Defining “Emergence”

I’m working on my first draft of Key Terms in Philosophy of Mind, a book under contract with Continuum Books. From time to time I’ll be posting draft entries on Brain Hammer, especially for controversial or especially difficult to arrive at definitions. Here’s “emergence”:

emergence, the arising of a property in a relatively unpredictable way from the interaction of other properties. Alternately, the instantiation of a property by a whole that is due to “more than the sum of its parts”, or, less colloquially, due to properties of the parts in a way more complicated than mere summing. Part of what’s difficult in supplying a viable notion of emergence is the task of characterizing a relevant notion of unpredictability that isn’t due simply to the current ignorance of investigators. Early proponents of the existence of emergent properties claimed that certain chemical properties like the liquidity or solubility of certain chemical samples were emergent on the grounds that they could not be predicted by knowledge of the nature and interaction of their atomic constituents. However, as chemistry and physics progressed, such claims were discovered to be false. Another difficulty in supplying a viable notion of emergence is in giving a precise meaning to the imprecise phrase “more than the sum of its parts”. We can see that there are clear cases in which the property of a whole is more than a sum of properties of its parts but that the properties of the whole are unlikely to be regarded by anyone as having emerged from the properties of the parts. For example, the temperature of a gas is the average kinetic energy of its constituent molecules. As such, it is thus not simply the sum of the molecule’s kinetic energy. It is instead the sum of their kinetic energy divided by the number of molecules. There’s a sense in which being a divided sum of its parts is more than the sum of its parts: since it involves division, it involves a further arithmetical operation than mere summing. However, this seems not to get at the sort of thing that emergentists have had in mind, perhaps since the result of the operation is insufficiently surprising or unpredictable. Emergentism, the proposal that there exist emergent properties, is closely related to non-reductive physicalism (see PHYSICALISM, NON-REDUCTIVE).

19 Responses to “Defining “Emergence””

  1. Two quick questions/comments:

    First, does the emergent property have to be new in *kind*? One way of reading your claim that an emergent property arises from ‘other properties’ suggests that maybe you think that there has to be a difference in kind. I think that Mill’s infamous assertion in A system of logic, that it is impossible to characterize ‘life’ by a mere summing up of the actions of the separate elements of a body, suggests that the answer might be yes. However, Durkheim’s argument for the existence of collective representations, as well as May Brodbeck’s arguments for group psychology and Margaret Gilbert’s arguments for collective intentions, are intended as cases in which the lower-level properties (representations, psychologies, and intentions) are *of the same kind* as the higher level property. I was just wondering how you were thinking about the notion of ‘other properties’ here.

    Second, is it really the case that emergent properties have to be ‘unpredictable’ in some significant sense? Some stock examples of emergent properties (i would think) are arches and columns (in a termite mound), traffic jams, and herding-schooling-flocking behavior. However, each of these are completely predictable from the right sort of theoretical standpoint. I would have thought that the interesting thing about emergent properties was that explaining them requires an appeal to the behavior *of the system as a whole* rather than an appeal to the behavior of the constituent parts. In part, i would guess, someone who buys into emergence might think that this is because appealing to the behavior of the parts fails to capture some theoretically robust generalizations about the behavior of the system; or, perhaps some counterfactual stabilities that are not captured by explanation in terms of the parts; or, something or the sort.

  2. Anibal says:

    What´s the difference among epiphenomenalism and emergentism if both describe in rough manner, according to my view, the arising of a new property with no direct connection to low level parts?

  3. Eric Thomson says:

    I would second Bruce’s second point that you might give a shout out to ‘weak’ emergence in which properties not inherent in the components are displayed at a higher level, and exist because of the organization of those components.

  4. Bryce said: “However, each of these are completely predictable from the right sort of theoretical standpoint.”

    But isn’t that already included in Pete’s definition, under the rubric “relatively unpredictable way from the interaction of other properties”, and his discussion of the problems of when can we say about something that is “relatively unpredictable”? Because it seems to me “predictable from the right sort of theoretical standpoint” could be “relatively unpredictable” given the complexity of the theory.

    I wonder if chaotic systems would be some interesting case, as their behavior seems to be in big part unpredictable, but still we get different kind of properties in the case of chaotic systems. (There was some theories that functioning of different organs in the body is such emergent chaotic phenomenon, what happened with those theories?).

    As for the example with the temperature, and why it isn’t emergent property, maybe we can say it is because the information about temperature can be ‘extracted’ from the information about each of the parts of the system. So, what is enough ,is for us is to have information about each of the molecules, and no information is needed about e.g. their configuration. Which is not the case with e.g. liquidity of the water.

  5. Pete Mandik says:

    Hi Guys,

    Bryce: I intended “other” to be neutral wrt differences in kind vs degree. The main thing I had in mind was the minimal entailment from emergence to non-identity. If e.g. temperature is identical to mean molecular kinetic energy, then it doesn’t emerge from mean molecular kinetic energy.

