related texts

Crick's treatment of consciousness

0. Introduction

Francis Crick's book 'The astonishing Hypothesis' (Simon & Schuster(1994), 0-671-71158-X) is an effort to chart the way forward in the investigation of consciousness. Compared to many other monographs in cognitive science, Crick's book is quite reasonable. However, it still contain many errors. In this text I highlight these errors.

Most of the reasoning errors that Crick is doing reflect common misconceptions and errors, and I document them in general documents. When Crick is doing any of these errors, I put here a pointer to the more general discussion of the error.

All the quotes below are exact, including italics, except the addition of bold type to highlight specific parts of the quote. This text is written for a person that has already read Crick's book, and has an immediate access to it.

1. The definition of Consciousness

The lack of definition of the term consciousness is the main problem with a discussion of consciousness (see in Methodological problem in discussions of human cognition, and in Reasoning errors for a discussions). Crick tackles it in several places.

1.1 The opening move

Crick declares his intentions in writing the book in the preface (P. XI). In the second paragraph he says:
Some readers will find this approach disappointing since, as a matter of tactics, it deliberately leaves out many aspects of consciousness they would love to hear discussed - in particular, how one should define it. You do not win battles by debating exactly what it meant by the word battle. You need to have good troops, good weapons, a good strategy, and then hit the enemy hard. The same applies to solving a difficult scientific problem.
This argument is plain nonsense, because the tasks of winning battles and solving scientific problems are different enough that you cannot learn from one on the other. You can find similarities between things that you already know about these tasks, but you cannot deduce new things on one of them by looking at the other.

So why does Crick write plain nonsense in the second paragraph of his book? The obvious answer is that he found that this approach makes it easier for him to convince his readers (or listeners). The interesting question is why does opening the discussion with a piece of nonsense makes it easier for Crick to convince his listeners.

The answer is that this is an extreme example of the Blatant nonsense effect. It works like that: When the reader reads the paragraph critically, shhe sees immediately that it is plain nonsense. However, most of the readers have too much respect for Crick to believe that he will write plain nonsense like that, so they have a dilemma: should they trust their judgement and believe that Crick writes nonsense, or should they trust Crick and discard their own judgement. Most of readers take the latter option, i.e. discard their own judgement, and from that point onwards reads Crick's ideas, particularly about consciousness and its definition, uncritically.

It should be noted that neither the reader nor Crick are necessarily aware of the effect, and I believe that both aren't. As far as Crick is concerned, he simply uses again an argument that proved useful in the past.

1.2 Do we know what consciousness is?

The second chapter of Crick's book is about 'The general nature of consciousness', which sound promising. It starts(P.13):
To come to grips with the problem of consciousness, we first need to know what we have to explain. Of course, in a general way we all know what consciousness is like. Unfortunately that is not enough.
The first sentence is promising, because it suggests that Crick is going to discuss the phenomena that make up 'consciousness'. However, Crick falls on his face in the next sentence (the one in bold). This assertion may look harmless, but it isn't. The usual meaning of the word 'know' (the way it is normally used) implies that the 'known' has an objective validity, i.e. it can verified by other observers than the 'knower'(currently, that means other humans). Thus to accept the bold statement as true, with the normal meaning of the word 'know', we need to assume that the consciousness that each individual knows can be verified by other individuals. However, the consciousness of each individual is not accessible to other individuals, so they cannot verify it. They can do this only if we assume that the consciousness of different individuals is the same, so individuals can use their knowledge of their own consciousness to verify the consciousness of other individuals. Thus this 'innocent' statement requires, and hence implies, the assumption that consciousness is the same across individuals (The 'sameness error', Reasoning Errors ).

Crick discussion in this book is based on this assumption, but he does not make it explicit. There is a good reason for this: to support this assumption, he needs to make explicit what 'consciousness' means, and he does not want to do that.

The last sentence in the last quote suggests that Crick is intending to tell us what needs to be explained, but that is not what he is doing. Instead, he describes suggested solutions to various problems that were suggested by various researchers.

1.3 More bogus arguments

Crick returns to the question of the definition of consciousness on p.20, where he says:
1. Everyone has a rough idea of what is meant by consciousness. It is better to avoid precise definition of consciousness because of the dangers of premature definition. Until the problem is understood much better, any attempt at a formal definition is likely to be either misleading or overly restrictive, or both.
The first sentence repeats the assertion that was discussed in [1.2] above. The rest of the paragraph is another piece of nonsense. A definition of a word can never be restrictive, because you can always use other words to express other ideas (to actually be restrictive, you must limit the pool of words as well). A definition of a word can be misleading, but only if it is different from the way the word is commonly used, and there is no reason why this should be the case. On the other hand, a lack of definition is guaranteed to be confusing.

