Go:

Visual Taxonomy

Links

Hide/Show:

Taxonomy Path

http://www.lunduniversity.lu.se/o.o.i.s?id=24732&postid=1717706

http://person2.sol.lu.se/JoelParthemore/papers/thesis/chapter7.pdf

Chapter 7
The Co-Emergence of Concepts and
Experience
The account of concepts I have offered so far has been ambitious, but it has had a striking omission:
both the account of concepts and the concepts themselves have been given a largely static presentation. Chapter Two looked at the essential nature of concepts from a historical and contemporary
point of view. Chapter Three considered their core properties; Chapter Five their limitations (as
well as the limitations on any theories about them). Chapter Four attempted to place them into
a context of agents and referents and – relevant to the present chapter – of application; but the
discussion of application there was necessarily brief and preliminary, the attention elsewhere.
Chapter Six laid the groundwork for the present chapter, drawing together threads from the earlier
chapters to show how Gärdenfors’ conceptual spaces theory could account for concepts as both
representations and (non-representational) abilities, and otherwise meet the basic requirements I
had established for a theory of concepts; furthermore, how an extension of it I call the unified
conceptual space theory could move it in an algorithmically amenable direction toward something
more directly implementable in e.g. a computer model and therefore, one might reasonably hope,
easier to test empirically. Indeed, one of the goals of this work is to put forward a theory that
can be tested. At the same time, ironically, the unified conceptual space theory appears to do
what Chapter Four said one should and indeed could not: to detach concepts from both the agents
possessing and employing them and from any particular context of application, to turn them into
some kind of free-floating entities and not things that are actively being lived.
If Chapter Six provided a bridge from the opening chapters to this one, then this chapter is
a bridge from Chapter Six (the theory) to Chapter Eight (the intended application). Its goal
is to tell a particular story of concept acquisition and application, as conceptually distinct but
logically unified processes: two sides of one coin, two contrasting perspectives that we cannot quite
conceptually unify. It will use the theory from the previous chapter for telling the story from both
perspectives, one cognitively bottom up and the other top down. It will make much use of the
twin concepts of circular causality (Section 7.1.2) and co-emergence (Section 7.1.3), both familiar
from the enactivist literature, both involving the conceptual limitations and hiding the sort of
paradoxes I discussed in Chapter Five. Finally, it will return us to the notion first suggested in
Chapter Two: that representationalist and anti-representationalist perspectives are, on their own,
critically inadequate; that whether concepts look more like representations or more like abilities
will depend, finally, on where the observer stands.
139
CHAPTER 7. THE CO-EMERGENCE OF CONCEPTS AND EXPERIENCE 140
7.1 Concepts and Experience: A Tangled Relationship
We frequently compare the experiences we are currently having to memories of earlier
episodes. Sometimes, we experience something entirely new, but most of the time what
we see or hear is, more or less, the same as what we have already encountered. This
cognitive capacity shows that we can judge, consciously or not, various relations among
our experiences (Gärdenfors, 2004, p. 4).
I should begin by saying that combining bottom-up and top-down approaches to understanding
cognition is far from a new idea. The approach described here shares something of the spirit, though
not the style, of Antonio Chella et al’s (2000) approach to modeling dynamic scenes, which is more
narrowly (and pragmatically) focused on visual understanding (much of it quite low level) and on
the action portion of the conceptual space, as well as on how these could best be implemented in
a robotic system1.
Concepts and experience exist in a certain unavoidable tension. Experience is typically very much
engaged in the moment and, in any case, grounded in the present. Concepts, on the other hand,
abstract away from the immediate experience of the moment, stepping back from it to take a wider
view. They have, as it were – to extend an idea from Lawrence Barsalou (Barsalou et al., 2007)
who borrows it from William James – one hand in the past and the other in the future. Not only
is it unclear whether “concepts” solely of application to the present moment would be of any use,
it is unclear to what extent they would qualify as concepts at all.
At the same time, concepts and experience are critically dependent on one another. Unless one
wants to postulate a large body of innate concepts – and even the latter-day Fodor (2008) seems
reluctant to do that, at least in the usual sense of “innate” – most concepts will require experience to
give rise to them. As I will argue in Section 7.2, that experience must ultimately be sensorimotorbased experience: the agent must be cognitively and physically engaged with its environment to
experience either environment or self. The standard reference here, of course, is the classic study
reported by Held and Hein (1963) in which kittens, deprived during their early development of
the ability to interact actively with their environment, appear subsequently unable to make visual
sense of that environment: they are effectively blind.
Experience is likewise dependent upon concepts. It is an open question and, perhaps, an unresolvable one, in what sense of the word a fully non-conceptual agent – one without any concepts or any
degree of conceptual abilities – could be said to have experience. Perhaps a goldfish encounters the
world in this way. In such an agent the “experienced” world would, in every instance, be entirely
new. It might, in some limited fashion, have memory, as we think of it; but there would be no
relating to the past as the past or to the future as the future, for that would imply some, at least
minimal, conceptual abilities.
For the conceptual agent, such experience uncoloured by concepts (if experience it is) is, if I
am right, no longer a possibility. Such an agent will never experience the “now” entirely on its
own (though perhaps forms of meditation may get her closer than she would get otherwise, as
may waking from a really deep sleep or from anesthesia, before the world resolves itself into its
usual conceptual forms). Instead, the “now” is experienced in the light of past moments and in
anticipation of future ones, as the opening quote from Gärdenfors suggests.
1The paper also provides an excellent literature review of work in the area of dynamic scene recognition.
CHAPTER 7. THE CO-EMERGENCE OF CONCEPTS AND EXPERIENCE 141
Concepts reliably shape and re-shape our experience of the world: that is to say, they shape and
re-shape our world-as-experienced. Though there is no reason to think that anyone is born with
an innate concept of DOORKNOB – to borrow Fodor’s (1998) example – once an agent has the
concept then, in ordinary circumstances, that agent cannot fail to see a doorknob as a doorknob.
Not only does the agent, in Fodor’s language, become a reliable doorknob tracker; she cannot
choose to unlearn it or even temporarily step aside from that role. In the language of someone
like Alva Noë, once an agent has a sensorimotor profile of doorknobs, that profile is inextricably
part of how that agent encounters and experiences her world, where “the sensorimotor profile of
an object is the way its appearance changes as you move with respect to it (strictly speaking, it is
the way the sensory stimulation varies as you move)” (Noë, 2004, p. 78).
