The computing power inside your skull
vastly exceeds that of any supercomputer
in the world. But for the past half century,
neuroscientists have generally supposed
that the process by which the brain
achieves its phenomenal performance is
fundamentally similar to the way
electronic computers work. According to
the conventional view, thinking takes
place through the aggregate action of
billions of simple elements--cells called
neurons--that are wired up in an
extremely complicated way.
Supercomputers are likewise built of
millions of interlinked simple switches,
consisting of transistors on silicon chips
by companies such as Intel.

Recent research is forcing a re-evaluation
of this standard model. Individual
neurons, it turns out, are not so simple
after all: experiments have shown that
they can actually perform surprisingly
complex calculations and register fine
discriminations. It is even possible that
networks of interacting molecules within
an individual neuron might perform
specific computations, Christof Koch of
the California Institute of Technology
reported in the January 16, 1997 issue of
Nature. The organ of thought is looking
far more complex than scientists believed
just a few years ago.

Koch's conclusions are based on studies
of the precise electrical behavior of cells
in the brain. Neurons conduct signals in
the form of tiny electrical impulses,
known as spikes. Messages travel from
one neuron to another as pulses of
chemicals that are released at specialized
junctions, or synapses; there are trillions
of such junctions in the human brain. How
and when synapses relay messages
between neurons is crucially important in
controlling mental activity. Moreover,
neuroscientists believe that learning
occurs through a change in the strength
of certain synaptic connections. A
frequently-used synapse becomes
stronger, whereas an infrequently used
one may grow weaker over time.

Researchers have long understood the
basic division of function in the neuron.
One set of branch-like extensions from
the cell bears incoming synapses; another
set of branches, usually located at the
end of a long threadlike extension,
processes outgoing messages. In the
traditional view of the neuron, which goes
back to experiments conducted in the
1940s, the cell functions as a fairly
simple on-off switch. A spike would be
initiated in a neuron if the total amount
of stimulation at all the incoming
excitatory synapses exceeded some
critical level. (Conversely, if the neuron
received enough inhibitory synaptic
signals, it would stop producing spikes.)

Yet as Koch observes, scientists have
discovered that neurons actually have
numerous electrically-active components
in the incoming branches. These active
components, which include the NMDA
receptor, a protein that spans the
neuronal membrane, modify the effect of
incoming messages. For example, the
active components ensure that spikes
received at synapses that are adjacent to
one another carry more weight than
spikes received at widely-separated
synapses. Computer simulations show
that active elements probably multiply
the influence of adjacent synapses, rather
than merely adding them together as the
traditional neurologists had supposed.
This finding adds a layer of complication
to the picture of how the brain works.

And the story gets still more involved.
Koch notes that the conventional idea
that the timing of individual spikes is
unimportant turns out to be quite wrong.
Researchers had generally supposed that
the representation of information in the
brain depends essentially on the overall
rate of firing of the neurons. But
experiments over the past few years have
shown conclusively that some cells in
monkeys' brains can adjust the intervals
between spikes in increments as little as
one hundredth of as second. Moreover,
the temporal patterns of spike activity
across different neurons is sometimes
controlled with an even finer accuracy of
about one thousandth of a second.
Contrary to the common wisdom, "the
brain appears to care a great deal about
timing," Koch says.

These results raise a new question: what
is the purpose of all of that very precise
neural timing? Koch points toward
breaking research that may offer a clue.
Spikes, once initiated in a neuron, do not
propagate only in the "forward"
direction--that is, toward the synapses
that relay outgoing messages. Rather,
experiments on isolated brain tissue
indicate that spikes also move backwards,
up the neuron's input branches.

The effect that these back-propagated
spikes have on the active components of
the brain--if indeed the phenomenon
occurs in intact animals-- is far from
clear. But a study published in the
January 10, 1997 issue of Science by
Henry Markram of the Weizmann Institute
for Science and his collaborators suggests
that back-propagated spikes can
dramatically influence the way a neuron
processes an impulse. The precise order
in which one spike arrives at a synapse
and another one back-propagates to the
receiving neuron greatly influences the
subsequent strength of a synapse,
Markram's group showed. If the
back-propagated spike arrives first, the
synapse is weakened; conversely, if the
back-propagated spike arrives second, the
synapse is intensified.

This unexpected phenomenon might
specifically boost the synapses that are
conveying messages while suppressing
random or unimportant signals. Other
research conducted in the past year has
shown that synapses can quickly adapt to
the incoming rate of spikes at different
synapses. In this way, the synapses can
remain highly responsive to any sudden
changes in electrical activity over a large
range of background levels.

Taken together, Koch believes, these new
insights into the capabilities of the single
neuron suggest the brain should really be
viewed as a hybrid computer, one that
employs both digital pulses (between
neurons) and analog computations (within
them). The brain, then, is quite unlike a
digital computer in its basic
underpinnings. Even if the brain is built of
hybrid digital-analog neurons, it need not
have any computational powers that are
utterly beyond those of a simple digital
computer. But simulating such a brain
using hybrid digital-analog elements may
take thousands of times longer than it
would take to simulate a brain built of
the same number of simple neurons,
points out Douglas R. Hofstadter of the
University of Indiana.

So humans have their unique uses after
all. Intel will not, it seems, be coming out
with a replacement for your brain anytime
soon.

--Tim Beardsley, staff writer