Emergence Unscathed:
Kim on Non-Reducible Types

Ron McClamrock
Department of Philosophy
University at Albany, SUNY
[email protected]


In The Electronic Journal of Analytic Philosophy 1993

ABSTRACT:

Jaegwon Kim has recently argued that the widespread assumption of the multiple realizability of higher-level kinds -- and in particular, psychological kinds -- conflicts with some fundamental constraints on both materialistic metaphysics and scientific taxonomy. Kim concludes that the multiple realizability of psychological kinds would leave them "disqualified as proper scientific kinds" (Kim 1992: 18), and that search for a scientific psychology should focus instead on more reductive or type- materialist possibilities. If correct, this would strikingly undermine a widespread assumption in the philosophy of mind and cognitive science. But it's not.


1. The Argument

Let me begin by restating the argument. Kim begins from what he sees as a (perhaps the) fundamental assumption of materialism: the rejection of any kind of "magical" emergence:

(i) "There are no causal powers that magically emerge at a higher-level and of which there is no accounting in terms of lower-level properties and their causal powers and nomic connections...." (Kim 1992: 18)

The next claim, which he calls the "Causal Inheritance" principle, is seen as following from (i): "to deny [causal inheritance] would be to accept emergent causal powers", and any "serious physicalist would find these consequences intolerable" (18).

(ii) "The Causal Inheritance Principle: If mental property M is realized in a system at t in virtue of physical realization base P, the causal powers of this instance of M are identical with the causal powers of P. " (18)

The next claim -- which is both the most pivotal and the most tendentious -- is taken as in turn following from (ii) (the hypothesis referred to here is in fact our thesis (ii)):

(iii) "Instances of M that are realized by the same physical base must be grouped under one kind, since ex hypothesi the physical base is a causal kind; and instances of M with different realization bases must be grouped under distinct kinds, since, again ex hypothesi, these realization bases are distinct as causal kinds." (18)

From (iii), we are to infer (iv): (iv) "Given that mental kinds are realized by diverse physical causal kinds, therefore, it follows that mental kinds are not causal kinds...." (18)

The next claim is another premise, rather than following in the chain allegedly leading from (i) to (iv). Kim calls it the principle of "Causal Individuation of Kinds"; and it is supposed to be a "broadly metaphysical and methodological principle about science" (19) which is seen as both "a plausible principle [and] widely accepted" (17).{1}

(v) "The Principle of Causal Individuation of Kinds: Kinds in science are individuated on the basis of causal powers; that is, objects and events fall under a kind, or share in a property, insofar as they have similar causal powers." (17)

Finally, we get to the striking (if true) overall conclusion, which is taken to be a consequence of (iv) and (v):

(vi) If mental kinds are multiply realizable, then they "are disqualified as proper scientific kinds" (18); multiple realizability "effectively rules out mental kinds as scientific kinds" (18); "psychological kinds are not scientific kinds" (19).

I believe there are many suspect inferences in this argument, and have, in fact, opposed some of them elsewhere.{2} But in the interest of brevity, I'll focus here on what I take to be Kim's most central problem: the role and interpretation of (iii).

2. The Problem

As noted above, (iii) is the locus of the most fundamental problem with Kim's position. To make this clear, let me first paraphrase (iii) for clarity by dropping out the explicit references back to (ii); I trust that in the context of the current discussion, nothing is distorted by this. But keep in mind that (ii) is still the fundamental justification for (iii). The leaner version:

(iii') Any physically type-identical instances of M must be grouped under one kind, and any physically type-distinct instances of M must be grouped under distinct kinds.

Note that (iii') (like the original (iii)) contains an obvious quantifier ambiguity. I take the two natural disambiguations to be these:

(iii-weak) There is some causal kind which groups together physically type-identical instances of M and does not group together any physically type-distinct instances of M.

(iii-strong) Any causal kind will group together physically type-identical instances of M and will not group together any physically type-distinct instances of M.

Now although the weak reading may well follow from (i) and (ii){3}, it is the strong reading of [3] that is required if we are to infer (iv) -- the claim that multiply realizable kinds are not causal kinds.

