Editor (T. Sejnowski) comments:
This paper should be rejected on the strength of two negative reviews and my reading, which confirm the conclusion that this paper does not deserve publication, nor is there a likelihood that any revision would improve rating.
[ This reviewer completely ignores the central argument of the paper. There isn't even a single referene to sections 1-6 of the paper.]
Review of "Neurons in the brain cannot implement symbolic systems", by yehouda Harpaz.
The author purports to show that brains cannot implement symbolic processing operations because they cannot copy tokens reliably. But he admits in the first paragraph of section 7 that people can in fact handle symbols. (And contrary to the claim made a few paragraphs later, such symbols are not necessarily tied directly to sensory inputs, since people are perfectly capable of manipulating sentences, formulas, diagrams, and concepts they dream up in their heads.)
People can do various things inside their heads, but not necessarily symbolic operations. Manipulating sentences, formalas, diagrams and concepts does not necessarily requires symbolic operations.So the author tries to imply that just because people can handle symbols, this does not mean that BRAINS can handle symbols. Actually, he doesn't say something quite that ridiculous; he argues (p. 14) that "the components of the brain" cannot handle symbol. But this is an entirely different claim than the paper start out with and cocludes with, i.e. that the brain as a whole is not a symbol processor. Since no one ever claimed tha individual neurons or small groups of neurons were symbol processing systems, the author's central argument is revealed as an attack on a straw man.
The claim that the brain is a symbolic system means that components of it do symbolic operations, namely manipulating symbol tokens. This is the position of the symbolic theorists, and the models of the brain that are based on it, and I discuss this in section 2. This reviewer clearly haven't read this section.
The paper concludes with a long series of silly statements, such as (p. 18) "handling language does not require any dynamic mechanism" because the meanings of words are static. (What about the need to dynamically construct representations of novel sentences?)
That does not necessarily require a symbolic system. The argument from language is based on the fact that the language itself is made of symbols (words), but because these are static, they do not require a symbolic system.And (p. 21) "Neuroscientists do not realize the importance of the stochastic connectivity for theoris of cognition".
So what is silly about that ?
Obviously, this is not a scholarly paper. And the central argument is nonsense.
Impressive statement, considering the fact that this reviewer did not discuss the central argument at all, and does not even give us a hint that he has read it.I recommend that the paper be rejected.
review of "The neurons in the brain cannot implement symbolic systems" by Y. harpaz.
In the manuscript the argumentis made, that the stochastic nature of cortical connectivity precludes the interpretation of brain function as a symbolic processor.
Very inaccurate. I am not talking about interpretation of the brain, but about what the brain is.Although I sympathize with this view, I do not believe that several of the basic assumptions are true, nor can I follow the line of argumentation. Therefore, I find the manuscript not acceptable for publication in JCN.
1) The connectivity ... does not change as part of the computation. here the author explicitedly excludes to consider changes in synaptic strength.
The reviewer gives the impression that I ignore changes in synaptic strenth. What I actually do is to define 'connectivity' to mean only existence or non-existence of connection. The changes in synaptic strenth are discussed in the next paragraph of my paper, which the reviewer seems to have skipped altogether.
I do not understand the exclusion of changes in synaptic efficacy. e.g consider the report by markram and tsodyk, that the effect of an presynaptic action potential very much depends on the afferent avtivity. Thus, an existing connection can rendered ineffective by the sheer occurence of an action potential shortly before. What is the difference on ineffectual and a non existing connection?
That non-existant connection will continue not to exist, while ineffective one will become effective. Is it realy too difficult for the reviewer to understand this?
Furthermore there is no positive evidence of anatomical changes on a timescale of thinking - what is that timescale anyway?
Since humans can do mental operations in less than a second, the time scale of thinking is less than a second, i.e. several tens to several hundreds millseconds. How come the reviewer does not know that?On the other hand, present day techniques can not exclude dynamic changes on a timescale of seconds, i.e. they might be relevant for solving symbolic manipulations.
The lack of evidence is in agreement with what I am saying. Maybe the reviewer missed a word here.
Present day techniques can easily exclude changes in times cales of seconds in the connectivity inside the cortex of mammals. The reviewer simply displays his ignorance.
At least, the asumptions made here do not rest positive text book knowledge, as claimed in the article.
Again, a display of ignorance.
2) The connections that each individual neurons form are stochastic (i.e. they are not specified accurately by any mechanism).
In the paper, this is the beginning of a paragraph that ends by saying that I will discuss it in the next section. The reviewer simply ignores this.
Along similar lines as above. Right now we do not understand the rules if any determining synaptic connective, Thus, local random connections is a sort of zero hypothesis. On the other hand, there is no positive evidence that this hypothesis actually applies,
Yes there is, and it come from comparison between brains, as discussed in the next section. This was pointed in the last senetence of the paragraph of the quoted sentence. The reviewer simply does not read the paper as it is written.and recent results on the connectivity between retina and lgn (stryker et al.) indicate that a pathway where no apparent order was visible for many years, might reveal such an order, when investigated with the appropriate means.
