40 • CHAPTER 2 Cognitive Neuroscience
is illustrated in ● Figure 2.21, which shows that Bill’s face
would be signaled by the fi ring of neuron 1, which responds
only to his face; Mary’s face is signaled by the fi ring of neu-
ron 2; and Ramon’s face by the fi ring of neuron 3. Thus,
specifi city coding proposes that there are neurons that are
tuned to respond just to one specifi c stimulus.
The idea that there might be single neurons that
respond only to specifi c stimuli was proposed in the 1960s
by Jerzy Konorski (1967) and Jerry Lettvin (see Barlow,
1995; Gross, 2002; Rose, 1996). Lettvin coined the term
grandmother cell to describe this highly specifi c type of
cell. A grandmother cell, according to Lettvin, is a neu-
ron that responds only to a specifi c stimulus. This stimulus
could be a specifi c image, such as a picture of your grand-
mother; a concept, such as the idea of grandmothers in
general; or your real-life grandmother (Gross, 2002).
But there are problems with this idea: (1) There are
just too many different faces and other objects in the
environment to assign specifi c neurons to each one; and
(2) although there are neurons that respond only to spe-
cifi c types of stimuli, such as faces, even these neurons
respond to a number of different faces. Thus, a neuron
that responds to Bill’s face would also respond to Roger’s
and Samantha’s faces. Because of these problems, the idea
of a highly specifi c grandmother-type neuron has not been
accepted by researchers.
The generally accepted solution to the problem of neu-
ral coding is that a particular face is represented not by the
fi ring of a single neuron, but by the fi ring of groups of neu-
rons. For example, let’s consider how the three neurons in
● Figure 2.22 fi re to a number of different faces. Bill’s face
causes all three neurons to fi re, with neuron 1 respond-
ing the most and neuron 3 responding the least. Mary’s
face also causes fi ring in all three neurons, but the pattern
is different, with neuron 3 responding the most and neu-
ron 1 the least. All three neurons also fi re to Ramon’s and
Roger’s faces, but with their own individual patterns.
Thus, each face is represented by a pattern of fi ring
across a number of neurons. This solution to the problem
of neural coding is basically the same thing as the idea of a
“chorus” of neural fi ring that we described when consid-
ering how feature detectors could represent a tree. This is
called distributed coding because the code that indicates a
specifi c face is distributed across a number of neurons. One
of the advantages of distributed coding is that the fi ring of
just a few neurons can signal a large number of stimuli. In
our example, the fi ring of three neurons signals four faces,
but these three neurons could also signal other faces, which
would have their own pattern of fi ring. (The similarity of the
terms distributed coding and distributed processing might
cause some confusion. For our purposes, distributed cod-
ing refers to the pattern of fi ring of a number of individual
neurons, and distributed processing refers to the activation
of a number of different areas of the brain.)
What all of this means is that our ability to identify
and recognize the huge number of different objects in our
environment is the end result of distributed cooperation● FIGURE 2.20 Firing rate, in nerve impulses per second, of a
neuron in the monkey’s temporal cortex that responds to face
stimuli but not to nonface stimuli. (Source: Based on data from Rolls &
Tovee, 1995.)Faces NonfacesFiring rate01020Bruce Goldstein● FIGURE 2.21 How faces could be coded by specifi city
coding. Each faces causes one specialized neuron to respond.
(Source: B. Goldstein, Sensation and Perception 8th ed., Fig. 2.21, p. 36.
Copyright © 2010 Wadsworth, a part of Cengage Learning. Reproduced with
permission. http://www.cengage.com/permissions.)Stimulus Neuron 1 Neuron 2 Neuron 3(a) Bill(b) Mary(c) RamonCopyright 2011 Cengage Learning. All Rights Reserved. May not be copied, scanned, or duplicated, in whole or in part.