related texts back to overview
Summary
This reviewer did not obviously skip complete paragraphs as the first
'reviewer' did, and he wrote more. However, this review is still not a
scientific review.
The response to (1) is based on irrelevant evidence from primate
brains, the argument is not to the point, and it is full of misleading
statements. Like the first reviewer, he ignores completely the
dimensions that I explicitly gives in the first sentence of my
criticism.
The response to (2) is the worst: The reviewer first misquotes my
review as to distort its meaning, hence misinterprets it and brings
irrelevant evidence. Then the text deteriorates, through sloppy
thinking to plain demagogy.
The response to (3) is just a word game.
Like the first reviewer, this one ignores completely the question of
reproducibility that I raised in the last paragraph but one of my
review.
1) The authors ignore the higher level nature of the stimuli, i.e
their conceptual emotional significance
- [1] The underlying idea is serial processing:...
- This is against the evidence from neuroanatomy, which shows that
the flow of information in the other direction is of the same
magnitude. This knowledge appears in any neuroanatomy textbook. The
reviewer, like the authors of the paper, ignore this knowledge.
- [2] The design of the experiment attempts to minimize the emotional or
conceptual properties....
- But far from enough. For example, in the coherent motion, the
authors could use one of he pictures in either optical flow or
biological motion, thus reducing the differences. They could also use
in biological motion an object that has less conceptual content (e.g. a
dog). It is obvious that the authors did not consider this question
at all. Not using a video of a real person is useful in reducing the
amount of information in the input, but as long as people can
recognize anything in the input, this recognition cannot be ignore.
- [3] whether they are additionally ..
- This sentences are a subtle effort to get the user to believe
that there are no conceptual effects, because in most of the cases the
subjects will not be 'prompted to imagine...'. However, he ignores the
dimensions of the stimuli that I gave in the first sentence of my
criticism, and other similar dimensions, like how amusing it is.
In addition, we haven't got a clue of what the subjects do
unconsciously, and they may be 'prompted to imagine', but do it
unconsciously.
- [4] More accurately, to show all ..
- The use of the term all here is misleading. Howard et
al marked as 'differentially active' only voxels with differential
activity over some threshold. They don't actually give the threshold
they use, but they do mention the probability of false positive as
0.00005 (in a conditional sentence), so I assume that threshold for
the discussion. That means that a voxel with a lower apparent
differential activity, which gives a probability of false positive of,
say, 0.1, is not marked as 'active'. However, of these voxels, only 1
in 10 is false positive, and the rest 9 of 10 are really positive,
i.e. have a real differential activity, albeit lower (on average) than
the differential activity of the marked voxels.
The authors don't give us any clue of how many voxels are at this
level of apparent differential activity, i.e. how many voxels would
give false positive at probability of 0.1. However, since the authors
use a very stringent threshold, which leaves as marked only a small
number of voxels, it is likely that there are many more voxels below
the threshold but still above the relaxed threshold of 0.1, than there
are above it. Since almost all of these voxels (9/10) are really
positive (have real differential activity), and there are many more of
them, the number of voxels with differential activity which are not
marked outnumber the marked voxels. Thus the voxels that the authors
present are likely to be only small part of the voxels with
differential activity, and to represent only peaks of differential
activity, rather than 'all the differences'.
Note that this analysis does not rely on the exact numbers, but on the
fact that the number of marked voxels is relatively small, compare to
the total number of voxels. It is not necessarily true in any case,
but it is, by far, the most probable assumption, and a counter-evidence is
needed before dismissing it. Instead, the reviewer tries to dismiss it
simply by words.
- [5] ... supported by a vast body of evidence from experimental work..
- This 'vast body' does not tell us anything on the distribution of
the peaks of differential activity in this territory (MT), nor does it
tell us anything about specialization for specific kinds of movement.
It is therefore irrelevant. Here the reviewer simply rely on the
reader not to be familiar enough with the literature.
- [6] It would be nothing but perverse to attribute ...
- This is not to the point. The conclusions of Howard et al
do not rely merely on activity in the MT ('this territory'), but on
the different locations of peaks of differential activity in it. Thus
the question is not if the activity in the MT is coming from higher or
lower level, but what determine the location of peaks of differential
activity in it. The reviewer did not bring any evidence about this
point, and simply avoid discussing it.
- [7] The smaller patch ..
- It is not obvious to me how this discussion is relevant to the
point I raised. It certainly does not give any reason why we can
ignore high level effects.
I suspect the reviewer tries here to show how 'taking conceptual
significance into account' leads to a dubious interpretation. However,
in my criticism I did not say that the higher level effects have to
be part of the interpretation, I said "These effects must be
controlled in some way, ..".
- [8] The doubts attach ..
- This statement probably rely on the analysis in the previous
paragraph, which are faulty (comments 4-6), and therefore it is wrong
as well.
2) The authors assume that high differential activity is
equivalent to functional significance.
- [9] "the bulk of information in the brain is not coded in level
of activity, but in the pattern"
- This is a gross misquote. My original sentence is:
".. the bulk of the information in the brain is not coded in level of
activity (how many neurons are active in a specific region), but in the
pattern of activity (which neurons are active and when)."
