Page 190 - Artificial Intelligence in the Age of Neural Networks and Brain Computing

180 CHAPTER 8 The New AI: Basic Concepts, and Urgent Risks

FIGURE 8.14
Brain circuit corresponding to Fig. 8.9.

cite here give more details. I am also grateful that every one of these great people
(except for Scheibel, Connors, and Bliss) took the time at least once to try to explain
to me the real intuitive content of what they saw in their data.
It will take a lot of new effort by many new researchers to test out all the details
of my theory, or to update it in a way which still fits the kind of neural network math-
ematics which has a sound mathematical and engineering foundation. For that
reason, I recently decided to extract just two key hypotheses from the theory for
further testing: Does the brain really implement a kind of discrete time “cinematic
processing” (as in Fig. 8.10), which requires a kind of hard-wired regular clock to
control the overall learning process? Does the cerebral cortex of the brain actually
employ a regular alternation of a forward pass to compute its outputs, in the first
part of a clock cycle, followed by a backward pass to calculate modulated derivatives
and to adapt the main synapses used to generate coherent predictions of its environ-
ment? Except for the word “derivatives,” this is exactly what Barry Richmond of
NIH said to me in a long discussion as we walked to or from a meeting at the
Dana Foundation, but I could not find it in print in the published paper he sent
me. (Of course, these are just two of the predictions of the theory I published
long before.)
Why argue about forward and backward passes in the fast calculations of the
brain, when we can simply look directly at real-time data on brain activity with
enough resolution in time to test what Richmond asserted? Two major studies did
exactly that in 2016, and both support these two hypotheses.
The beautiful figure in the middle captures exactly what Richmond said he saw in
his data as well: alternating windows of processing, which seem to go in alternating
directions, separated by windows of relative quiet. These results came from a state of