comparing the
BOLD activation between the two groups, and finding activity in areas not typically associated with
motor processes (such as somatosensory cortex), Schlegel and his team concluded that the motor
network’s processing is part of a larger mental network that recruits multiple regions in carrying out
processes— such as rotation. Such results are especially interesting when considered in relation to studies
such as Downing et al. that found evidence for modular processing using the same fMRI BOLD
activation techniques.

Moving to research methods that aren’t often utilized in support of modularity, there are papers
that argue in favor of distributed processing with evidence based in logical and statistical models. These
papers, such as David Plaut’s “Double Dissociation without Modularity: Evidence from Connectionist
Neuropsychology” often specifically argue against the possibility of modular processing or the evidence
used to support, invoking arguments that are similar to those found in the psychological literature of the
debate. Although such papers are too complex to analyze here, however, there is another method used to
support the theory of distribution in the neuroscience community: reviews. In “Control without
Controllers: Toward a Distributed Neuroscience of Executive Control”, Eisenreich et al. discussed a range
of research papers hailing from different disciplines, and showed how the evidence in each could be used
to support distribution. One such example in the field of neuroscience is their discussion of the dorsal
anterior cingulate cortex. Here, Eisenreich and his colleagues noted over twenty different research papers
attributing the dACC to different functions (such as reward, adaptation to changing environments and
control of actions) through different kinds of experiments (physiology, neuro-imaging and lesion studies).
With so many distinct functions, they reason, it’s logical to state that the dACC is involved in multiple
processes rather than any specific one— and the same can be said of other regions.

Having reviewed the research above supporting both sides of this debate, and considering other
knowledge regarding human cognition, I believe that the organization of the brain’s processing system is
actually a combination of both modularity and distribution. The strength of the evidence for either side is
hard to refute, but once one looks closely at the nature of each study’s results, there seems to be a pattern:
most studies of lower-level cognition display a modular architecture, whereas most pertaining to higher-
level processing seem to be distributed. For example, in Downing et al. and similar studies, the brain

Page 4 of 6

regions implicated in modular processing are specific to basic functionality, such as identifying the human
body. On the other hand, more complex actions, such as visual rotation in Schlegal et al. or taking in an
entire visual scene instead of a single object in Valdez et al., yield evidence of distributed processing. And
this occurrence makes sense. It’s logical for more complex tasks to require large scale processing, and
there are advantages to both systems: modularity yields faster, more efficient processing, distribution
yields slower, more calculated processing. Furthermore, even an evolutionary argument can be made. The
brain could have evolved to handle new complex tasks by recruiting multiple brain regions, all with
previously distinct functionality, to carry out these processes instead of evolving new cortical regions.
Thus in the future, studies comparing human cognition with that of our primate cousins could shed some
light on the debate, along with studies that aim to address the possibility of both modular and distributed
processing within the same system.

Overall, this debate on how processing is organized in the brain is just as alive today in 2017 as it
was in 1986. Despite advances in research technology and the many creative methods that scientists have
devised to study this, there is yet to be that breakthrough paper that tips the evidence either way.
Nevertheless, the advancement of the technology that would benefit from a resolution continues forward,
if perhaps slower. With the state of the debate, it’s possible that developments in artificial intelligence or
more advanced decoding devices in neuroprosthetics could one day stumble upon a processing secret
that’s evaded neuroscientists all these years. As BMIs move from laboratories to clinical settings— and
eventually consumer technologies— researchers are forced to continue to developing new decoding
mechanisms without the desired knowledge of how neural encoding works. However, such bottom-up
methods of making better technologies might just make a connection that experiments haven’t. In the
event of such an occurrence, tech could provide that breakthrough that the scientific community needs to
advance to the next level in this debate, thus keeping up the rhythm of science and technology that so
drives the wheel of progress.