    Tanasije: I think you do a nice job of saying what I would have said about predictability. So, ’nuff said.

  6. Eric Thomson says:

    I would mention the common distinction by name (i.e., ’strong’ versus ‘weak’ emergence) since it comes up a lot.

  7. Pete Mandik says:

    Thanks, Eric. I struggle, though, to understand what that distinction would amount to such that (1) there’s such a thing as weak emergence and (2) it’s not vacuous (e.g. all properties of wholes, like their masses and velocities are emergent).

  8. Pete Mandik says:


    I think that while emergents may arguably be epiphenomenal, it shouldn’t be part of the definition of emergence that emergents are epiphenomenal. One may try to argue for epiphenomenalism about emergents via an explanatory exclusion argument. Maybe sometime soon I’ll post my entries for “epiphenomenalism” and “explanatory exclusion”.

  9. Eric Thomson says:

    Pete: I think there are some ways to make it less vacuous. For instance, Bedau seems to have his finger on an important point when he says that X can only be predicted via simulations of the system.

    Also, neuroscientists sometimes talk about the ‘emergence’ of larger-scale neuronal patterns via the interaction of individual neurons without mysterious connotations. (E.g., the paper I reviewed here says ‘The model exhibits behavioral regimes of normal brain activity that were not explicitly built-in but emerged spontaneously as the result of interactions among anatomical and dynamic processes.’)

    Or, perhaps you could argue that scientists (in A-life, neural nets, complex systems, cellular automata, whatever) should stop talking about emergence and use some less loaded term. That would seem strange to me. Your definition does capture the initial use of the term by the emergentists, but why let them have the last word on the meaning of the term? A mention of the use of the term in the above disciplines would be more thorough (though perhaps take too much space…I realize these entries must be very hard to keep short).

    In general, it a useful enough concept that I am not too worried about it being vacuous. Even if we get rid of the word ‘emergence’ we’ll still need to have some word for features of a system that are produced via the interaction and organization of its parts.

  10. Eric Thomson says:

    On my first paragraph in previous post. To be fair to you, you did highlite unpredictability as the key claim by the emergentists. I think Bedau has something right, as there are some systems for which we know the nonlinear differential equation literally has no analytical solution. It isn’t due to our ignorance, but can actually be proven. In such cases we must simply simulate the system.

    But there is a second more innocuous sense used by many that I also mentioned: even if we could predict the higher-level feature or whatever, we’d call it emergent because it comes about due to the interaction of a bunch of parts organized a certain way.

  11. Eric Thomson says:

    For some reason the post I referred to previously is not there, so perhaps I screwed up comment submission. Basically the point I made there (aside from mentioning Bedau) is that the notion of ‘emergence’ is used in so many disciplins (neuro, A-life, complexity science, neural network modelling) that it might be a mistake to leave their weaker notion out, even though it doesn’t conform to the early emergentist’s definition in terms of predictability (and why let them have the last word anyway?).

  12. I like Andy Clark’s definition:
    “a phenomenon is emergent if it is best understood by attention to the changing values of a collective variable.”

    A collective variable is a “variable that tracks a pattern resulting from the interactions among multiple elements in a system.”

    This explanatory definition thus cuts across all the arguments what whether something is *really* emergent or not. Emergence just turns out to be a theoretical stance that happens to be useful for explaining some phenomena and not others. It is really just a special case of reductive explanation, except it different in the *ways* that low-level properties interact to bring about higher-level ones.

  13. As far as I can tell from this discussion, every property is emergent. Can somebody give an example of a non-emergent property?

  14. Eric Thomson says:

    The mass of a system isn’t emergent, as it doesn’t depend on an interaction among the parts. Just add up the masses of the parts to get the mass of the whole. And you can solve for the mass given the masses of the individuals (no insoluble nonlinear differential equations involved there). So, using the usual criteria for ‘weak’ emergence, mass works. As would volume. And a zillion other things.

    Where you might get a contradiction is with a linear system whose equations have an analytical solution, but in which “interesting” behavior emerges due to the interaction of the parts. E.g., this can happen in electrical circuits. The people who like the ‘property that emerges due to the interaction of parts’ people would call that emergence, the ‘we must simulate the system to see how it works’ people would not. I’d personally be happy to call it emergent, as the soluble equations formulation seems sort of arbitrary, but at least it does connect with the old-school notion of predictability used by the emergentists. (For me it seems arbitrary as it seems to confuse the formal and material modes of speaking, and many of the neo-emergentists use the material mode, so the types of equations involved don’t matter, but the properties and how they are generated).

    I’m not sure if temperature would be considered emergent by the pro-emergentists, but unclear cases don’t make the distinction vacuous.

  15. I stand by my remarks that these definitions are all pretty much empty.

    I’m seeing things like an emergent feature is one of the “features of a system that are produced via the interaction and organization of its parts” (quoted above by Eric Thomson). Or from Andy Clark, “a phenomenon is emergent if it is best understood by attention to the changing values of a collective variable” where a collective variable is a “variable that tracks a pattern resulting from the interactions among multiple elements in a system.” (quoted above by Gary Williams). Or “the arising of a property in a relatively unpredictable way from the interaction of other properties” (quoted by Mandik at the start). Of course, Mandik properly criticizes the vacuity of “relatively”.