Concept acquisition and application go hand in hand. Acquiring concepts is a process of applying
concepts, which may themselves change in the process of acquiring new concepts. Within the
context of the last chapter, our conceptual spaces, individually and collectively, are both the
product of our interaction with our environment and the basis for it. To borrow a phrase from
dynamical systems theory, they constitute dynamically coupled systems.
7.1.1 Dynamical Systems
Perhaps the most distinctive feature of dynamical systems theory is that it provides a
geometric form of understanding: behaviors are thought of in terms of locations, paths,
and landscapes in the phase space of the system (van Gelder and Port, 1996, p. 14).
According to dynamical systems theory, “the cognitive system is not a computer, it is a dynamical
system” (van Gelder and Port, 1996, p. 3), where time and change over time play key roles. A
given system is a dynamical system if it is describable in terms of:
• A set of state variables, corresponding to a set of real numbers or points in state space, whose
changing values determine the present state of the system.
• A set of parameters or fixed points.
• Either differential or difference equations for relating them, depending on whether time is
being considered in a continuous or discrete manner.
Dynamically coupled systems are ones that interact in sufficiently rich ways as to constitute a
single, joint dynamical system.
For all of the common ground – notably, the common preference toward describing cognition in
terms from geometry – conceptual spaces theory and dynamical systems theory must part company
at certain points, particularly the tendency by many dynamical systems theorists to think that
all cognition can and should be described in terms of dynamical systems, and to deny any role to
“symbolic” or “computational” accounts. Consider:
. . . Dynamical and computational systems are fundamentally different kinds of systems,
and hence the dynamical and computational approaches to cognition are fundamentally
different in their deepest foundations (van Gelder and Port, 1996, p. 10).
CHAPTER 7. THE CO-EMERGENCE OF CONCEPTS AND EXPERIENCE 142
Or:
According to this ambitious doctrine the domain of the computational is empty, and
dynamical processes will eliminate their computational competitors across all aspects
of cognition (van Gelder and Port, 1996, p. 31).
More to the point, dynamical systems theory provides the wrong sort of tools with which to
separate concept acquisition from application, and to tell a linearly ordered and causal story for
each, as one might reasonably want to do. For that, we need a different tool, albeit one favoured
by some of the same theorists: i.e., circular causality.
7.1.2 Circular Causality
Figure 7.1: A model of circular causality.
Figure 7.1 presents the most basic model of circular causality: experience (a) gives rise to concepts
(b), which in turn give rise to experience (a).
Although philosophers – particularly, perhaps, those within enactive philosophy2 – often talk about
circular causality as though it were just an alternative account of causal relations, it is not; at least,
it is not that simple. Conventionally speaking, “cause” and “effect” are different things, related both
by order (cause is supposed to precede effect) and necessity (the cause is sufficient for the effect
necessarily to follow). In circular causality, what is cause and what is effect are purely matters of
perspective, and every effect is, by tighter or wider loop, its own cause. If this sounds paradoxical,
it should. Indeed, it should bring to mind the talk of eternally receding targets from Chapter Five.
I should not be understood as favouring a traditional, linearly causal model. Linear causality has
its own issues, as David Hume (2003) is perhaps most famous for pointing out – most particularly
for where the necessity relationship lies: between cause and effect, or only (as Hume himself
concluded) in the perspective of the observer. It might seem that one is presented with nothing
more than reliable co-occurrence (b has always been observed to follow a) and assigns it the status
of logical implication (if a then b; a, therefore b).
2I have in mind someone like Thompson (2007), though the idea is important in (Varela et al., 1991) as well.
CHAPTER 7. THE CO-EMERGENCE OF CONCEPTS AND EXPERIENCE 143
Neither am I rejecting circular causality, which I require for the account I wish to offer of concept
acquisition and application. I am only pointing out what should, perhaps, be obvious: that the
two perspectives offered by Figure 7.1, that a causes b on the one hand and that b causes a on the
other, only make sense if considered separately from each other: i.e., as two independent instances
of linear causality. Considered jointly, together with how causality is being defined in the first place,
no consistent interpretation can be given.
Figure 7.2: A dragon swallows its own tail. (Photo downloaded from Wikimedia Commons:
http://commons.wikimedia.org/ and edited.)
A circularly causal account may be very useful – even, in some cases, a necessary fiction; but push
any such account too far, and one will derive a contradiction (see the discussion in Section 5.2.1.3,
and compare Figure 7.1 with Figure 7.2). Consider the science fiction scenario of the person who
travels back in time to become his own father (and therefore his own paternal grandfather, and
his own great-grandfather, etc.). Although it may make for good story telling, it is impossible to
make coherent sense of it.
7.1.3 Co-Emergence
So, we are left with the perspective toggling introduced in Section 5.4, where each perspective is
like one half of the circle in Figure 7.1. I can account for the emergence of concepts from experience
(Section 7.2), and I can account for the emergence of experience from concepts (Section 7.3). I can,
and will, argue (Section 7.4) that no priority can logically be made between the two accounts (even
though a linear text insists that I put one before the other, and both logic and convention suggest
placing acquisition first: one cannot make use of what one does not possess). What I cannot do is
offer one unified account, based on one unified perspective. That is not, however, the fault of the
account.
Concepts and experience are co-emergent: each (to borrow Varela’s phrase) “brings forth” the
other. It’s like the chicken-and-the-egg problem: which comes first? Something, of course, must
CHAPTER 7. THE CO-EMERGENCE OF CONCEPTS AND EXPERIENCE 144
logically start things off, but it need not be either concepts or experience as we understand them;
especially if, as I have been suggesting, each presupposes the other. Caught within our conceptual
perspective (see Section 5.2.1.2), we cannot step outside of it: we cannot simply put our concepts
or our conceptually structured experience aside.
7.2 Concepts Emergent: The Acquisition Story
What transforms the “booming, buzzing confusion” that enters our eyes and ears at
birth into that orderly world we ultimately experience and interact with (Harnad,
1990c, p. ix)?
Concepts – as described by Fodor or Prinz or Millikan or Gärdenfors or pretty much any theorist
of concepts – are paradigmatically abstract and cognitively high level. If this is most clearly
true on the concepts-as-representations accounts, it is only marginally, if at all, less true on the
concepts-as-abilities accounts.
Sensorimotor engagements are paradigmatically concrete and cognitively low level. On the face of
it, the two could not be further different and removed from each other.