For the slightly more sophisticated readers, which have not surrendered yet, Crick adds another challenge. In a footnote to the last quote he says:

If this seems like cheating, try defining for me the word Gene. So much is now known about genes that any simple definition is likely to inadequate. How much more difficult, then, to define a biological term when rather little is known about it.
This is less blatant nonsense, because it requires some acquaintance with genetics to see that it is nonsense. gene is simply defined as the 'functional unit of the genome'( where genome is 'the entity(ies) that carry(ies) the design of the organism'). The knowledge that we have about the genome means that the description (which is factual statements about biology) of what is gene is complex, but it does not affect its definition (which is a statement about the way the term gene is used in communication). Similarly, we need to learn more to be able to describe consciousness (i.e. make factual statements about it), but we don't need more information to define consciousness ( i.e. make statements about how the term consciousness is used in communication).

1.4 Speculations about consciousness

Crick strongly opposes the idea of giving definition to 'consciousness', and gives nonsense arguments against it. However, his main speculations in the book, on page 251, are centered around this term. As a result they are actually meaningless.

Interestingly, Crick does define 'conscious' in one place in his book (p.266, first paragraph) as the capability of immediate recall. However, this definition is not used in the rest of the book, and does not fit with it, because recall does not feature at all in Crick's discussion.

2. The basic ideas of the research

Crick introduces his basic ideas of researching consciousness on pp.19-20. The first idea is introduced (after some blurb) on p.20:
It seems probable, however, that at any one moment some active neuronal processes in your head correlate with consciousness, while others do not. What are the differences between them?
The problem with this question is that its meaning is dependent on the meaning of the word 'consciousness', and without a definition, this meaning varies freely between different people. Thus the meaning of the question is variable, and trying to answer it tends to lead to confusion.

Crick continues (p. 20):

Our second assumption was tentative: That all the different aspect of consciousness, for example pain and visual awareness, employ a basic common mechanism or perhaps a few such mechanisms. If we could understand the mechanisms for one aspect, then we hope we will have gone most of the way to understanding them all.
Crick does not give us any idea why this assumption should be true, or any research program to test it. It looks to me that the sole reason for this assumption is that Crick thinks he needs a reason for justifying his interest in vision, and uses this assumption. To my mind, at least, interest in vision does not require any justification, so Crick can do well without this highly optimistic assumption.

3. The homunculus, symbols and neural correlates

Crick discusses the homunculus idea on p.24, and says that with his hypothesis the problem of vision is seen different. He then says (pp.24-25):
In short, there must be structures or operations in the brain that, in some mysterious way, behave as if they correspond somewhat to the mental picture of the homunculus.
That is plain nonsense, because the homunculus is not an observation, so there does not have to be anything that corresponds to it in the real system. That may seem like a negligible point, but its importance comes out when Crick starts talking about symbols.

Crick gives an explicit definition of the word "symbol" in P.32:

A symbol is something that stands for something else, just as word does.
That is a reasonable definition, and in agreement with the way the term 'symbol' is understood, but it means that for something to be a symbol, there must be something that interpret it, i.e. uses the symbol to refer to what the symbol stands for. Since Crick takes for granted that the brain uses symbols, this definition leads him to an interpreter, i.e. the homunculus.

Crick himself seems to have understood this point in some level, because in a footnote on p.33 he says:

The use of the word symbol should not be taken to imply the literal existence of a homunculus. it merely means that the firing of the neuron (or neurons) is highly correlate with some particular aspect of the visual world.
Here Crick tells us the he does not use the term symbol to mean what he himself defined earlier ('stands for', p.32), but something else('highly correlate with'). These two are very different things, e.g. words stand for entities in the world, but the words, in general, do not correlate in any sense with the entities they stand for. On the other hand, for example, foot size highly correlates with height, but does not stand for it. Thus these definitions are completely different.

Why does Crick uses the term 'symbol' when he means something else? I can think of two explanations. The first is that Crick simply tries to conform to the mainstream of cognitive psychology, with its emphasis on symbolic systems.

The other explanation is that Crick wants to introduce the idea of correlation between neural activity and the real world, but cannot find any good argument for it. Crick clearly believes in this correlation, but it is difficult to find any argument for it. It is an interesting question if Crick himself is aware of the way he is avoiding the argument for correlation (my gut-feeling is that he does not).

4. Attention and Memory

In chapter 5 Crick discusses 'Memory and Attention.' In this discussion Crick follows the normal myths and errors in cognitive psychology, which are discussed in Myths[2] and Myths[3] and in Reasoning Errors. Crick here is clearly far away from his main line of interest, but his uncritical adoption of all the common errors is quite disappointing.

