[From Bill Powers (2004.12.30.0630 MST)]
Bruce Gregory (2004.1229.1738)–
Fair enough. Here follows a
brief statement of Hawkins’ predictions. I
leave it to you to describe the corresponding PCT-based
predictions.
- We should find cells in all areas of the cortex, including
primary
sensory cortex, that show enhanced activity in anticipation of a
sensory event, as opposed to in reaction to a sensory
event.
If “anticipation” is the setting of a reference signal to a
nonzero value, it would be expected that the output functions of the
systems doing the anticipating would be active, sending reference signals
to the systems that are anticipated to receive the sensory event in
question.
When you say sensory “event”, are you specifically referring to
a stereotyped set of configurations and transitions that create a
space-time pattern with a beginning, a middle, and an end? That is how an
“event” is defined in HPCT. Or does “sensory”
refer to what are called intensities in HPCT? Of course if you’re
speaking of anticipation, the implication is that something is to
“happen”: or “occur” at an identifiable moment. It
would be hard to identify the onset of a perception like
“spinning” or “beside” or “glissando.”
I presume that Llinas mentions the work of E. Roy John, 1972. John says
“A substantial body of data from human beings not only supports the
contention that these released patterns of electrical activity actually
correspond to the activation of specific memories, but establishes
unequivocally that there is a subjective correlate to the appearance of
these released potentials.[HERE IS THE RELEVANT PART] When an
event expected by a man does not occur, a cerebral potential appears at a
latency [delay] similar to that of potentials usually evoked by the
expected stimulus. Evoked potentials elicited in man by absent but
expected events have been reported by numerous workers, including most
recently Weinberg et al. We earlier reported similar findings in the
cat… Working in our laboratory, Herrington and Schneidau demonstrated
that in some subjects the shape of the waveshape released when a
particular geometric form was imagined closely resembled the waveshape
evoked by the actual presentation of that visual stimulus.” (B:CP:
p. 227).
This could be interpreted as I propose above, or it could mean that the
reference signal that is set results in a match through imagination, so
the anticipatory potential is actually a replay of a memory of the
stimulus into the perceptual pathways (as proposed in B:CP). John’s
emphasis is on memory phenomena, but it clearly applies to anticipation
as well.
- The more spatially specific a
prediction can be, the closer to
primary sensory cortex we should find the cells that become active
in
anticipation of an event.
I interpret “more spatially specific” (a rather vague term) as
“lower order.” HPCT would then agree with this item. Of course
it would not agree with the contention that “prediction” is
involved.
- Cells that exhibit enhanced
activity in anticipation of sensory
input should be preferentially located in cortical layers 2,3, and
6
and the prediction should stop moving downward in the hierarchy in
levels 2 and 3.
I’ve said that I expect the imagination connection to exist mainly at
higher levels. It can’t exist at order 1, and when it exists at order 2
we tend to treat the result as hallucination, not memory or imagination.
In HPCT reference signals, not predictions, move downward in the
hierarchy.
- One class of cells in layers
2 and 3 should preferentially receive
input from layer 6 cells in higher cortical regions.
In HPCT, cells in comparators at lower orders receive input from higher
orders, but I don’t know anything about what specific layers do. Of
course systems at all levels receive input from systems at the next
higher level. The standard labeling of cortical layers does not necessary
correspond to position in the HPCT hierarchy.
- A set of “name”
cells described in prediction 4 should remain active
during learned sequences.
Those cells weren’t described in your item 4 so I can’t say anything
here. Is a “name cell” connected with categorizing lower-order
perceptions?
- Another class of cells in
layers 2 or 3 (different from the name
cells referred to in predictions 4 and 5) should be active in
response
to an unanticipated input, but should be inactive in response to an
anticipated input.
Nothing specific to say about this, but see next item. “Active”
isn’t a very specific term for what a cell’s output means, is
it?
- Related to prediction 6,
unanticipated events should propagate up
the hierarchy. The more novel the event the higher the
unanticipated
input should flow.
This just corresponds to the perceptual hierarchy in HPCT. Of course the
only way to know that an anticipated event produces internal signals is
for the event NOT to occur; it is then created in imagination (see
above). If it does occur, there’s no way to measure the imagined signal;
it’s replaced by the sense-based signal. So “unanticipated”
just means “real” or “based on actual sensory input.”
And then HPCT says that the results of a disturbance should propagate
upward to whatever level is controlling the changed perception (the level
that “expects” it), but not above that (if control is
successful).
Completely novel events
should reach the hippocampus.
Too vague to comment on. What the heck is a “completely novel
event?”
- Sudden understanding should
result in a precise cascading of
predictive activity that flows down the cortical
hierarchy.
Reorganization that alters a high-level control system should result in
altering reference signals, with resulting control activities, at all
lower levels.
- The memory-prediction
framework requires that pyramidal neurons can
detect precise coincidences of synaptic input on thin
dendrites.
Does this mean that the logic level is implemented by pyramidal cells?
There’s no other reason I can think of that precise coincidences should
matter.
- Representations move down
the hierarchy with training.
I mentioned yesterday that tasks initially considered cognitive would
tend to be replaced by control at the lowest level where the same result
could be obtained.
- Invariant representations
should be found in all cortical areas.
In HPCT, computation of invariants is what perceptual input functions do
at all levels of organization.
Best,
Bill P.