Yet it would be a mistake, I believe, to take these either as fully independent systems (such that
an exploration of the former could take place without any consideration of the latter)3; or as semiindependent and interacting systems needing somehow to be fitted together. At the same time, the
one cannot be eliminatively reduced to the other, as some dynamical systems theorists are inclined
to do (see Section 7.1.1), and as Vittorio Gallese and George Lakoff favour in (2005). Rather, they
are positions toward either end of a continuum, neither of which can be divorced from the other;
and experience straddles both.
Mental content and meaning do not “come for free”: at some point they must begin with, or be
caused by, that which is not mental content, that which does not already have meaning. That
is to say, they must be grounded. Noë (2004), Harnad (2007), and others have written that all
mental content, conceptual or otherwise, must be grounded in specific sensorimotor engagements,
and so sensorimotor engagements are partly constitutive of that content. I agree. Note that this
makes mental content not fixed but contingent: if I do this, then I will experience that; if the
agent does this, then the agent’s mental content will be that. Mental content is defined out of
interaction; there is no logical separating of input from output, as the dynamical systems theorists
would strongly concur. This is why one talks, not of a sensory system and a motor system, but of a
sensorimotor system. Note, too, that this is consistent with (indeed should be seen as a refinement
of) the classical empiricist tradition that grounds cognition in experience.
3I have in mind someone like Fodor.
CHAPTER 7. THE CO-EMERGENCE OF CONCEPTS AND EXPERIENCE 145
7.2.1 Noë’s Sensorimotor Account
. . . We ought to reject the idea – widespread in both philosophy and science – that perception is a process in the brain whereby the perceptual system constructs an internal
representation of the world (Noë, 2004, p. 2).
Much of what follows is consistent with Noë’s brand of enactivism, as noted in Section 6.1.2.
Nonetheless, I do take issue with Noë’s account at a number of points4.
• Noë favours a more linearly structured account than I.
Noë’s attention in his 2004 book is on how cognition consists of or is built upon continuous
sensorimotor engagement. Perception is never truly passive but should be understood, even in
its apparently passive instances, as an engaged physical activity. This is important, and either
downplayed or missed entirely in many older accounts of cognition, which are not unfairly
caricatured as over-intellectualizing matters. But one could equally turn the perspective
around to look at how sensorimotor engagement consists of or is built upon perceptions: how,
as a matter of principle (on anti-realist accounts) or practice (on certain realist accounts), the
mind constrains the world. To wit: Noë’s causality is linear, from sensorimotor engagements
to sensorimotor profiles to higher cognition; my preferred model of causality, as noted above,
is circular.
• Noë’s account is overly forward looking.
I believe that Noë’s account is much more forward- than backward-looking: Noë has a lot
to say about where expectations take us but relatively little to say about where they come
from (as opposed to e.g. Chrisley’s Expectation-Based Architecture approach (Chrisley and
Parthemore, 2007b)). This might in part be a consequence of Noë’s strong externalism;
talking about where expectations come from might seem to require a more internalist-looking
perspective, one that has more to say about the mental life of the agent.
• Noë’s account is strongly externalist; I wish to reject the extremes of both internalism and
externalism.
As Anthony Morse and Tom Ziemke have written (2010), Noë focuses on sensorimotor contingencies to the exclusion of any consideration of the agent’s bodily states in general or
emotions in particular. Without an account of emotions and, consequent to that, of motivations, Noë has no explanation for why some affordances in the environment are salient and
others not. This follows directly from Noë’s externalism. The sort of enactivism I have endorsed (see Section 6.1.2) views either internalist or externalist perspectives, when taken on
their own, as deeply misleading. One of the important things to remember about affordances
is that they are not in the environment (as Gibson (1986) is often read) but (metaphorically
speaking) in the interaction, always relative to the perspective of the agent at a particular
moment in time (as a more nuanced reading of Gibson might allow).
• Noë’s account is strongly anti-representationalist, whereas I favour a qualified representationalism.
Finally, and perhaps most critically for my account, is the reason why I classified Noë (see
Section 2.5.3) on the “abilities” side of the abilities vs. representations divide: Noë eschews
representational language. As should be clear by now, I take that to be a mistake. At the
same time, I want to endorse fully the quote of Noë’s that began this section.
4Much of the content of this section I owe to informal discussions with Anthony Morse, my co-author on (Parthemore and Morse, 2010).
CHAPTER 7. THE CO-EMERGENCE OF CONCEPTS AND EXPERIENCE 146
7.2.2 More General Issues With Sensorimotor Accounts
There are further problems, not particular to Noë’s account, but common to any accounts that
might be seen to focus too narrowly on sensorimotor explanations. The difficulty is how one gets
beyond specific sensorimotor engagements: how one generalizes to the sensorimotor profiles needed
to explain specific affordances, nevermind abstract conceptual thought. This is what Harnad has
called “the problem of extracting reliable categories from experience” (1990a, p. 538).
Take for example the account of Gallese and Lakoff (2005), according to which the most abstract
of concepts is, on any occasion one can name, no more than a specific sensorimotor engagement,
albeit with parts of it (e.g., the full activation of the motor cortex, with consequent movement)
suppressed. The concept they focus on is that of grasping. The similarity between their account
of grasping and George Berkeley’s account of triangles (1999) is striking. For Berkeley, the general
concept of a triangle is nothing more than a specific instance of a triangle with details suppressed:
the length of the sides or the measure of the angles. For Gallese and Lakoff, the general concept
of grasping is likewise a specific instance of grasping with details suppressed: i.e., the “actual”
carrying out of the action.
One might be tempted to cede the point for grasping and reserve concerns for concepts like that of
democracy or enlightenment, or such extreme abstractions as the concept of a concept itself; but
that would, I think, be a mistake. To the extent that one’s concept of grasping is a representation
of grasping – and I have argued that, for the conceptually reflective agent, all concepts take on
this representational aspect – the representation may have as much (or as little) to do with the
represented as e.g. a representation of a dog (such as a painting) has to do with a dog. I have
further argued that the role of all concepts is to simplify (Section 5.2.2) and to abstract away from
the particulars of context, from (pace Gallese and Lakoff) any particular application; in which
case, some additional mechanism or mechanisms are needed. I propose that the unified conceptual
space meets this requirement.
Of course I, too, want to reject a sharp class/instance distinction, but not by merely eliminating
it. By trading on an essential ambiguity – according to which all instances can (within practical
limits) be treated as classes and all classes can be treated as instances of more general classes (see
Section 6.1) – I will show how one can get from either one to the other, on whatever cognitive
level.
7.2.3 An Alternative Account: Sensorimotor ++
How can one generalize from single observations to general laws (Gärdenfors, 2004, p.