5. Theories of Vision

Crick lays down his approach to vision research in the end of chapter 6 (pp.77-78). There he says he mostly agrees with Ramachandran. After rejecting deduction, resonance and elaborate equations as a way to approach perception (on which I agree ) he quotes agreeably Ramachandran as believing that perception (p. 77)
"uses rules of thumb, short-cuts, and clever sleight-of-hand tricks that are acquired by trial and error through millions years of natural selection. This is familiar strategy in biology but for some reason it seems to have escaped the notice of psychologists who seem to forget that the brain is a biological origin..."
The first half of the first sentence is reasonable, but the second half (in bold) isn't. First there is a problem of the tense, which makes sense only if the time of acquisition is short (e.g 'that are acquired through childhood'), but not when the time of acquisition is millions of years. Hence it probably meant to say 'That WERE acquired..'. With this modification, this idea is clearly wrong, because evolutionary 'tricks' must be coded with precise connectivity to actually work, and in the human brain the connectivity is mostly stochastic (see in Stochastic Connectivity section 4 for a discussion of the stochastic connectivity). Thus Crick (following Ramachandran), like most of cognitive scientists, ignores what we already know about the brain.

Crick may have felt that something is wrong here, because on the next page he says in a footnote (p. 78):

There may be only few basic learning mechanisms underlying all this complex activity. The final explanation is likely to be in terms of the basic patterns of connectivity laid down in normal development, plus the key learning algorithms that modify those connections and other neural parameters.
This is much more realistic, but it still assumes some 'basic patterns of connectivity', and it is not clear what these are. In particular, it is not clear if these 'basic patterns' are supposed to code the 'rules of thumb etc.' (which is wrong because the stochastic nature of the connectivity), or the latter are supposed to be acquired by learning (which contradict the quote from Ramachandran). It is only a footnote, and even though Crick puts key learning algorithms in italics, he does not develop this point anymore in the book.

6. Part II

In part II of the book Crick describes the brain and visual system. However, as the discussion get closer to Crick's areas of interest, he starts to put in his ideas, and makes several conceptual mistakes. Some of them are simply a result of taking other people's ideas uncritically. Here I discuss only part of those misconception which Crick does not present as other people's ideas.

6.1 Neurons and layers

In page 132 Crick starts to discuss the cortex, and says:
The most noticeable feature of the cortical sheet is that it is layered. It is important to know something about these layers as the neurons in the different layers do somewhat different.
From this sentence, and from the discussion that follows, the reader may get the impression that each neuron belongs to a specific layer. This is simply false. While we can say that the neuron 'belongs' to the layer that its soma is in, functionally this is nonsense. The interaction of the neuron with other neurons, which is what a neuron in the cortex is doing, is done (mainly) through its dendrites and axon trees, and these traverse the other layers. Thus functionally, a neuron does not belong to a layer.

Paying too much attention to the layers in the cortex is one of the common myths in cognitive neuroscience(see in Myths [8]).

6.2 The 'intelligent neuron' misconception

During all the discussion, Crick repeatably uses terms that imply 'intelligent-neurons', e.g. "the V2 neurons, as a whole, are interested ..." (p. 147) (see the discussion Myths [6]). Crick, however, is at least partially aware of the problem that this language hides, i.e the underlying connectivity that causes the neurons to behave the way they behave. On p.143-144 he discusses the input side of this point in two paragraphs. However, even though he seems to think that this is an important problem, he ignores it completely in the rest of the book.

In fact, we find statements like this (p. 237):

One characteristic of an awareness neuron is that its firing is often likely to be the result of a decision by the neural network involved.
Which gives a very strong impression that Crick believes that the firing of other neurons is not a result of the network involved. Certainly non-experts readers will get this impression.

The output side of this problem, i.e. the connectivity of the axon tree, is mention in a single paragraph on p.146-147,without any real discussion. Here he gives the impression that he does not actually understand the problem, when he says:

The projective field will probably play an important role in any discussion of "meaning." It is unlikely that the activity of a neuron whose axon has been severed can have much meaning for the brain.

This statement gives the impression that Crick thinks that as long as we don't discuss "meaning", we do not need to understand the connectivity of the axon of a neuron to understand its function. It may seems unbelievable that somebody like Crick actually believes this, but the rest of the book confirms this. Through all the discussion of the 'neural correlates of awareness', which Crick takes for granted to be firing of some neurons, he does not consider the question of how these neurons affect the rest of the system. As far as Crick is concerned, their activity is an end to itself.

