Yehouda Harpaz
27 Feb 96 
 related texts 

Methodological problems in discussions of human cognition

1. Introduction

Various responses to the model of 'The Mechanisms of Human Cognition' raise some methodological points. By 'methodological points' I mean points which do not have anything to do with the correctness/incorrectness of the model, but with the way people evaluate it. In the this text I will try to deal with some of them.

2. Proof And Evidence

Some commentators said they want to see a proof for the model. First thing to note is that, except in logic and mathematics, we don't have proofs in the absolute sense. We have supportive evidence, which can be very strong, and then we call it a proof. Thus the question is really what is the evidence for (or against) the model.

Evidence for a model can be of two main types:

  • Negative evidence - This is made of observations(*) which are incompatible with the model. I claim there is no such evidence, and nobody have claimed otherwise yet.
  • Positive evidence - This is made of observations which are Note that the second condition is critical: if the observation is compatible with the model and also with alternative models, it cannot help us decide which model is correct, so it is not an evidence (or maybe 'neutral evidence'). The problem , therefore, is to find evidence which is compatible with the model which I present, and is incompatible with alternative models (the null hypothesis in this case, i.e. that the brain is a random heap, is obviously incompatible with the facts). This raises the question: what are the alternative models?

    In general, the alternative model to any new model is the currently accepted model. In the case of human cognition, there is no generally accepted model. Rather, there are many models, of all kinds. The evidence that distinguishes between my model and each of these models is different between models. Thus, what is considered to be the evidence for my model is dependent on which model is regarded as alternative. Hence, when a proof of the model is requested, the alternative models must be specified.

    This requirement may seem counterintuitive, but this is because our intuition is based on experience in other domains. In these domains the alternative model is implicitly given (the accepted model), so intuition tells us it needs not be specified. When it comes to a domain which does not have an accepted model, like human cognition, intuition fails to take this fact into account, and therefore is wrong.

    Note that to show that the model is wrong, there is no need to specify alternative models. All that is required is to find observations that are not compatible with it. If the model is really wrong that should be reasonably easy. Thus, it is much more useful to discuss potential evidence against the model, as opposed to look for supporting evidence.

    Though it is not 'the accepted model', the symbolic system approach is the most common. Here is the evidence against it.

    3. What a model of cognition should be able to do

    Some people object to the model because it cannot be used to predict human behavior in details. This objection is valid only if human behavior is actually predictable in details. There is no evidence that this is true, and in fact the most consistent finding in psychological experiments is the large variability in human performance. This is true even for very simple tasks, and in general there are no models that can predict human behavior in details. The only way of 'achieving' this is to model the average of human performance (effectively doing the 'sameness assumption' error, see in Reasoning errors ), on a small number of experiments. None of these models can actually predict anything more than the data they are was based on.

    My model actually predicts that human behavior in details is not predictable in principle, as it is learned on top of stochastic connectivity. This is compatible with the data, but it is at variance with the hopes of CogPsys, who spend their time trying to achieve detailed predictions. Thus CogPsys tend to reject this model immediately, without any consideration of the possibility that it is actually correct.

    To some extent, this is a result of the history of the field. In the early days of the Information Processing Paradigm, there were high hopes that explanations of human thinking are just around the corner. Many models that 'explain' human thinking in details were advanced. With time, most of people realized that these models do not actually explain anything, but they failed (and continue to fail) to realize that even the belief in detailed explanations may be wrong.

    4. 'It is Too complex'

    Some people objected to the model because it is too complex. This can only be based on the implicit assumption that the human cognition can be explained by some simple model, which is obviously nonsense. That does not seem to be enough to prevent people from actually using this argument. Even for critics that did not express this objection explicitly, I suspect that some of them object to the model mainly because of this. It is important to realize that there cannot be a simple explanation of human cognition, and understanding it will require a large mental effort.

    5. 'I don't like it' : The Problem Of Biased Intuition

    In principle, nobody would reject a model because shhe does not like it. In practice, however, when a person doesn't like a model, it affects his/her intuition, such that his/her judgment is biased against the model. This means, that when a person does not like a model, his/her intuition is unreliable when it comes to judge the model or related points.

    When the model is simple, that may be only a small problem. However, when considering a complex model (or any other complex entity), there are many points which require judgment, and significant number of them would have to be decided intuitively. This means that a person that doesn't like the model is almost guaranteed to misjudge in part of the points. Obviously, that is true for a person that does like the model, but the misjudgments would be in the other way.

    The proper solution to this problem is, when intuitive judgments of some point differ, to drop intuition as a means for generating judgments. Instead, these points have to be decided by a deeper analysis, or, if deeper analysis is not successful, regarded as 'undecided'.

    Most of people seem to be unaware of the problem of biased intuition, and continue to rely on their intuition for generating judgments even when it differs from other people's intuition. This means that if somebody does not like a model, it is extremely difficult to convince his/her to accept it, no matter how good the arguments for the model are.

    As explained above, CogPsys don't like my model, mostly because it predicts that it is impossible to predict human behavior in details, and because it is complex. Their intuition is therefore biased, but they still rely on it , many times explicitly.

    6. It does not explain consciousness

    The problem of explaining consciousness is that the term 'consciousness' does not have a generally accepted meaning, i.e. there is no well defined set of real phenomena associated with it. Since realistic models can only explain real phenomena, it can explain consciousness only after this set is defined. At the moment, each person interprets the term 'consciousness' (and related terms) in his/her own way, so what will count for explanation for some people will not count as explanation for most of the rest.