205)?
The ANN [artificial neural network] model I outline here is based on Kohonen’s. . .
self-organizing maps. The purpose of the method is to reduce the representational complexity of the input in an efficient and systematic way. Thus the proposed method can
be seen as a way of answering [the] question. . . on the subconceptual level (Gärdenfors,
2004, p. 221).
As the unified conceptual space theory was presented as an extension of Gärdenfors’ work, so the
position I would like to take on sensorimotor grounding of cognition should be understood as an
extension of Noë’s work, whilst addressing the above reservations. One might be tempted to call
CHAPTER 7. THE CO-EMERGENCE OF CONCEPTS AND EXPERIENCE 147
it “sensorimotor plus plus”5: sensorimotor engagements plus somatic and other bodily information (per Damasio (2000) and Morse and Ziemke (2010)) plus (with appropriate qualifications)
representational language, as located within a conceptual spaces framework.
As Gärdenfors’ diagnoses, the emergence of cognition in general and concepts in particular from
their pre-conceptual and pre-cognitive origins is largely a matter of induction. Although we may be
most comfortable thinking about induction as a very abstract and high-level cognitive process, an
analogous process must logically, it seems, be going on at the lowest levels of cognition. Gärdenfors
couches his account of low-level induction in terms of Kohonen maps, a special kind of artificial
neural network that preserves the topological relations of the input space. The details of Kohonen
maps is not important here. I want to offer a different but, I hope, complementary account, first
abstractly and then through the specific mechanism of the unified conceptual space theory.
As on standard connectionist or associationist accounts, the story begins with pattern recognition,
with the caveat that there should be some, minimal, pre-existing ideas in the system about what
sort of things patterns are (in the most general possible way). These constitute a minimal set of
primitive building blocks – proto-concepts: mental structures that are concept-like yet fail to meet
one or more of the intrinsic properties of concepts, such as being under the agent’s endogenous
control. Such structures should be governed by nomic relations (of the kind endorsed by Fodor
(1998) for concepts in general) rather than experientially derived. In terms of conceptual spaces
theory, they provide an initial partitioning of the unified conceptual space (see Section 7.2.4), one
from which all subsequent partitioning is derived, through a process of recognizing regularities
in the perceptual stream between one moment and another and another: repetitions that are
recognized as somehow salient.
I will not attempt to offer any proper account of salience. Such an account lies far beyond the
scope of the present work. Nonetheless I find intuitively appealing the attempt by many in the
enactivist camp – recently e.g. by Evan Thompson and Mog Stapleton (2009) – to ground minimal
salience in the survival of the organism. What is salient is what enables the organism to survive,
and other saliences should follow directly or indirectly from there. More of the detail can be fleshed
out on a somatosensory account such as offered by (Morse and Ziemke, 2010).
The most basic perceptual regularity, then, is a mapping between a set of points in “raw” (but
still proto-conceptually structured, and hence minimally interpreted) experience: something was
observed, and then it was observed again, and again. To borrow an example from robotic vision,
it could be something as simple as a recurring pixel at the same location in a visual display, or a
sudden change of pixel value.
An agent can derive many such regularities. Just as one’s “raw” perceptions define a perceptual
space, so, too, these regularities collectively define a space of their own: a space of minimal
regularities, albeit one that preserves something of the topology of the original perceptual space.
The trick is that, just as patterns could be discovered in the original perceptual space, so too,
patterns can be found in the higher-level regularity space. The latter are patterns of patterns,
regularities in the regularities.
5The name was suggested by Gärdenfors.
CHAPTER 7. THE CO-EMERGENCE OF CONCEPTS AND EXPERIENCE 148
Figure 7.3: Patterns, patterns of patterns, and patterns of patterns of patterns.
Consider Figure 7.3. In the first row, digits are highlighted where there is a repetition of three or
more of the same digit in any column or row of a grid of digits. In the second row, a second-order
pattern, that of a common rectangular shape, is extracted. In the third row, the shapes are placed
inside one another relative to their size: a third-order pattern. Of course, this same idea could be
extended through any number of steps with other examples.
So regularities are found among the first-order regularities, yielding second-order regularities; regularities are found among the second-order regularities, yielding third-order regularities; and so
on. At each level, the agent steps further back from the moment, and the “moment” itself (i.e., the
minimal individuable unit of time) becomes more and more stretched out.
Regularities in the regularities, and regularities in those regularities, yield increasingly complex,
increasingly abstract mental content that, as the process iterates over and over, eventually becomes
recognizable as first- then higher-order concepts, iconic then symbolic representation, unconscious
(non-introspectible) then self-conscious (introspectible) content. The result is an associational
hierarchy that at its base is strongly associational and at best very weakly symbolic, at its summit
strongly symbolic and at best very weakly associational.
CHAPTER 7. THE CO-EMERGENCE OF CONCEPTS AND EXPERIENCE 149
Figure 7.4: Perspectives on a cognitive continuum.
More accurately, the “hierarchy” should be understood as a continuum. At one end, one is, as it
were, drowning in a sea of detail; go too far in the other direction, however – toward the very
rarefied, the very abstract – and no useful detail remains.
See Figure 7.4: somatic markers (the term originates with Damasio (Damasio et al., 1991)) combine
with, on the one hand, sensory capacities and motor abilities; on the other hand, proto-concepts.
Together these give rise to sensorimotor profiles. From there, one can follow either of two paths,
representing two contrasting perspectives. The narrow blue line down the middle represents the
unresolvability of these two perspectives.
As one ascends through the associational hierarchy (following the solid black lines), the richness
(dimensionality) of the referring structures is reduced at each step and the richness (dimensionality)
of the referent structures (the target descriptive space) is likewise increased; compare Gärdenfors
in describing the working of this process in the case of Kohonen maps:
. . . Points closely related in the high-dimensional space are mapped onto closely related
points in the low-dimensional space. Since dimensionality is reduced, this entails that
regions of the high-dimensional [perceptual] space are mapped onto points in the lowdimensional [conceptual] space (2004, p. 222).
Look back at Figure 7.3: the second row simplifies and abstracts away from the first, the third row
from the second. At the same time, the semantic content – the represented pattern – is enriched
at each step.
CHAPTER 7. THE CO-EMERGENCE OF CONCEPTS AND EXPERIENCE 150
In the limit, the referring structures come increasingly to look like arbitrary symbols, whose form
bears no obvious relation back to any particular context; while the referent structures are maximally
structured. The referring structures become mere pointers to richly structured sub-regions within
the unified conceptual space. In this manner, the unified space is gradually transformed from a
largely unstructured space to a space of many spaces.
