Page 106 - Artificial Intelligence in the Age of Neural Networks and Brain Computing

94 CHAPTER 5 From Synapses to Ephapsis

the neuron measure microscopic activity, electrodes outside the neuron measure
mesoscopic activity of a group of neurons.
Freeman chose the olfactory perception for detailed study because it is dominant
in most animals. Freeman’s team implanted 8 8 electrode arrays arranged in
4 4 mm spatial layout on top of a small section of olfactory bulb in several rabbits.
It is known that even if most of the bulb has been surgically removed, animals can
still discriminate odors regardless of which part remains. This holographic-like
encoding enabled Freeman to detect the pattern of activity covering an entire
bulb, even though 4 4 mm sensor window covers only a fraction of it. The com-
mon wave has a different amplitude at each location of the bulb, so the wave serves
as a carrier wave in the gamma range with a spatial pattern of amplitude modula-
tion (AM) throughout the bulb. Freeman was surprised to find a spatial pattern of
bulbar activity that was nearly invariant with the same odorant. Remarkably, the
oscillations of the dendritic potential have the same waveform over the whole
bulb. This must mean that the common wave is due to the interactions of neurons
throughout the entire bulb. The patterns are therefore created by neuron popula-
tion, not imposed from outside. The conclusion is that neural populations form
fields, not networks.
Each pattern is as unique for each rabbit as is the individual history of the animal.
Information processing view would require the brain to store, accumulate, and
average sets of AM patterns in a training period and then retrieve the average pattern
as a standard template against which to compare all incoming patterns during a test
period, not only with one average AM pattern but with all other average patterns, to
find the best match. Freeman’s data shows that brains do not have the neural machin-
ery to perform these engineering operations and even if they did, they would not
have the time needed to run them. For Freeman, Hebbian assemblies modeled by
neural networks are a mere ignition mechanism for the sustained ephapsis (e-field
created by oscillating neurons) which provides continuous fuel for the real-time
embodied cognition.
Initially, mainstream researchers did not pay much attention to Freeman’s
finding. More recently however, the discovery of place cells, for which Edward
and May-Britt Moser and John O’Keefe were awarded Nobel Prize in the 2014, pro-
vides empirical confirmation of Freeman’s insight. O’Keefe and his student Michael
Recce provided experimental support showing that the spikes of place cells shift sys-
tematically relative to the phase of the ongoing theta oscillation. They called the
phenomenon “phase precession.” Thus the phase of spikes and the animal’s position
on a track are correlated. Hippocampus seems to employ gain control mechanism to
assure that the relationship between the position and spike phase is independent of
firing rate or the speed of the animal and depends only on the size of the place field.
The phase-precession is the first convincing example of the critical role of oscilla-
tions in brain function.
Buzsaki [3] rejected the idea of place cell forming a simple map and concluded
that the current position of the animal is embedded in the representation of the past
and the expected future in terms of both location and time. The currently coded item