For example, on p.251, where Crick advances his speculations about consciousness, he says:

Not all cortical neurons in the lower cortical layers can express consciousness. The most likely types are some of the large "bursty" pyramidal cells in layer 5, such as those that project right out of the cortical system.
So what does 'express consciousness' means? The term was not discussed earlier in the book, and consciousness is explicitly undefined, so this is an undefined term as well. However, it does not seem to do anything with the output connectivity of the neurons, because otherwise Crick would have discussed how the projection of these neurons 'express consciousness'. He does not.

6.3 Activity related to seeing

On page 147 Crick says:

It is probably a good general rule that when we see some visual feature explicitly (as opposed to merely inferring it), there will be neurons in some areas of our brains that are firing to it.
I don't really know how to interpret this sentence. Crick seems to think that when we 'merely infer' something, there is no neural activity which corresponds to it (can you think on another interpretation?). This interpretation suggests that Crick himself does not believe his own 'Astonishing Hypothesis'.

Another interpretation is that for explicitly seen features, there are specific (i.e. the same ones each time) neurons which are active, while for inferred features, there will be some neural activity, but not by specific neurons. While this is not obvious nonsense, it is still a very strong assumption, which, if this is what he means, Crick assumes implicitly without any supporting argument.

I couldn't find in the rest of the text anything to make it clear what Crick actually means, and it seems that he is quite confused about how to think of non-explicit mental operations.

6.4 Regions, projections and hierarchy in the cortex

Like many others, Crick believes in the importance of distinguishing regions in the cortex (see the discussion in Myths[9]). He puts a lot of significance to their 'hierarchical' organization (p.155-157). This, however, is based on a very skewed view of the data, with connections that don't fit the theory simply called 'exceptions'. Except from the unique position of the V1 region as the main input of visual information, there are no signs for this 'hierarchy' in the connectivity of the cortex itself.

The other 'evidence' for hierarchy is that in areas that are further from V1 the neurons respond to more complex features and larger receptive fields. This is exactly what we should expect if the information is being spread as it flows (by neurons activating other neurons) inside the cortex. To take this as evidence for hierarchy is a typical 'misanalysis of the null hypothesis error' (Reasoning Errors).

6.5 The myth of dissociation

Crick did not figure the nonsense of the dissociation myth (Reasoning Errors), as evident when he says of P.166, after discussing the various different cases of prosopagnosia:
These results suggest that different aspects of facial recognition are handled in different parts of the brain.

7. The usefulness of artificial neural networks

In chapter 14 Crick discusses neural networks. Towards the end of the chapter he says (P.197):
A possible criticism of these neural networks is that since they use such a grossly unrealistic learning algorithm, they really don't reveal much about the brain.
He then try to answer this criticism. The first counter is :
One is to try algorithm that appear to be biologically more acceptable.
Good idea, but this does not help the current networks to contribute anything. Instead, this argument should lead to rejection of implausible models. That is clearly not what Crick has in mind.

The second counter is not better. Crick says that "The other is a more general and powerful one.", and then, relying on David Zipser, says:

The claim is that if the architecture of the network at least approximate the real thing, and if enough constraints on the system are known, then backprop, being a method that minimizes errors, will usually arrive at a solution that, in its general characteristics, will resemble the real biological one.
This statement is either a tautology or false, depending on what is the meaning of 'enough constraints'. If this is defined as 'enough constraints to reach a solution that resemble the real one', then this is a tautology, and we are left with the question whether we have enough constraints. If 'enough constraints' is defined otherwise, the statement is simply false.

It is fair to say that Crick continues with a cautious approach, but he does not debunk these bogus arguments. Notwithstanding all his caveats, he believes that current neural network, even though they are biologically implausible, already contributes to our understanding of the brain. There doesn't seem to be any base for this belief except these arguments.

8. The binding problem

Like many other, Crick is paying much attention to the 'binding problem' (see the discussion Myths [4]). He introduces the problem on page 208, where he says:

Because any object will have different characteristics (form, color, motion, etc.) that are processed in several different visual areas, it is reasonable to assume that seeing any one object often involves neurons in many different visual areas. The problem of how these neurons temporarily become active as a unit is often described as "the binding problem."
In what sense is 'active as a unit' different from 'active' or 'active at the same time'? Why do the neurons have to 'become active as a unit'? Presumably, Crick has some answers to these questions, but he does not tell us. Maybe the beginning of the next paragraph contains a hint:
Our experience of perceptual unity thus suggests that the brain in some way binds together, in mutually coherent way, all those neurons actively responding to different aspects of a perceived object.
Thus it seems that Crick believes that the neurons have to become active as a unit (whatever that means) because of our 'experience of perceptual unity'. If that is true, This is an example of the 'unity' misconception (Myths [10]). Crick does not states what 'perceptual unity' is, but maybe he means the fact that we see objects as objects, rather then separate attributes (this interpretation is supported by the first paragraph of chapter 17). This would be a typical 'modularity error' (Reasoning Errors), of assuming that the neurons that deal with different attributes are separated into modules.

