12 Jan 98 Yehouda Harpaz Harlequin Ltd. Barrington Hall Barrington Cambs. CB2 5RG UK Yeh_harlequin.co.ukDear Axel Cleeremans,
Re: ms. #39-97:Harpaz. "The neurons in the brain cannot implement symbolic systems."
I got your letter rejecting the paper above and the reviews. After reading the reviews, I request that you reconsider your decision, based on the detailed comments which I attach. These comments demonstrate that all the counter-arguments that the reviewers bring are vacuous. Neither has found any real flaw in my argument.
Considering your summary of the reviews, You say:
... Your argument rests on poorly-supported speculations about the role and nature of stochasticity in the pattern of connectivity between neurons in the brain
My text does not contain any speculations, and when Dr. Perruchet's tries to claim that it does he simply demonstrates his ignorance. All the assertions about neurobiology, including the stochasticity of connections, are well established knowledge. To that end, it probably be useful if you ask a neurobiologist to act as an "expert witness", and to review sections 3 and 4 only. This will establish the correctness and the universal acceptance (by neurobiologists) of my assertions about neurobiology.
... and on erroneous assumptions about the implications of the physical symbol system hypothesis
I discuss the requirements, not the implications. Neither of the reviewers point to any erroneous assumptions. Dr. Jimenez has expressed doubts, but these are based on lack of knowledge about the implementation of symbolic systems. Again, I suggest you use an expert on symbolic system as an "expert witness", to review the requirements that I list in section 2.
(2) that you fail to offer any convincing (alternative) account of how human beings are capable of processing symbols that is not a rehash of well-known ideas, such as the ideas embodied in the connectionist approach.
You also say:
My argument does not need any such alternative explanation which is different from connectionist approach. You probably have been misled by the treatment of Dr. Perruchet of section 7. In this case my comments below will clarify the point.
I don't think that this is true. I did not find any discussion of the problem of implementation of symbolic systems on top of stochastic connectivity. The two reviews of my paper confirm this, because both of the reviewers clearly lack any familiarity with the facts and concepts associated with this topic.
I Attach the full text of both reviews (in italics), with my comments indented.
The European Journal of Cognitive Psychology.
Author: Yehouda Harpaz
Title: The neurons in the brain cannot implement symbolic systems.
The author claims that the neurons in the brain cannot implement symbolic systems, because the connections between neurons are not well defined at the individual level. Although this view may appear, at first glance, as challenging and stimulating, I believe that it does not hold after further scrutiny, for several reasons.
First, the author does not question the claim that the brain can implement symbolic systems, as he contends, but, far more restrictively, he challenges a strict computer metaphor of the central nervous system.
That is a blunt lie. I make it very explicit what is that I claimed is not possible (symbol tokens), and this is not restricted to computer metaphors. This kind of lies would make a political speech writer proud, but has no place in a scientific text.
What is not possible, according to the author, is that a symbol token, implemented as a pattern of neuronal activation, may transfer from a neuronal structure to another one.
Apart from some distortion (I discuss all neuronal features, not only patterns of neuronal activation), this is reasonably accurate. However, this is the last reference to symbol tokens, and in the next sentence the reviewer returns to his bogus line of attack.
The underlying model is the computer structure, in which the information may transfer from a system (for instance, ROM or RAM) to another (for instance, CPU) without any losses.
Repeating the blunt lie from two sentences earlier.
The problem is that I doubt that we can find anyone willing to advocate that actual brains are built as computers. Among many other reasons, there are far more direct neurobiological counter-arguments than the one put forth by the author, such as for instance those stemming from studies on brain injury (the author himself mentions in P.18 that the assumption that the brain has a computation centre "is in conflict with evidence from brain damage"). Obviously, we can think of symbolic systems without relying on a computer metaphor of the central nervous system.
Nice blurb, but irrelevant to my text, which is about symbol tokens, not computer metaphor.
My second concern is about the so-called stochastic nature of the connections of individual neurons.
That is a demonstration of ignorance, as the stochastic nature of the connections is a well-established fact.
The author infers this from the observation of differences between brains.
Another piece of blunt demagogy. I repeatedly say that this is a well establish fact, not `my own inference.' (abstract, first paragraph of section 3, last five paragraphs of section 4).
I wonder why we should expect to have the very same network between individuals, and between the two hemispheres in the same brain.
We shouldn't, and I never wrote we should.
The claim that (p. 8): "Since the low-level connectivity is different between individuals, it cannot be specified during development (by the genes or otherwise), and hence must be stochastic" is groundless.
This declaration is based solely on the ignorance of the reviewer. The quoted assertion is regarded as obviously true by anybody that has any knowledge in neurobiology.