But (iii-strong) not only fails to follow from (i) and (ii); it's also simply false. A complete specification of the causal powers of an instance of a mental property and at a time will of course be a complete specification of the causal powers of the physical state implementing it, given token identity. (iii-strong) only follows if we assume that the only specification of causal kinds can be in terms of complete characterizations of causal powers of tokens; and we've been given no reason to assume this. Instances of a causal type needn't share every single property in common; types are essentially more general features that various instances might share. The existence of higher-level causal types needn't be predicated on causal powers that emerge at the higher level and for which there's no lower-level account. Higher-level causal powers can be understood as just more general and abstract characterizations of the causal powers implemented in the physical structure of the system.

So even if objects and events do fall under a kind only in virtue of sharing causal powers{4}, we've been shown nothing which prevents us from drawing the boundaries on types with more abstract characterizations of those causal powers. The fact that some taxonomy constitutes kinds on the basis of their complete physical causal powers does not entail that some other taxonomy can't group together kinds via some, say, more abstract and less complete characterization of their causal powers -- thus allowing the grouping together of physically type-distinct instances.

Varieties of causal taxonomies group things in virtue of various kinds of causal powers they might have. The taxonomy of orbiting bodies will specify the causal powers of the object had in virtue of its mass, position, and movement, and will abstract away from other features -- e.g., chemical composition, or the presence or absence of life on the body. Similarly, the taxonomies of information-processing structure will abstract from the details of physical implementation (e.g., the material structure of the semiconductors) and focus on the device's causal structure at a more abstract grain (e.g., registers and operations on them).

Without any further motivations for the question-begging (iii-strong), it's hard to see what force the argument has left. Perhaps there is another way to take (iii), or a plausible suppressed premise that will bridge the gap from (i) and (ii) to (iv). But I don't think so. The considerations offered against the strong reading of (iii) seem to be also reasons to reject (iv). In any case, the burden would seem to be on Kim to fill in this missing link.

3. Real Pluralism

There is no stronger nor more important theme in the philosophy of mind of the last 25 years than that of turning toward placing these issues of reduction and identification in the context of the philosophy of science -- of recasting the mind-body problem as the physics- psychology problem, if you like. But Kim's discussion exemplifies an all-too-widespread tendency to accept this shift in name but not in spirit or detail. To bolster this admonition, let me very briefly review a few points about the real uses of multiple realization (and more generally, implementation-independence) in the higher-level sciences.{5}

The real higher-level structures characterized in many sciences are not only multiply realizable in principle, or even in fact across different overall implementations of the higher-level structure. Multiple realizations of not only higher-level properties, but even of the very same token higher-level entity often occur within the same system. For example: the implementation of a higher-level primitive in computer programs, such as a particular variable, will have significantly different machine-level implementations at different moments in the running of the program. It will not only hold different values, but will also reside at different real memory locations (as garbage collection may re-allocate variable memory), be in RAM cache or on disk, be in compressed or uncompressed format, and so on (see, e.g., Wilson 1991). These different physical structures implement that particular variable in virtue of their global role in the higher-level processes of the system, and will typically lack any common properties at the physical (or even the machine) level that they don't share with implementations of other variables used in the system.

Not only does multiple realizability allow different lower-level structures to implement the same higher-level feature; but in addition, context-dependence allows the very same lower-level structure to implement different higher-level features in different contexts. Such context-dependence permeates scientific taxonomies: the positions of protein sequences in DNA with respect to the rest of the genetic material are critical to their genetic status: type-identical nucleic acid sequences at different loci can play different hereditary roles -- that is, can be different genes. Similarly, type-identical CPU operations (e.g., storing a register to an indexed address) can differ completely in terms of their functional properties -- on one instance implementing "set the carry digit from the last addition", and at another, "add the new letter onto the current line of text".

Further, context-dependence doesn't just allow the same local properties of a device to play different overall functional roles. Facts about context also will influence the higher-level properties of a given local physical structure by making different properties of that local part functionally salient. Not only can a different surrounding context make a transistor switch -- still taken as a digital switch -- represent different functions; but varying context also can make that transistor's switching into a different higher-level event, because the context forces different (lower-level) properties of the transistor to be the ones that are salient to the overall functioning of the system. When a switching transistor goes from `off' to `on'{6}, there's lots of lower-level complexity in the event -- not only the large-scale change in overall DC voltage level, but a fair bit of high-frequency noise known as switching transients. By putting that complex local lower- level electrical event in the surrounding context of a radio signal generator, we will make most salient the high-frequency switching transients of the transistor, and will reduce the difference between 0's and 1's represented by different DC voltages to the status of noise. Here, context determines what part of the mass of lower-level activity in the object is an implementation of some higher-level function at all. The example thus illustrates how understanding the behavior of a complex system requires knowing which aspects of the complex mass of lower-level properties are significant in making a contribution to the overall behavior of the system; and this in turn depends on having some sense of the higher-level functioning of the system.