    [9 Mar 2003] Here is somebody trying to find what people think 'consciousness' means, and here the results when he tries an explicit questionaire to get more results. Notwithstanding he fact that my name appear on the message, my role was just to encourage him to publish the results of the first effort and to proceed to the second stage.

    That people tend to accept using an undefined term is best explained by the 'sameness assumption' error (Reasoning Errors). Based on the fact that they have some subjective concept of consciousness, they believe that other people would have the same concept of consciousness. The fact that we cannot find a good definition for consciousness shows that this belief is wrong.

    That people tend to use the term "consciousness" (and related terms) is because it is useful to create confusion. Confusion is useful when the speaker/writer needs to defend nonsensical ideas that they hold. Typically these are the idea of totally free will (sense 2 in section 7 below), or some form of dualism (section 8 below), but there are others. In general, whenever somebody uses "consciousness" as central concept in their discussion, you can be sure that they are trying to hide some non-sensical ideas, because otherwise they would have used clearer language.

    When the meaning consciousness is specified, the most common ingredients are qualia and the ability to reflect. These are explained by the model. However, many people do not find the explanations convincing, because they (the people) ascribe unrealistic attributes to the ingredients of consciousness. These are typically:

    Many people implicitly include additional ingredients in their concept of 'consciousness', and reject the model because it seems not to deal with these additional ingredients. These are probably always either easily explainable by the model, or are fictitious, like the belief in the accuracy of reflection. However, since they are not stated explicitly, it is not possible to argue against them. Only when all the ingredients of consciousness are explicitly stated, it is possible to check if the model can explain it or not.

    In some cases, a more explicit definition of consciousness is given. These come in two varieties:

    Both varieties are useless in serious discussion, and their main effect is to create confusion.

    7. It does not explain Free will

    The term 'Free Will' also suffers from blurred definition, because it may mean either of

    1. Will that is free from external rules.
    2. Will that is free from external rules and also free from the physics of the brain.
    It is quite clear that humans have free will the first sense, but this is not a great mystery. I am not aware of any suggestion for a criterion to distinguish between an intelligent system which takes care of itself and a system with free will in the first sense. Without this criterion, all that requires explanation is intelligence and selfishness.

    Many people believe (mostly implicitly) in the second sense. There is no evidence to support the second part of this, and it is in contardiction with our understanding of the world. At least some of these people get confused between the senses, believing that we must have a free will in the second sense, because when they think whether it exists, they use the first sense. The confusion about the term "consciousness" and related terms is useful to hide this confusion.

    I have seen some places where 'Intentionality' was regarded as an important ingredient for free will, but 'Intentionality' itself doesn't have any real meaning.

    8. It does not explain the vivid/rich/life-like experience/images/feelings we have

    This objection may be implicitly based on the assumption that vividness or richness of images and feelings is an objective feature of these images or feelings. This is obviously wrong, i.e. the vividness is subjective, and simply means that those neuronal patterns that are associated with the concpet of 'vividness' are being activated by these images and feelings. This explanation is compatible with all the data that we have, but for anybody that implicitly assumes an objective feature it seems inadaquate. For other people, this explanation seems to be too simple, most likely because they feel it diminishes the value of the human mind.

    I think a more plausible explanation for the underlying assumption of this objection is the refusal to accept that all our mental and emotional life is made of neural activity. This is shown by the more general statements of the kind:

    "It is difficult/impossible to see how subjective experience (vivid images, rich internal life etc.) can be explained by neural activity."
    This kind of statements are used for arguing that consciousness is a "hard problem", which cannot be explained by plain neurbiology, or maybe not by science at all.

    The statement is clearly nonsense if we assume that all our mental life is neural activity, as seems to be the case from the way we understand the world. The statement makes sense only to people that believe (a-priory) that at least some of the mental life is not neural activity.

    See also here.

    9. The alternative to innateness

    A common confusion in the psychology is about what is the alternative to innateness. In most of the cases, people would tend to take it for granted that it is the environment or "nurture", or "culture". This is simply stupid. The alternative to innateness is learning (in its functional definition, i.e. a robust improvement in performance). The environment is the source of stimuli that are part of the "input" to the learning mechanism (i.e. it affects the activity inside the brain).

    The main problem with talking about the environment as the entity that produces acquisition of mental skills is that it is clearly false, because the environment does not have any knowledge about the brain and how to form changes in it that produce skills. Thus when a person that regards innateness and environment as the only possible entities that produce skill-acquisition really considers acquistion of a mental skill, innateness will look as the only reasonable option.

    "Nurture" and "Culture" just affect some of the environment (not all, for example neither can make objects fall upwards), so it is even less sensical to regard them as producing skills.

    Large part of the appeal of Evolutuonary Psychology is based on this confusion.


    (*) In this text, I completely ignore the philosophical question of whether we can really observe the world etc. I take it for granted that we can.

    (**) An interesting point to note that that when discussing qualia, it is very common to use qualia of colours as examples. This ignores the fact that colour is a unique property (with other few properties like sound frequency) as much as the cognition is concerned, because it is discriminated already at the level of the receptors. The bulk of the qualia (e.g. dog, table, shoe, friend, democracy) are not discriminated at this level. Thus colour is special, and even if there are shared responses to colour that are not learned, it does not tell us that this is true for other properties. Since currently there is no known such unlearned shared response, this is currently a moot point.

    The other important point about colour is that human perception of the world is strongly based on colour discrimination, and colour discrimination is built-in by the receptors. Therefore, colour discrimination is used heavily by humans almost from birth, and hence there is a lot of learned associations with different colours. This confuses the discussion, because some people may believe that some of these associations are not learned.