We know that individuals differ by both genes and environment on the one hand, and that neural structure growth depends on both kind of factors on the other hand. It should be a deep mystery if all nervous systems turned out to be strictly identical at the end. This does not mean that they are stochastic.
That is nonsense, because both the effects of the environment and of genetic variability are stochastic with respect o the connectivity in the brain, so any differences they cause will be stochastic as well. To account for a different connectivity which is not stochastic, there has to be a factor which is different between individuals, yet has high correlation with the connectivity in the brain. There clearly isn't such a factor (and the reviewer cannot think of any), and that is why all neurobiologists agree that the connectivity is stochastic. I make this point in the top of p.9.
The third main problem is about section 7, entitled:"If neurons cannot perform symbolic operations, how can the whole person perform symbolic operations?". It is indeed a good question, but I find no good responses in the paper.
The reviewer uses here a a more subtle demagogy than before, so it is important to understand what happens here. Section 7 of the paper is not part of the main thesis of the paper, which is made in section 2-6, so even if the reviewer could refute what I am saying in section 7, it is not affecting the main thesis of the paper.
Anyway, the reviewer does not even come close to refute what I am saying. In section 7, I am answering a possible objection, and all that is required is to show potential differences between the capabilities of the whole person and components of the brain. To that end, I bring two differences. Each one of these is sufficient, so the reviewer needs to refute them both. Instead, he ignores completely the first one, and argues against the second one by distorting it.
The second argument, "the brain implements a learning and thinking system", is specially puzzling. It is told here that we "are not limited to the capabilities that are inherited in the genetic make-up of the person".
After dismissing the first potential difference simply by ignoring it, the reviewer goes on to distort the second one. The last quote is accurate, but out of context. In the text, I say: `These new skills are not limited...'. The reviewer changes this to: `We are not limited...'. This is important, because if the reviewer has quoted me properly, the reader of the review would have noticed the sudden appearance of `new skills', and may be tempted to re-read my text again. In this case, the reader would have found that I gave a list of actions that the whole person can do while components of the brain cannot do, and are used to acquire these skills. The reviewer tries to make the reader forget about this.
Therefore, according to the author, the brain implements a learning system.
Now that reader forgot my text (hopefully), the reviewer introduces a bogus interpretation of it.
However, "components in the brain do not have these learning capabilities, so they are limited in a way that the whole person isn't." I am not very familiar with the various textbooks on which the author relies, but if these textbooks do not mention that a neural network has learning capabilities, they are in need of serious revisions.
And using the bogus interpretation, the reviewer tries to show that I am ignorant of neural networks.
Given these major drawbacks, I pass over in silence a lot of other problems (such as the allusion in p.20 to author's own model -a model that "can explain thinking"- without any details nor references).
The reference to my model is in a side comment. As far as I know, a reference to the author's own unpublished work is acceptable provided that it does not form part of the main thesis of the paper. If this is the worst of the `lot of other problems', it probably means there aren't many of them.
Overall, I believe that this paper is unsuitable for EJCP.
Manuscript: The neurons in the brain cannot implement symbolic systems.
Reference: ECP 39.97
Referee: Luis Jimenez
General Recommendation: Resubmit/not suitable for EJCP
The central thesis of this paper is that, given the stochastic nature of connections of individual neurons, it is not possible for the brain to implement symbolic systems. The paper begins with a very brief description of the essential features of symbol systems, then includes a description of the main features of neurons in the brain, with special emphasis on their pattern of connectivity, and finally proceed to the discussion of its main thesis, namely that the symbol systems cannot be implemented through such kind of hardware. The rest of the paper is mostly a restatement of this thesis against several counter-arguments.
Although I am sympathetic with some of the arguments defended along the paper, i'M afraid that my general appraisal can not be positive, and that I can not recommend publication of the manuscript as it stands now. Maybe a substantial revision of the paper could be considered as a resubmission, but it would require a change in the focus of the article, and specially more convincing discussion of the its central thesis. This general appraisal is based on three converging reasons that are discussed below.
First, my reading of the paper has been hindered by what I construe as a problem of focus and style. The author adopts a polemicist style that struck me as inappropriate, mainly because he does not contend with other papers addressing the same topic at a similar depth,
I could not find any other discussion of implementation of symbolic systems in a network with stochastic connectivity. The last two paragraphs of section 2 and section 10 of the paper are dedicated to this point.
but only with the content of some textbooks that, of course, can not avoid to be limited in scope and relatively simplified. Although this kind of provocative style can sometimes be considered as healthy, I would have liked to find (more?) references to the current discussion between defenders of the classical framework and proposers of non-symbolic alternatives.