If one inverts the associational hierarchy (following the dotted black lines downward), then what
at first looked very much like symbols will shift into iconic representations and gradually lose
themselves in context as their meaning becomes more and more defined by that context, until most
(if not all) of what we understand by symbols disappears, and we are left with “bare” interactions.
I can now, finally, return to and refine the definition of “concept” first offered in Chapter One:
A concept is a synchronized pattern of relatively abstract, relatively higher-order association between some aspect of the mental world of the agent (“self”) and some matching
aspect of her experienced environment (“non-self”).
A concept is recognizably a concept to the extent that it abstracts away from the particulars of
context, even as it is always then applied back to particular contexts. It both abstracts away from
and is structurally isomorphic to its referent in perception. It need not necessarily correspond in
any way to whatever caused that perception, only vary as whatever caused that perception varies6.
I will return to and revise this definition once more before I close the chapter.
7.2.4 Partitioning the Conceptual Space
Clearly, no organism is born a blank slate. Some categories are innate (Harnad, 1990b,
p. 2).
Presumably, the child masters only a few prototypes in animal space and these prototypes are used to generate a partitioning of the entire space. Consequently, the child
will overgeneralize a concept in comparison to its standard use. . . . When the child
learns more prototypes for other animal concepts, however, it will gradually adjust an
early concept to its normal use since its partitioning of the animal space will become
finer (Gärdenfors, 2004, p. 125).
The coming to be of a conceptual mind is the progressive partitioning7 of a conceptual space that
is, at the same time, a space of spaces: what I have termed the unified conceptual space. I have
already suggested, at several points, that the conceptual mind does not start out as a tabula rasa
but is already – albeit minimally – structured. This is by virtue of a small set of innate protoconcepts and a small set of “rules” (identifiable as such only by some theorist) for breaking them
apart (to make more subtle distinctions), combining them together, adjusting their boundaries,
and otherwise matching them against experience. Such is the symbolic (or rather, symbol-like)
heart of what, at the most basic level of cognition, is otherwise an all but entirely non-symbolic
6Compare Gärdenfors (2004, p. 109): “. . . Representations need not be similar to the objects they represent. What
is important is that the representations preserve the similarity relations between the objects they represent. . . .”
7Compare Dominic Massaro’s notion of categorical partitioning (1990). Note that, for Massaro, this partitioning
is not already present in perception but is added later.
CHAPTER 7. THE CO-EMERGENCE OF CONCEPTS AND EXPERIENCE 151
matter (see again Figure 7.4). The suggestions for proto-concepts are taken directly from the
earlier discussions in sections 4.2.2 and 6.2.2. These are the great-great-great-etc. grandparents of
all the other, “true”, concepts.
It is important to emphasize, again, that proto-concepts are not concepts: that is, the structuring
of the unified space begins at a level far below the conceptual. At the same time, proto-concepts
are not entirely non-conceptual, either, nor is the conceptual space they structure: otherwise it
would not be a conceptual space. As noted earlier, one need not accept McDowell’s position that
experience is fully conceptual “all the way out” to allow that it is conceptually coloured all the way
out (see Section 5.2.3.2).
7.2.4.1 Initial Partitioning
This initial situation is described in Figure 7.5 (cf. Figure 6.6). Object- and action-type things
dominate the space, all of which is compressed below the level of first-order concepts (but above
zeroth-order).
Figure 7.5: Initial partitioning: minimal structure.
A few caveats are in order. First, the distinction between zeroth-, first-, second-, nth-, and higherorder concepts should not, per Section 4.1.1, be taken as a series of discrete levels but rather rough
positions along a continuum. Those positions relate to the degree of explicit reflection required
to grasp the concept: that is, higher-order concepts, like the concept of a second-order concept,
require not just thoughts about thoughts but some level of active awareness of thoughts about
thoughts. Second, one might be tempted to substitute in Rosch’s (1975; 1999) labels of subordinate,
basic-level, and superordinate for first-order, second-order, and higher-order; but that would be a
mistake, for although Rosch’s labels do, indirectly, relate to how abstract categories are on a scale
of concrete/physical to abstract/mental, they are first and foremost about the relationship between
“parent” and “child” categories. That is, following the discussion in Chapter Six, they lie along the
axis of generalization (Section 6.2.1.1) not the axis of abstraction (Section 6.2.1.4), allowing that
those axes do, at one extreme, converge.
CHAPTER 7. THE CO-EMERGENCE OF CONCEPTS AND EXPERIENCE 152
7.2.4.2 Subsequent Development
As the partitioning proceeds and the space expands into the area above first-order concepts,
object-type things (now recognizably objects) and action/event-type things (now recognizably action/events) continue to dominate. See Figure 7.6. Particularly at the proto-conceptual level, the
space becomes recognizable as a Voronoi tessellation (see Figure 6.1, which the proto-conceptual
layer in this figure is meant to resemble). Initial tendencies will be to overgeneralize, as Gärdenfors reminds us (see the opening quote for Section 7.2.4). Notice that, in line with the comments
about Rosch above, although conceptual development does proceed from the sub-conceptual to
the conceptual to the meta-conceptual, it does not proceed from the subordinate to the basic-level
to the superordinate. Rather, it jumps first to the basic-level and expands to subordinate and
superordinate from there, as Rosch notes in describing some of her empirical results:
. . . Basic objects were shown to be the first categorizations made by young children,
and basic object names the level of abstraction at which objects are first named by
children and usually named by adults (1975, p. 587).
Figure 7.6: Initial first-order concepts and distinctions.
That said, one of the most basic “object” distinctions to be made is between self and non-self (see
Section 4.1.3), and one of the most basic “action/event” distinctions is precisely between (intentional) actions and (non-intentional) events (see Section 4.2.2.2). These would seem, intuitively, to
be foundational to all our other conceptual distinctions: so e.g., a child must be able to distinguish
self from non-self before proceeding to recognize, or name, basic objects.
CHAPTER 7. THE CO-EMERGENCE OF CONCEPTS AND EXPERIENCE 153
7.2.4.3 Advanced Partitioning
Figure 7.7: Higher-order concepts and further distinctions.
As meta-cognitive abilities develop (the capacity for thoughts about thoughts, and for thoughts
about those thoughts), the partitioning expands further toward higher-order concepts, giving more
and more of the space the appearance of a Voronoi tessellation (see Figure 7.7). Properties take on
increasing importance, even while sub-partitioning of object and action/event subspaces continues.