4. Last Words

It's striking that a paper like Kim's which attempts to make such a far-reaching claim about scientific methodology and inquiry gives almost no consideration to any particular science, and certainly none at all to the moderately successful special sciences -- e.g., biology (from genetic to cell to ecological), information science, or even macro-physics (like the physics of waves, falling/orbiting bodies, etc.). It settles instead for a fairly standard version of a simplistic and philosophically idealized micro-physics. In fact, his only real example of a multiply realizable kind (outside of the case of the mental, which is explicitly up for grabs) is the now well-worn example of jade as having two distinct implementations (jadeite and nephrite).

But from the perspective of contemporary pluralism in the sciences, this is a degenerate case, lacking all of the interesting complexity of the paradigmatic cases of higher-level types in science. I, for one, am happy enough to let the reductionists have this one: there may well be no interesting causal taxonomy which groups together jadeite and nephrite and nothing else. But the existence of some such cases is hardly enough for Kim's point, unless no cases go the other way; and the kinds of the moderately successful special sciences are at least prima facie examples of such types. These other sciences might well provide more appropriate models for psychology than does idealized micro-physics; and surely their relationship to "the physical" is the first place to look for guidance in sorting out the "new" version of the mind-body problem.

The widespread temptation in philosophy to explicitly or implicitly damn a view for not meeting the honorific standards of "science" often levies these criticisms from the perspective of a monolithic view of science: a kind of reductive perspective on the relation of imaginary micro-physics to the other sciences that's largely been superseded by a more pluralistic view of the relationship between investigation in the various sciences. Those of us interested in the philosophy of the less-developed sciences might do well to be less driven by metaphysical preconceptions, and more motivated by something like the following working hypothesis: if it shows up in more established sciences (like genetics, and even computer science), don't be so worried when it shows up in slightly less mature ones -- like psychology.


References:

Fodor, Jerry (1987) Psychosemantics: The Problem of Meaning in the Philosophy of Mind. Cambridge MA: M.I.T. Press.

Kim, Jaegwon (1992) "Multiple Realization and the Metaphysics of Reduction." Philosophy and Phenomenological Research 52: 1-26.

McClamrock, Ron (1991) "Methodological Individualism Considered as a Constitutive Principle of Scientific Inquiry." Philosophical Psychology 3, no. 4.

---- (1993a) "Functional Analysis and Etiology." Erkenntnis 38, no. 3.

---- (Forthcoming) Existential Cognition: Computational Minds in the World. Chicago: The University of Chicago Press.

Wilson, Paul (1991) "Pointer Swizzling at Page Fault Time: Efficiently Supporting Huge Address Spaces on Standard Hardware." Computer Architecture News, June 1991.

Notes:

Thanks to Paul Wilson, Robert Howell, Stuart Glennan, Daniel Gilman, and Irene Appelbaum for their comments on an earlier draft.

{1} Kim notes in passing (Kim 1992: fn. 34) that (ii) and (v) are almost identical to Fodor's (Fodor 1987: chapter 2) metaphysical point ("causal powers supervene on local microstructure", and his methodological point ("categorization in science is characteristically taxonomy by causal powers") (Fodor: 44). Fodor obviously intends these as compatible with multiple realizability, offering them instead as part of defense of "individualism" (see McClamrock 1991).

{2} For example, something very much like Kim's (v) is opposed on the grounds of the level-relativity of context-dependence of types in (McClamrock 1991).

{3} I believe it does, as long as we keep in mind that (a) a complete physical characterization of some class of instances of M will satisfy (iii-weak), and (b) (iii-weak) does not claim that M must contain only one such physically characterizable subset.

{4} See (McClamrock 1993a) for some reasons for thinking that this too is false, particularly for functional properties.

{5} For a fuller account, see Chapters 1-3 of (McClamrock 1993b).

{6} I'm simplifying by constraining our consideration to the transistor's electrical output; when we drop this simplification (and, e.g., allow in the heat produced, or the electrical effects on the input side) the point of the example is only strengthened.