This reviewer missed the whole point of using textbooks knowledge, which is to demonstrate that these are not speculations, but well established knowledge.
The current discussions about symbolic systems ignores completely the question of implementation by neurons, as I discussed in the end of section 2, with some references.
On the other hand, the problem of focus arises at the very description of what a symbolic system is. Here, I simply fail to follow the author in his description of the description of the symbolic framework: I don't know why the two requirements that he points out in p. 5 are the "essential" features of symbolic systems and why the whole discussion of the paper should be based precisely on these two features. The author may be right, but my understanding of his pints would have been largely benefited from a thorough review of the literature from which the author has inferred such conclusions.
Without the The two requirements on p.5 the system cannot manipulate the symbols. This is obvious to everybody with even the slightest aqcuaintance with the question of implementing symbolic systems. That the reviewer does not know that shows that he has never thought about this problem. There is nothing wrong with that, but it is not a reasonable basis for rejecting my paper. He can either consult somebody with more expertise, consult basic books about the subject, or simply accept my word for it. It is not sensible to expect me to argue for what is an obviousity for anybody with basic knowledge in the relevant area.
Second, and besides the problems of style and focus, the main shortcoming that I find in the paper is that it does not present convincing evidence to persuade the reader about its main thesis: that neurons in the brain cannot implement symbolic systems. OK, he convinces me that it is difficult to do so, but I was already persuaded about that before reading the text. In my view, then, what the paper fails to do is to demonstrate that it is not possible, EVEN IN PRINCIPLE, for a brain made up of neurons to implement a symbolic system. The discussion in section 5 is an attempt to do so, but I think that it is not a successful one.
Until now, the reviewer just states that he is not convinced by the argument, but did not point to anything which could be wrong with it.
My final impression is that, although the original pattern of connectivity among single neurons is stochastic, some learned patterns of connectivity over groups of neurons could present enough systematicity to be considered as symbolic. Paragraph #4 in p.12 actually suggests this possibility, but it is discarded too soon, in my opinion, based on the absence of a learning supervisor.
This is a demagogical distortion of my text. I don't mention `supervisor' in the text. Instead, I explicitly states what is the information that is required for such learning. The importance of this point is revealed in the next sentence.
I don't think that such a supervisor would always be necessary for learning,
A supervisor is indeed not necessary, but the knowledge that I wrote about is. Talking about the supervisor instead of the knowledge allow the reviewer to ignore what I wrote.
and I could imagine, at least in principle, a system that develops a symbolic structure out of non-symbolic, stochastic connectivity.
And in the end, all that the reviewer can bring against my argument is his imagination. This is plain nonsense, because being able to imagine something does not make it possible.
Anyway, my purpose is not so much to demonstrate that the thesis of this paper is wrong, as to show that it has not been satisfactory demonstrated.
But the reviewer did not show that. Apart from using his imagination as a counter-argument, and asserting that the argument is not convincing, he did not touch the argument at all.
Finally, there is also a third reason that I envision as a real threat for the conceptual nucleus of the paper. As the author has accurately pointed out by way of counter-argument, the fact that people can handle symbols is enough to demonstrate that the thesis of the paper MUST be wrong.
I have answered this counter-argument in section 7.
So to be entirely consistent, the author should state one of the following three claims: (1) people cannot handle symbols, despite the fact that they appear to do so; (2) the brain does not manage symbols, but there is some other structure, outside the brain, which does that;(3)some part of the brain does manage symbols, but it is not the part that is made up of neurons.
That is extreme nonsense. The obvious alternative claim is that the [brain + senses + mechanical ability] of the person (in other words, the whole person) manages symbols, and obviously this is the alternative I am assuming in section 7.
All these alternatives seems equally unlikely, so that I would prefer to stick at the position that it is the brain made up of neurons which can, somehow, implement a symbolic system, although we currently are unable to know how.
Of course all the alternative are unlikely, because the obvious one has not been considered. The shallowness of this argument is quite amazing, and shows that this reviewer is not intending to seriously evaluated my paper. All he wants is to find arguments to reject it, no matter how stupid these arguments are.
I don't want to be harsh, but I think that the alternatives proposed by the author are simply pointless, and that they do not worth serious consideration.
Apart from being extremely arrogant, that is also far below the standard of scientific writing. If my argument is wrong, the reviewer has to point where it is. If he cannot, he must at least give it the benefit of a doubt.
The paper does not reach the requirements to be considered for publication in EJCP. The issue is relevant enough to deserve discussion in a journal like this, but the paper fails to present a complete description of the relevant framework in which such a discussion should take place, and the arguments presented do not succeed in sustaining the thesis of its author. resubmission could be considered if the author is prepared to face those two challenges.
Signed: Luis Jimenez
Back main page