To the extent that the space is still largely unstructured, however, there will remain a consequent
tendency to over-generalize. Per Gärdenfors’ account, this problem will persist (and perhaps
never entirely go away: note e.g. the categorization differences between bird watchers and bird
“novices”, or between expert mushroom gatherers and the greater number of people who cannot
tell a mushroom from a toadstool).
7.2.5 Mapping Conceptual Space Onto Conceptual Space
. . . Humans have powerful abilities to detect multiple correlations among different domains. . . . In the theory of conceptual spaces, this kind of inductive process corresponds
to determining mappings between the different domains of a space (Gärdenfors, 2004,
p. 228).
The process I have just described, where the unified space begins as largely if not entirely unpartitioned and ends up intricately partitioned, is a movement from the maximally general (something
is salient) to the maximally specific (the salient thing is e.g. my weight on the scales in my bathroom at 7:03 this morning). Here are concepts under what I termed the “first description” (see
Section 6.2.1), as well-behaved shapes in the unified space .
However, there is a competing perspective that is just as valid, whereby concept formation is a
movement from the maximally specific (applicable to the narrowest possible range of contexts) to
the maximally general (applicable to the widest possible range). This is because, at the same time
that the unified space is being partitioned, it is becoming more and more structured in another
way, as parts of it map onto each other, both through conceptual reference, in those cases where
the referent also lies within the unified space, and through filling in the details of components,
CHAPTER 7. THE CO-EMERGENCE OF CONCEPTS AND EXPERIENCE 154
parameters, and contextuals. Here are concepts under the “second description” (Section 6.2.2), as
“sets of logical relations”.
These mappings are initially very limited, rendering the proto-concepts and their immediate derivatives applicable only within quite narrow contexts, things to be used and quickly discarded (or
recycled). Over time they become increasingly sophisticated, allowing agents to detect “multiple correlations among different domains”. So on one end of the continuum, one has maximally
general proto-concepts that are applicable only within extremely specific contexts (everything is
just a [unique] thing); at the other end, one has maximally specific concepts (relating only to one
particular thing) able to locate their referents in the broadest possible range of contexts. Compare
this to the earlier discussion in Section 7.2.3.
The description of space mapping onto space in this way, given in Section 6.2.2, was very high
level. At its most basic level, it should begin with distinguishing proto-conceptual instances one
from another by some proto-conceptual “understanding” of their different properties8, using those
properties to begin to organize them into a hierarchy of classes. Some of them will have parts
(components) relating to each other in a predictable fashion. Proto-conceptual object-type things
will get associated both with other object-type things and with various action/event-type things;
the same will happen in reverse for proto-conceptual action/event-type things.
Figure 7.8: Mapping between different parts of the unified space.
Consider Figure 7.8. The purple lines represent components: a rock can “consist of” rocks or sand;
sand consists of grains. Cyan lines represent parameters: both rocks and sand have colour, which
itself has the parameters of hue, saturation, and brightness. Both throwing and falling have the
parameter of speed. Red lines are associational: both rocks and sand are associated with beaches.
Rocks, but not sand, are associated with falling and throwing, both of which are associated with
(among other things) rocks.
8. . . That is, their parameters (or, in Gärdenfors’ terms, integral dimensions).
CHAPTER 7. THE CO-EMERGENCE OF CONCEPTS AND EXPERIENCE 155
7.3 Experience Emergent: The Application Story
A brain is interposed between the stimulus and the response, totalizing the sensory
input of the organism into an internal representation of what goes on in and about it
(Torey, 2009, p. 6).
Of course if the account given so far – of concept acquisition – was all there was to be said,
then concepts really would be static entities, with no means for update or obsolescence. But
concepts have no value, no meaning, unless at the same time they are being acquired they are
being applied. On the account I am offering, they are at least as much skillful abilities as they are
expressible knowledge (cf. (Morse and Ziemke, 2007)). Furthermore, on any conceptual-spacesderived account, they will be fundamentally dynamic entities.
To borrow a page from the classical definitionists (Section 2.3.1), the concept acquisition account
just outlined can be turned on its head, verifying instead of discovering, disassembling instead of
assembling, in the same spirit in which definitions are neutral as to whether they are defining new
concepts or identifying and verifying old ones. Before, concepts were being abstracted away from
experience, away from the particulars of the moment. Here, concepts are being applied back to
experience, back to the particulars of the moment.
Unfortunately we do not have, as we had with Noë and the mostly bottom-up-driven process of
concept acquisition, a similar guide to the mostly top-down-driven process of concept application.
I will need, instead, to help myself to several diverse guides.
As do many in the enactive camp, I reject as not very useful (as a general model for cognition, even
viewed top down) the cognitivist input-output-based model of cognition, exemplified by SMPA:
sense-model-plan-act. Although there are, of course, many differences, what cognitivist approaches
generally have in common is a splitting apart of sensory input from motor output and a treatment
of higher-level cognition as independently explainable from lower-level details of “mere” implementation. That said, they remain surprisingly popular: witness the recent publication of Torey’s book
(2009), which ironically is, in many ways, much more traditional than it is revolutionary.
Torey’s approach, as neatly summarized in the opening quote, is precisely the sort of representationalist account I do not want to give. Remember that representations as I intend them (see
Section 2.6.4) are neither internal nor external, nor are “mental” representations mental in a way
that other representations are not. Rather, representations are an ineliminable part of our firstperson perspective. They stand not between the pre-experiential agent and his pre-experiential
world but between the (self- or other-) perceived agent and the perceived world.
Questions about the nature and existence of representations have formed a running theme throughout this work. Representations are less ontological reals than perspectives we take – perspectives
that, at the same time, we cannot simply set aside. Perhaps one should, in the end, talk not of
representations but of a representational stance. I will return, one final time, to the question of
representations in Section 7.4.
CHAPTER 7. THE CO-EMERGENCE OF CONCEPTS AND EXPERIENCE 156
7.3.1 Concepts as Expectations
. . . The less we just stare at the hammer-Thing, and the more we seize hold of it and
use it, the more primordial does our relationship to it become, and the more unveiledly
is it encountered as that which it is – as equipment (Heidegger, 1978, p. 98).
With one metaphorical eye to the past and one to the future, concepts are the expectations that
drive experience. Consider concepts as a tool that, once you have it, you literally cannot imagine
doing without, to the extent that the tool may be incorporated into your core self-image. (See the
earlier discussion in Section 4.4.3.)