My sentence was completely unambiguous. However, after the reviewer chopped
away the clarifying brackets and the last two words, the result is
ambiguous, because it can be interpreted as:
"the bulk of information in the brain is not coded in level of
activity [of each neuron], but in the [temporal] pattern [of spikes of each neuron]"
and this is the way the reviewer is going to interpret in the
following text (comments 10-12 below).
- [10] ".. as it essentially implies all scanning experiments are
worthless."
- This presumably refer to the reviewer misinterpretation. The
original statement cast doubt on the way these experiments are
interpreted, not on the data itself.
- [11] "However no evidence is cited to substantiated this
criticism..."
- The original sentence is obvious to anybody that knows anything
about the brain, and it is hard to believe that a neuroscientist would
ask for an evidence for it. The misinterpreted version of it is indeed
doubtful (in my opinion, it is completely wrong). However, this is
not relevant to my original sentence, or to the argument as a whole.
- [12] There is some, very recent ...
- These two sentences give a clear confirmation that the reviewer
did the misinterpretation that I outlined in [9] above. Apart from
this, they are irrelevant, as the question is about the importance of
the rate of differential activity in voxels, each of which contain in
the order of magnitude of million neurons, and single unit recording
cannot tell us anything about this.
- [13] The scanning evidence is also beginning to build a
self-consistent picture,...
- This statement is simply false, and neither Howard et al
nor the reviewer could bring any support for it. In fact, howard et
al could not even match their results to their own previous
results, as they themselves admit. There is evidence for correlation
between peak of activity and processing only for some special cases,
and only in coarse grain, as was discussed in the response to the
first reviewer.
- [14 It seems more likely that scanning may give too exuberant a
picture of activity,...
- I am not sure what the reviewer says here, but it is clearly not
opposite my criticism, which is: "differential neuronal activity does
not entail functional significance". It is probably opposite his own
misinterpretation, whatever that is.
- [15] In his initial comments Harpaz writes .... This is a clear
contradiction.
- That is plain nonsense. Processing is a change in pattern, but
that does not mean it cannot cause peaks of activity here and there.
- [16] The other reasons given, (b) and (c) are completely
incomprehensible to me...
- The reviewer seems not to believe that any input can escape
processing. That is simply sloppy thinking. There is no way to know
in advance what stimuli are going to enter the visual system, so
it cannot be designed perfectly to process all the input in a useful way.
There will always be a large amount of noise, i.e. activity that is not
process usefully. This 'noise' is even larger due to the fact the
system discards part of the input most of the time (the part that it
'doesn't pay attantion' to).
the claim that (c), which says "completely irrelevant effects", is
incomprehensible, cannot even be described as sloppy thinking. It is
obvious that the reviewer did not want to admit that irrelevant
effects are possible, but could not come up with any sensible argument
why they are not possible.
- [17] It is worth noting ...
- This paragraph is plain demagogy of the worst kind. The reviewer
intentionally mislead the reader to thinks that I put a 'complete ban'
on coincidence between activity and functional significance. What I
actually wrote is:
- " .. the processing of the input is not necessarily done where the
highest differential activity is seen."
- "Hence, regions where the processing take place does not necessarily
appear as peaks of differential activity."
- "Thus the equivalence cannot be assumed, and must be supported by
evidence."
- [18] I agree with the final ..
- Again, the reviewer repeat the blame that I ignore the 'body of
evidence', even though this body is irrelevant (5 and 8-12 above).
3. The authors regard localization as specialization.
- [19] Very likely both factors are at play ...
- This looks like a benign assertion, but it is in disagreement
with the facts. We know that the input-output wiring of the cortex
(including cortical neurons with axons that go outside the cortex)
does cause different parts of it to be dedicated to different
activities, so this factor is indeed very likely. As yet, nobody could
work out how any of internal features that vary across the cortex can
cause parts of it to specialize in anything, even though many people
tried to, and there is no experimental evidence for it. Thus, at the
moment, local structure and physiology of the cortex don't seem to
determine which region of the cortex is dedicated to what activity.
- [20] but Harpaz wants to...
- That is not 'harpaz wants'. The word 'specialization' is almost
universally used to refer to intrinsic property(ies) of the
specialized entity, rather than to the effects of other entities on
it. More relevantly, this is the way it is used in neuro- and
cognitive-sciences.
- [21] This might be a reasonable definition but it is not the one
used in cortical physiology.
- Using such a common word as 'specialization' in a different way in
cortical physiology is misleading for both neuro- and
cognitive-scientists, especially here, because it is the main 'take
home message' of the paper. It is even more important in a journal
like Current Biology, where most of the readers are not
cortical physiologists.
I also have my doubts about the claim itself. My impression is that
when Howard et al (and other cortical physiologists) say
'specialization', they do mean some physiological modifications, like
the rest of the world does. This is supported by the fact that the
reviewer fails to tell us what 'specialization' does mean in cortical
physiology.
- [22] the alternative "there is functional specialization for
motion in the mid-temporal" is neutral in this respect.
- But what does it mean? The reviewer did not tell us what
is the meaning of 'specialization' in cortical physiology. We
can interpret the reviewer as saying that it means something like
"assigned to" or "dedicated to", but it is significant that the
reviewer himself did not actually give any definition.