Multiple Realization Poll

Philosophers, both at Brains
and elsewhere in the philosophical blogosphere, seem to like
participating in and reading polls, so how about another one here?

Is the property of having dichromatic color vision multiply realized in humans? If so, why? If not, why

I say “yes”. There are three
familiar forms of dichromatic vision in humans, each corresponding to
the loss of one of the three normal cone pigments.


  1. gualtiero

    I’m not sure that having dichromatic color vision is the kind of property to which the notion of multile realization is designed to apply. I thought putatively multiply realized properties are properties all of whose instances are “functionally equivalent”. Are people with different variants of dichromatic color vision “functionally equivalent”?

  2. Anna-Mari

    Hi Ken,

    I´d follow the argumentation of the Master (of the Brains): Are these “some color discriminations” you mentioned functionally equivalent? Or are there functional differences?


  3. kenneth aizawa

    I’m assuming that the property of being a dichromat is the property of making color discriminations based on two distinct photopigments.   There may be more implicitly to it than this, but this is what the vision scientists make explicit. 

    The three most familiar forms of dichromacy are protanopia, deuteranopia, and tritanopia.  There is a simulation of what it is like to have these forms of color vision at wikipedia:

    Each form of dichromacy is equally dichromacy, but there are different types of dichromacy.

    Note that both you and Big G assume that there must be some sort of conceptual connection between a property’s being MR and some notion of functional equvalence.  But, it could be that all a property’s being MR requires is that the property be the same realized property in the distinct realizations.  In other words, sameness of property involved in MR could be functional equivalence, or it could just be sameness of property.

  4. Carl Gillett

    This will be long-winded, but will get around to what I hope is a further clarification of Ken’s kind of point in response to earlier comments about ‘functional equivalence’

    I say “yes” to Ken’s original question, but qualify that by saying I take my “yes” to refer to what we might term causal-mechanist realization: the kind of ‘realization’ one finds posited in mechanistic explanations, ie a compositional relation between properties. So, I say, with regard to causal-mechanist realization, the properties appear to be multiply realized.

    Now, of course, there are OTHER notions of ‘realization’ out there, for example the further notions associated with computation, or models, or Turing machines, or Ramsey sentences, and so on. One better be careful to distinguish which of these VERY DIFFERENT notions of realization one is talking about or trouble results.

    To take the case at hand, I am not sure it makes much sense to talk of ‘functional equivalence’ with causal-mechanist realization — though one could have a go (for example, instances under a condition that contribute the same powers?). However, I am damn sure a notion of ‘functional equivalence’ of the kind associated with abstract mathematical functions make no sense.

    But perhaps this is what Ken is up to, trying to prompt us to be clearer about WHICH KIND of realization is at issue in this kind of case?

    Just my two cents worth, looking forward to seeing folks at the SSPP, best, Carl

  5. anna-mari

    Hi Carl G,

    Yes, I agree with you that it is important to stress the notion of realization at hand.

    However, there is something I´d like to say about this:
    “To take the case at hand, I am not sure it makes much sense to talk of ‘functional equivalence’ with causal-mechanist realization — though one could have a go (for example, instances under a condition that contribute the same powers?).However, I am damn sure a notion of ‘functional equivalence’ of the kind associated with abstract mathematical functions make no sense”.

    Well, yes. But isn´t it right that this depends on the notion of functional equivalence?
    If one, for instance, defines the functional equivalence as the similarity of the information processing problem-solving tasks and gives them a proper formal, “abstract”, say mathematical, description, it is then possible to say that these tasks can be multiply realized at the “algorithmic” level. And one can, I guess, then well adopt the position advocated by Master (of the Universe)in which the individuation of those “algorithms” (G does not like that expression, he prefers “computations”) is done (causal)- mechanistically. But one just have to remember that the abstract task-level description (of functional equivalence) cannot be done causally or causal-mechanistically.

    This is the standard marrian/chomskian maneuver, or at least we will argue so in EuroCogSci07 in Delphi with Otto.
    We will also talk about these “abstract mechanisms” at ISH in July, and I guess Otto will give that presentation, since I´ve promised to deal the EuroCogSci.


    P.S What does this mean “instances under a condition that contribute the same powers?”?

  6. anna-mari

    I´d like to add that the other speakers in the ISH-special spession titled “Marr`s Vision 25 years” in Exeter will be Oron Shagrir, Amir Horowitz and of course, Matti Sintonen.

    So, we all will be mainly talking about computational explanation in neurocognitive sciences.

  7. The answer to the question is yes, using the traditional notion of MR.

    Originally (in Putnam etc) multiply realizable properties were taken to be those properties that can be realized in objects composed of different types of stuff (e.g., silicon versus carbon based information processors). Under this more traditional definition, mass, energy, temperature, brittleness, being a liquid, etc are all multiply realizable physical properties (hence the leap from MR to the falsity of type-identity theory is illicit).

    It is strange that someone would say that by definition, a property is multiply realizable only if it is a functional property. That would take all the bite out of the original arguments against identity theory using MR, as they would become clearly question-begging.

    Why do intro philmind books (and presumably classes) still use multiple realizability in their exposition of “problems” with identity theory?