Perhaps concepts are like language in this way. For many people, language is so much a part of
their thinking that it seems that their thought just is structured as words of a (spoken or written)
language. As noted in Section 3.3.2, Torey (2009) follows Davidson (1987) in making language the
basis of his account of cognition: no language, no thoughts, and no mind.
Consider conceptualized experience as an emergent projection over top of non-conceptualized experience, all but obscuring it. Once we become aware of past as past and future as future, we
cannot help experiencing the present moment in light of both. In Damasio’s language (2000, pp.
195-233), we begin telling the narrative that gives us our rich sense of autobiographical self.
If concepts are a tool, then perhaps the metaphor is Heidegger’s hammer (1978). Only when the
hammer breaks or the nail bends – only when the hammer fails somehow to perform as a hammer
– do we stop and see the hammer as a hammer. We see, hear, and feel what our concepts lead us
to expect until the match between expectations and current experience breaks down in a manner
that we cannot ignore. Only then are we forced to take a closer look, at which time at least some of
our implicit conceptual expectations are made explicit. The unintended and seemingly paradoxical
consequence is, as Heidegger notes, that we lose the “unveiled” experience of the very thing we are
trying to get clear about.
This account of concepts as expectations is very reminiscent of Chrisley’s Expectation-Based Architecture (EBA) (Chrisley and Parthemore, 2007b), even though Chrisley grounds those expectations
(correctly, I think!) in non-conceptual content. Like the sensorimotor++ account (Section 7.2.3),
the EBA is trying to combine the best elements of enactivism (in the case of the EBA, specifically Noë’s version of it) and a kind of representationalism, while avoiding many of the common
criticisms of both. Its key insight is to introduce “counterfactual representational vehicles”: not
presently tokened representations, but representations that would be tokened were one’s (conceptual or non-conceptual) expectations to be met.
CHAPTER 7. THE CO-EMERGENCE OF CONCEPTS AND EXPERIENCE 157
Figure 7.9: Concepts-as-representations-as-control: Consider three students in a tree (middle) and the recognition
of three students in a tree (left and right). At a proto-conceptual level, that recognition will be straightforwardly
picture- or painting-like (left). At a conceptual level, that recognition will be much more abstract, focusing on the
logical relations of parts of the image to each other (right) - cf. Figure 7.8 – even while retaining links back to its
proto-conceptual roots.
It is reminiscent as well of Imogen Dickie’s (2006) notion of “representation as control”, whereby
representational (conceptual) expectations, based on past experience, guide and massively simplify
the agent’s interaction with its environment. (Dickie presents her own approach as preserving the
best parts of the early Wittgenstein’s “picture theory” (Wittgenstein, 1922). In brief, rather than
the world being in any meaningful sense a “picture”, it is rather that we represent it to ourselves
and to others in a certain picture-like way. See Figure 7.9.) The problem, as Gärdenfors noted
for us in Section 5.2.2, is that “. . . the information received by the receptors is too rich and too
unstructured” (2004, p. 21).
Not surprisingly, concepts-as-expectations present several trade-offs. As noted in Chapter Five,
one simplifies in order to understand. But if one over-simplifies, one no longer understands.
The set of patterns potentially discernible in any perceptual context is unbounded and may be
infinite (see the discussion in Section 4.3 and (Dennett, 1991b, pp. 33-35)). Any context can be
perceived from any of a bewildering variety of perspectives. Attention is limited by finite resources;
there is ample psychological evidence that working memory can attend to only a small number of
items at any time (somewhere around seven (Miller, 1994)).
Ask the people watching a basketball game to count, and report, the number of times the ball
bounces, and they will consistently fail to see a gorilla walking across the court (a phenomenon
known as inattentional blindness); a control group with no such instructions will be far likelier to
see the gorilla (Simons and Chabris, 1999). Martin Langham reported on people who pull out in
front of motorbikes, who “look but fail to see”. What he found was that inexperienced drivers look
all over the place. Experienced drivers minimize where they are looking (Langham, 1999).
The lesson more broadly is this: experience teaches us to become more and more selective of what
we attend to. A simplified experience of the driving scene leads, in most instances, to improved
performance. A simplified experience of the world in general may, in many instances, lead to
improved performance, even better survival and reproductive opportunities. A simplified world
is easier to understand and respond to. But sometimes the simplified model will make mistakes,
CHAPTER 7. THE CO-EMERGENCE OF CONCEPTS AND EXPERIENCE 158
because the simplified model is not the original. Something has been lost. The driver pulls out
right in front of the motorbike, even though he swears, truthfully, that he looked and saw no one.
But beyond all of this, the more conceptual knowledge we have, the more we come to rely on it. As
a wealth of further psychological evidence shows, most of the time we see not what is in front of us
but what we expect to see. Perception is not independent of reality, nor is it solely constituted by
it! Instead of simply revealing the world, concepts help to construct it even as they are constructed
by it.
Concepts have this dual nature: if their nature is increasingly abstract, their application is always
specific to a context. To return to and refine my earlier working definition – one last time, for now!:
A concept is (or could be described as) a synchronized pattern of relatively abstract,
relatively higher-order association between some aspect of the mental world of the agent
(“self”) and some matching affordance(s) of her experienced environment (“non-self”),
such that the affordance(s) implicitly or explicitly specifies the necessary, sufficient,
and customary (or contextual) conditions for its application relative to any particular
moment.
With this revised definition, we can begin to see how conceptual spaces theory and the unified
conceptual space theory come into their own. Just as representations are neither internal nor
external, being relational entities, standing between an agent and her perceived environment; just
as concepts are, likewise, juxtaposed between the agent and her environment, created out of their
dynamic interaction; so a theory of concepts properly belongs between associational and symbolic
accounts of cognition, showing how conceptual mental content arises from the dynamic interaction
of symbolically interpretable structures with symbol-free associations, the continuous interaction
of the cognitively abstract with the sensorimotorly concrete.
7.3.2 Mapping Conceptual Space onto Perceptual Space9
Structuring the unified space was described as a bottom-up, layer-by-layer hierarchical (or continuous) process of pattern recognition: finding patterns in patterns, patterns in patterns of patterns,
and so on, until the nth generation patterns have all but lost their connection back to their
perceptual origins. Putting the resulting space into use is, by contrast, best understood as a topdown-driven, layer-by-layer process of pattern matching, a kind of “de-layering”: a return down
through levels of the hierarchy (or through the continuum), toward particular encounters and toward parts as opposed to wholes. If an X violates expectations, consider previous experiences with
similar X s or X -like things, or decompose the X into e.g. its functional parts. (For an illustration
of this, see Section 7.3.3 and Figure 7.10.)