I don't know what the Stryker paper is, but the retinal to lgn is clearly relatively ordered connection (compared to the cortex), and you cannot deduce from one to another. The argument for stochastic connectivity is not based on 'order is not visible', but on differences between brains. You need more than hope to argue against that.As a second example might serve the specificity of tangenial connections in the cortex. Where a bias of cells with similar physiology properties to form anatomical connections has been found.
This kind of bias does not help at all in reaching order, as the ratio between number of neurons and number of types is huge, so there is huge number of neurons of each type. The selection between these is still stochastic, and therefore the difference the bias is making is tiny.Thus, given the present rate of results, it is very well possible that we think about precision of cortical connections in different terms in only few years.
Again, using hope as an argument. As I discuss in section 4 of my paper, the 'precision of cortical connections' is already out of question, and the reviewer's assertion is simply a declaration of ignorance.
3) In too many places phrases like "It is clear ..." It is obvious..." are used. Sorry, most are not clear or obvious to me. take the second paragraph on page 8, incidententally a crucial one for the whole argument.
First, it is NOT clear that the structure of axon trees of individual neurons are not well specified. To conclude from the subjective impression of the complexity of axonal arbors to the randomness of connectivity is just wrong.
In the text, I say:"The evidence for this [stochastic connectivity] is from comparison of the axon trees of different neurons, within the same brain and from brains of different animals of the same species. It is clear that the structure of the axon trees of individual neurons are not well specified." i.e. it is clear because of the comparison between brains. The reviewer discusses the second sentence as if the first one did not exist.
Clearly, the reviewer cannot be stupid enough to have missed the first sentence, so he must be pretending.
Second, what is the relevance of the comparison of different brains?
And now the reviewer discusses the first sentence as if the second sentence (and the rest of the section) does not exist.Is a computer not a symbol processor even if it is the only one of its kind?
No. The differences between brains are evidence for the stochastic connectivity, not directly against symbolic systems, and the text is totally clear about that. The reviewer here obviously pretends that he is stupid enough to fail to make the connections between the two sentences that I quoted above. Note also that the whole section is about stochastic connectivity, and I don't start discussing symbolic systems until the next section, so by default every sentence in it should be about stochasticity.Furthermore, it ignores the influenece of development. The influence of experience onto the formation of cortical connections has been shown in many areas. That this experience is not exactly matched between different individuals may lead to differences in the physical appearance of the respective brains, so what? Is this an argument that, as the brain look different, they cannot implement the same rules?
It took me time to understand what the reviewer actually wants to say, so maybe I still misinterpret him. My interpretation is that he wants to say that even though the brains look differently, the initial connectivity is the same, and it is the environmental effects that cause differences in the physical appearance. That is another demonstration if ignorance, because the connectivity of the brain is stochastic at birth as well.
4) The term stochastic is used inconsistenly. On one hand it refers to the comparison between different brains, on the other hand on the mapping of different activation patterns. Where is the connection?
I defines explictly what I mean by stochasticity (lack of relations between features inside the same brain) and discuss it in section 4, and use to mean the same thing in all the paper. It certainly does not refer to 'the comparison between different brains', this is just the evidence for it. As to mapping, except a mention about topographic mapping in section 4, the word does not appear in my paper.What is the metric to compare different activation patterns? In any instantiation of a brain there is a specific mapping, thus what does it means mean to call it stochastic?
Why do I need a metric? I don't actually compare activation patterns. The only question is whether they 'point' to the same thing, and if they are stochatstically related, they can't.
What is the 'specific mapping', and how it ralates to the neurons and connectivity of the cortex ? My discussion is about the real brain, not about abstract terms. The reviewer seems to have some theory of how the brain actually works, and based these question (and probably his judgement in general) on this theory, rather han real phenomena.
5) The time scale of storage must be shorter than the time scale of computation.
This relates to the argument above.
Here the reviewer is being very misleading. The statement is actually taken from my discussion of symbolic system in section 2, before all the argument above. It is part of the requirements of symbolic systems.In the cortex, the expression of LTP(if we take this as related to memory) might take minutes, however, it can be induced on a very short time scale.
How does that relates to the requirements of symbolic systems? They need the result of the storage, not only inducing it, and anyway, the argument is that the LTP must be be based on activity, which is based on connectivity, which is stochastic.Furthermore, there might be different mechanisms for storage for different time scales, i.e. reverberating activity - incidentally, this can be developed to an argument against random connectivity, otherwise, how could the sequence X->Y->Z->Z lead to reactivation of X? - which then might lead to long term storage.
Symbolic systems require immediate storage, in addition to long term storage, and the problem of stochasticity is the main one.
The nested argument (beteen 'incidentally' and '?') is amusing. The reviewer assumes that there are reverberating activities, and hence that the connectivity cannot be random. The possibility that there aren't reverberating activities (at least not ordered ones) in the cortex seems not to even cross his mind.
6) Why is it necessary to able to store any symbol everywhere?
Because symbolic systems require it (and it is storing symbol tokens, not symbols). Another demonstartaion of stupidity.Can you store visual icons in auditory cortex?
A visual icon is not a symbol token, so this is totally irrelevant.What is the meaning of anywhere as long the functional unit of brain dynamics has not been elucidated, i.e. locationalist and holistic theories compete?
My argument is independent of the meaning of 'anywhere', as long as it is in the cortex.
Back main page