    It’s alleged that mass isn’t emergent. How’s that? Nobody understands the origins of mass (and if it comes from some sort of Higgs mechanism, then it certainly seems to be emergent on all these definitions). And consider issues involving inertia. It would seem that the mass of anything is utterly dependent on the masses of all other things in the entire universe, all interacting via gravity.

    It’s alleged that volume isn’t emergent. Volume of a gas? Well that depends on temperature and pressure, both of which involve interactions. pv = nrt, as they say. The volume of any material thing depends on all sorts of forces. Forces involve interactions.

    On these definitions, the only properties that arent’ emergent are the intrinsic (non-relational) properties of pure ontological simples. I have no idea what the simples would be or what there basic intrinsic properties would be. Are the simples the basic particles in the Standard Model, that is, quarks and leptons? Their charges seem to be mediated by bosons - thus charge is emergent. I have no idea which properties would be non-emergent. Spin maybe?

    And the possibility remains that all physical properties are relational. If that’s true, then indeed they all do emerge from interactions with other things.

    I’d happily accept a definition that says P is emergent iff P is defined by a non-linear equation, or by an equation of this or that form. That’s at least clear. But it’s hardly what people seem to have in mind.

    So I stand by my claim that emergence as defined here is utterly vacuous.

  16. Eric Thomson says:

    I think you are pushing for more precision than the term is mean to have.

    Well, we have one clear use: insoluble equations. That is mathematically nonarbitrary, though ontologically it is arbitrary (that is, the actual systems described by such equations are not any more inherently interesting or complex than those described by solvable linear equations).

    Mass and volume: red herrings.The mass of X (composed of a bunch of Ys) is a sum of the masses of the Ys. It is to change the subject to start to talk about the ‘origins of mass.’

    And with volume similar just tendentious and overconfident. Sure, PV=nRT in gases. Fine. have a block of glass with a certain volume. I add a bunch of such blocks together.

    Emergent isn’t the same as relational. Being 20 feet from is a relational property, but not emergent.

    I have the feeling that this is a good case where scientists and others are using the term in a perfectly vanilla, relatively uninteresting, but not vacuous way, but the philosophers are jumping all around demanding more because of the history of the term within philosophy. It has a new meaning, but it is like we are saying we think gods exist or something when we use this silly little term. Properties emerge when things interact with one another. We need a word for this, and many use the word ‘emergent’ for that higher-level property.

    There is gray area, but so what. This is getting silly.

  17. Why is it getting silly? If our goal is not to be precise, then what’s the point? Why bother to give any definitions at all? Why do philosophy?

    I’m just following the definitions that were given, which I quoted. On those definitions, mass, volume, etc. are all emergent. The definitions don’t say that emergent properties are more specific than non-relational properties.

    There probably are good, mathematically rigorous and scientifically responsible analyses of emergence. I suspect notions of complexity (also troublesome) will play some crucial role. But none of the discussion so far gets even close.

    If aiming for precision is silly, then philosophy is silly.

    All of us should either aim to get it right or just not bother to do it at all.

    And that goes for me, too.

  18. Eric Thomson says:

    The mass and volume discussion was silly. They aren’t emergent on all of the definitions given. Relational properties aren’t ipso facto emergent under the definitions.

    Also, if you look at what I wrote I said you may be trying to add too much precision, by which I meant more precision than is necessary for the term to be meaningful and useful. And more precision than is even appropriate. If the goal is to give a dictionary definition, which in this case it is, it should reflect the precision (or lack thereof) with which the term is used.

    That said, sure we should be precise, as precise as we can presently be with a term. But just because we can’t be as precise in our definitions as you’d like that doesn’t mean the term is vacuous, has no extension, etc.. I have given what I take to be a reasonable few descriptions which are precise enough to give a sense for how the term is actually used in practice.

    I think you may be right that ultimately the extension of the term as presently used may end up with a rough set of necessary and sufficient conditions, perhaps involving complexity of interactions, also likely strength of interactions among parts, complexity of the emerged phenomenon, and that this polydimensional creature will create a space in which certain regions are clearly instances of emergence, others not, and some just unclear.

    I’m really saying two things. As used, it is a perfectly reasonable term even if we can’t give a set of necessary and sufficient conditions. Secondly, a dictionary definition is not the place to work that stuff out, but to give a definition that is as precise as is responsible given present usage.

    Note Pete’s ‘key terms’ book is not the same as a dictionary, so I may be wrong in this sentiment. But to dismiss it as vacuous just because we can’t make the old-school analytic philosophers happy with a set of necessary and sufficient conditions for the use of a term…that’s just silly.

  19. Eric Thomson says:

    This reminds me a lot of the discussion of that awful Bennett book which we had here.