  8. kenneth aizawa

    Your first paragraph captures where I think MR applies.  (Maybe Carl would disagree.)  Your second paragraph captures my sense that a definition of MR should not require appeal to functional properties.  That’s what I was trying to say in reply to both Anna-Mari and Gualtiero.  I agree with what Carl was driving at, but that’s just not what I was driving at.

  9. Carl Gillett

    Hi all, one quick question for Eric: Why is it strange to use multiple realization to sink the identity claims generally in the philosophy of science?

    I ask ‘cos I am clearly not tracking something, since it seems to provide a reason. Multiple realization of a property H implies failure of coextensitivity of H with lower level properties — hence Leibniz Law fails between H and lower level properties.

    Why isn’t that a good argument against identity theories — at least to start the circus show? But maybe you meant MR under some other conception? Best Carl

  10. Anna-Mari


    I cannot say, what G is thinking. But I thought that the question G posed was relevant, because Ken´s example was about visual system.

    At least I took it, and I still take it, for granted that in that example the properties “to be multiply realized” must be defined sort-of-“functionally”, since the examples were about visual neurocognitive abilities. I really did not mean to say that all properties (that can be, may be or) are multiple realized _must be_ functionally defined.

    But in the light of Ken´s original example I still think the relevant individuation criteria for the properties is the functional, not some other, criteria. Is this really in contradiction with Putnam`s original account of MR, as Eric seems to suggest…? Well, I do not know. Eric?

  11. Carl: I think the reason MR was able to have such an effect on identity theory is because the identity theory back then was usually the ‘mind-brain identity theory’, and the identity theorists didn’t appreciate that the right-hand side of the identity doesn’t need to refer to the molecular identity of the stuff on the left hand side. This is also the weakness of your formulation.

    Consider the temperature-mean kinetic energy identity. The right hand side refers to the statistical properties of physical stuff, but not to the identity of the particular molecules involved.

    MR seems to show that a molecular-constituent identity theory is wrong. But there are other types of identity theories. E.g., the mind could be identical to some other natural property analagous to brittleness, temperature, or mass, that doesn’t depend on molecular constituents but higher-level statistical (but not functional) properties of a physical system. So the mind could be to the brain as brittleness is to inanimate matter.

    Note I am a functionalist, not because of MR arguments, but because I am an externalist about content. I talked about this here“>”>here.

  12. kenneth aizawa

    One of the things that Carl and I are working on is the idea that MR of a property is indexed to levels.  So, we say that, e.g., memory consolidation is MR at the biochemical level.  We will also argue that it is MR at the neuronal level.  This is consistent with the idea that there can remain other levels at which a given property is not MR. 

    So, regarding the temperature-mean kinetic energy example, I would say that temperature (of a gas) is univocally realized as the mean kinetic energy, but multiply realized in different sets of masses and velocities of different collections of particles.

    We think that something like this relativization to level is implicit in much of the familiar discussion of MR.

  13. At least I took it, and I still take it, for granted that in that example the properties “to be multiply realized” must be defined sort-of-“functionally”, since the examples were about visual neurocognitive abilities.

    It isn’t clear to me that having dichromatic vision is a functional property. I’m not sure.

    The way I understand it is that being a functional kind is sufficient for being MR, but not necessary.

    I don’t have Putnam’s original article handy, but my memory is that he doesn’t give much in the way of an analysis of multiple realizability.

    It sounds like others here are more up to date with the literature, so I’d defer to the experts.

  14. kenneth aizawa

    Maybe “having dichromatic vision” is a functional property, something like, thing that processes color features of light by using exactly two distinct color photopigments.  But, this specification would seem to me to be in the nature of that particular property and not required in order to explicate multiple realization.

    Speaking more directly to Eric, it does seem to be in principle possible for there to exist a functional property that can only be realized in one way, that is only uniquely realizable.  You can get the idea how this might be from the property of being a “diamond scratcher,” where this is a thing that scratches a diamond.  Maybe there will turn out to be only one way or thing that does this.  So, it seems to me to be a largely empirical matter if being a functional property is sufficient for being multiply realizable.  Of course, there is also some room to worry about the modality involved here.

  15. I would resist saying that the temperature of a gas (or rather, gasses) is univocally realizable (the same temperature property can be realized in a silicon or carbon gas).

    But also, temperature in general is multiply realizable (temperature of the air, of a glass of water, of a block of iron).

    Your ‘diamond scratcher’ example is interesting. Would diamond scratchers made differently be multiply realized diamond scratchers? E.g., a little hand-held one versus a big machine with hyrdaulics that lowers the material to the diamond?

  16. kenneth aizawa

    Regarding temperature, I think it was Berent Enc, 1983,  who proposed roughly that temperature is identical to mean kinetic energy, but then that mean kinetic energy is multiply realized in different ensembles.  So, switching gas would simply be another way to get the same mean kinetic energy.  One high level property, temperature, is univocally realized by one middle level property, mean kinetic energy, which is in turn multiply realized by lots of lower level properties (masses and velocities).

    I didn’t flesh out the ‘diamond scratcher’ example, but you would probably have to do something like “standard hardness test diamond scratcher bit’.  So, that only a particular part of the contraption, the bit, has to be real hard and the rest is allowed to be multiply realized, e.g. in the different constitutions of the parts the hold the bit and align it with the test surface, etc.

Comments are closed.