Crick himself may have felt that he is on shaking ground, because in a footnote on p.208 he says:

It is not completely certain that the binding problem (as I have stated it) is a real one, or whether the brain gets around it by some unknown trick.

Crick continues by giving us his idea of what binding is:

In other words, if you are currently paying attention to a friend discussing some point with you, neurons that respond to the motion of his face, neurons that respond to its hue, neurons in your auditory cortex that respond to the words coming from his face, and possibly the memory traces associated with knowing whose face it is all have to be "bound" together, to carry a common label identifying them as neurons that jointly generate the perception of the specific case.
Here Crick seems to interpret 'active as a unit' as 'carrying a common label'. Why do these neurons need to carry a label? Crick does not explain.

Crick favorite idea about 'binding' is that it is done by correlated firing, which he raises as a possibility on p.211. Chapter 17, which is called 'Oscillation and processing units' is dedicated to it, and significantly, this is the chapter that contains the main speculations of the book (p.251). The nonsenseness of oscillation as a significant part of information processing in the brain is discussed in Myths [5].

9. The main speculations

On page 251, Crick presents his main speculations about consciousness. In the next page he admits the he himself regard these as 'house of cards'. However, these are the main speculations of the book, so I comment on them paragraph by paragraph.

Consciousness is associated with certain neural activity. A Plausible model could start with the idea that this activity is largely in the lower cortical (layers 5 and 6). This activity expresses the local (transient) results of "computations" taking place in other cortical layers.
Because 'Consciousness' is undefined, this is almost meaningless. The only clear thing is that Crick thinks that you can consider the activity of layers 5 and 6 separately from the rest of the layers, which is obviously nonsense (section [6.1] above).

Not all cortical neurons in the lower cortical layers can express consciousness. The most likely types are some of the large "bursty" pyramidal cells in layer 5, such as those that project right out of the cortical system.
This was discussed in [6.3] above.
The special lower-layer activity will not reach consciousness unless sustained by some form of very short term memory. It is plausible that this needs an effective reverberatory circuit from cortical layer 6 to the thalamus and back again to cortical layers 4 and 6. If this is lacking, or if layer 4 is too small, it may not be possible to sustain these reverberations. For this reason only some cortical areas will express consciousness.
That the thalamus is essential for the working of cortex is obvious, and connections in both directions are essential (mainly brain damage studies). Since we don't have a way to distinguish activity from 'reverberations', the talk about reverberations does not add anything. The references to layers are nonsense as above. We still don't know what 'consciousness' or 'express consciousness' means.
A processing unit (only some of which are associated with consciousness) is a set of cortical areas at the same level in the visual hierarchy, each projecting into each other's layer 4. Each set of such cortical areas is strongly connected to just one small region of the thalamus. Such a region coordinates the activities of its associated cortical areas by synchronizing their firing.
The fictitious nature of the hierarchy was discussed in [6.4] above. It is possible that the thalamus has a 'map' of the cortex, i.e. that specific regions in the cortex are mainly in contact with specific regions in the thalamus. How synchronization can help in information processing is a mystery, though maybe Crick is referring to the 'binding problem'.
The Thalamus is intimately involved in attentional mechanism. Special binding, where needed for operations such as object tagging (especially figure/ground discrimination), takes the form of coordinate firing, often with rhythm in the 40-Hertz range.
As above, the thalamus is essential to all mental activity, including attentional mechanisms. What is 'intimately' means, and how it differs from being essential, is not clear. The idea that oscillations are important in information processing is discussed in Myths [5] The idea of 'the binding problem' is discussed in [8] above.

In summary, Crick's idea are either meaningless or nonsense, except the possibility of a 'map' of the cortex in the thalamus.

10. Free Will

In the end of the book, Crick adds 'A Postscript on Free Will.' The first two pages (265-266) are a coherent debunking of the idea that free will is a special feature which requires an explanation. The only mistake is that Crick assumes that there is a specific part of the brain that does this.

In the next two pages, Crick 'finds' the location of free will, based on one brain damage case. The nonsenseness of this is quite amazing, as there is large number of brain damage studies, and they definitely contradict the idea that free will is located in a specific place.


Yehouda Harpaz