180 HOW THE BRAIN MAKES POTENTIALby experience, and not by built-in rules and linear pro cessing. Th is is
shown in many ways.
We can now be fairly sure, for example, that the messages provided by
sensory systems are self- organized into more compressed form. Th e sys-
tems do not simply pass on in de pen dent signals. Knowing a grammar is
far more effi cient than having to store volumes of individual items. Th e
fact that the ret ina contains more than one hundred million photorecep-
tors feeding into one million ganglion cells already indicates that the
network must perform signifi cant data compression. We do not know
exactly how to describe the message sent from the ganglion cells, except
that it must be some aspect of that deeper structure.
In the brain itself, it is now clear that neurons function as cooperative
ensembles that learn from experience. At the network level, this is attested
to by the many studies revealing the interdependence among diff er ent
centers (e.g., visual, hearing, olfactory) and among neurons in centers. At
the synapses, dendrites integrate multiple signals as complex correlations,
not as in de pen dent inputs. Across the network, “the majority of interac-
tions in highly interconnected systems... are multiplicative and syner-
gistic rather than additive.”^11 Th is is sometimes called “context- dependent
coding” and makes it clear that the language of the brain uses the structure
of multidimensional, spatiotemporal batteries of signals, not discrete cues.
Th e early studies of neuron responses used single recording electrodes
inserted into single neurons (in anesthetized animals). And they used
simple artifi cial stimuli, such as light spots or lines fl ashed on the eye. It
is now pos si ble, however, to rec ord si mul ta neously from hundreds of neu-
rons in the same or diff er ent brain centers. And more realistic, natu ral
stimuli have been used. Th e results have been striking.
Tai- Sing Lee and colleagues fi rst developed movies of three-
dimensional natu ral scenes and analyzed the statistical structures in the
data (including its fractal, or deeper correlational, structure). Th en they
implanted microelectrode arrays in experimental animals so as to rec ord
si mul ta neously from hundreds of neurons in the visual cortex. By then
exposing the animals to the movies, the researchers could assess whether
the neurons actually use that structure.
Th ey showed, fi rst, that the neurons do indeed respond preferentially
to the statistical structure previously mea sured in the movies. Th en they
This content downloaded from 139.184.14.159 on Tue, 17 Oct 2017 13:54:00 UTChttp://www.ebook3000.com