For concept acquisition, concepts looked more like abilities. Associations and association building
were in the driver’s seat. For concept application, concepts look more like representations. Initially,
at least, symbols and symbol application are the more appropriate level of description (though only
some part of this need be consciously articulable by the conceptual agent herself).
9It may be useful to compare the discussion here with that of Section 7.2.5.
CHAPTER 7. THE CO-EMERGENCE OF CONCEPTS AND EXPERIENCE 159
The basic idea is this: unified space can be fitted to perceptual space, concepts in the unified
space matched against their non-conceptual analogues in present experience, attempting the closest
match possible: a particular concept (most likely at the basic-level on Rosch’s hierarchy – see
(Rosch, 1975, p. 587)) to a particular sub-region of present experience. A precise match may push
one to further identify the match at the subordinate level. A mismatch, on the other hand, may
force one to jump to some other basic-level category entirely. (That “dog” in the shadows is not a
dog at all; it’s a rock. That “wailing cry” was not a woman in distress; it was the call of a loon.)
Alternatively, one might, as it were, relax the resolution: resorting to superordinate categories that
match against larger and larger portions of the unified space, going more and more general until
a match finally does occur. (At some point a match must occur, since everything perceivable is,
minimally, a something.)
So: sometimes a mismatch can be resolved by jumping to another location in the unified space.
Other times it can be resolved by “zooming out” and in again. Other times, however, there will
be breakdowns: failures of our conceptual expectations. Breakdowns suggest four strategies for
dealing with them, which we can order from least to most radical10.
1. Adjust the logical structure of the closest matching concept (Section 6.2.2) in terms of its
parameters, contextuals, and, if appropriate, components, so that a match now does occur.
Formerly you conceptualized all swans as white; now you conceptualize swans as either white
or black.
2. Alternatively, perhaps you did not conceptualize swans in terms of colour at all, but now you
must, in order to distinguish the white from the black swans. This is the equivalent, in terms
of traditional conceptual spaces theory, of adding an additional integral dimension (namely
colour).
3. Partition an as-yet-unpartitioned sector of the conceptual space. Formerly you recognized
only a general category of swans, all of which were white. Now you recognize two distinct
sub-categories of swans, one of which includes all the white swans and one all the black swans.
4. Most radically, remove partitioning from some sector of the conceptual space and re-partition
it: i.e., divide it into a different set of (convex) shapes. As opposed to conceptual change via
(1) or (2), this is conceptual obsolescence and replacement. This would be if your encounter
with black swans forced you to e.g. re-structure your whole “bird” space, classifying all birds
differently and not only swans.
10Cf. the discussion in (Aisbett and Gibbon, 2001, p. 210)
CHAPTER 7. THE CO-EMERGENCE OF CONCEPTS AND EXPERIENCE 160
7.3.3 On Encountering a Door (A Thought Experiment)
Figure 7.10: A door may be taken as an undifferentiated whole, a set of functional parts, or a similar instance to
some previous door encounter.
Consider a door that is in front of you (see Figure 7.10). Does your present experience of that
door as a door match your expectations at the most abstract conceptual levels with respect to
doors? If you don’t need to see the door as anything more than a whole with no parts (like an
“unstructured” symbol), then you won’t: it will register as an undifferentiated door (top of figure).
Of course, depending on where your attention is focused, you may choose or be motivated to look
more closely, by considering the door as a special type of door (like a trap door, or the kind of
door one finds on a bank vault). Alternatively, you might focus on its functional parts (lower left):
where is the handle?. . . where are the hinges?. . . does the door have a sign on it? Again, you
might consider the door in the light of particular past door experiences (lower right). Does this
door remind you in some way of one of them?
How you encounter the door will depend, in part, on what you want to do with it. If you need to
pass through the door, you will look, minimally, for how the door opens. If it has a handle, you’ll
probably be inclined to pull it. If it has a flat metal plate where the handle would be, you’ll be
inclined to push it. The more closely you examine the door, the more directly your sensorimotor
capacities with respect to that or other doors will be brought to bear, on-line (directly engaging
with the door) or off-line (simulating that engagement).
Only if the door has something perceptually un-door-like about it will you be forced to examine it
yet more closely: e.g., if the door has a handle but is meant to be pushed instead of pulled. One
could imagine that the “door” is only a painting on the wall, or has been painted or nailed shut.
Unusual doors will focus your attention and shift it from the abstract and general to the concrete
and immediate, from doors as some platonic-like entities to specific door encounters.
In the process, you may come to change your understanding of doors a little; or you may derive a
concept of a new kind of door. You might re-structure your entire “door” space. Of course, at some
CHAPTER 7. THE CO-EMERGENCE OF CONCEPTS AND EXPERIENCE 161
point, the unusual door in front of you may confound all attempts at conceptual understanding,
and you may resort to brute sensorimotor engagement with it!
7.4 Which Takes Precedence? (Some Final Thoughts on Representations)
The ordering of sections 7.2 and 7.3 was not, as noted at the start of the chapter, purely arbitrary.
Application preceding acquisition seems intuitively as impossible as effect preceding cause or – for
the empiricists at least (of which I admit to being one) – understanding preceding experience.
7.4.1 Locating the Observer
Furthermore, as much as I owe a debt to the phenomenologists, my background is rather from
analytic philosophy, which strives to be a scientific discipline. As noted in Section 5.2.1.1, empirical science has accomplished a great deal by pushing the observer into the background or
ignoring the observer’s role altogether. The story I have told of concept acquisition is the one
that has likewise backgrounded the observer; while the story of concept acquisition has brought
the observer back into the foreground. So in telling the acquisition story first, I am exposing my
analytic roots. At the same time, recall the conclusion I made in the final section to Chapter Five:
. . . Sciences of concepts and of consciousness remind us of what we should have remembered all along: that the observer is always there. . . that the subjective is inseparably
bound up with the objective, that science yields up not timeless understandings freed
from cultural and historical contexts but working hypotheses.
Figure 7.11: Necker cube: Does it extend from or recede into the page?
Often the backgrounding of the observer is a good strategy, perhaps even the right strategy – but
not always. Sometimes, the order of things needs to be turned around. Sometimes we require
a contrasting perspective; and sometimes our perspective inverts of its own accord, as when we
view a Necker cube (see Figure 7.11), or when we look at the famous illustration by W.E. Hill of
CHAPTER 7. THE CO-EMERGENCE OF CONCEPTS AND EXPERIENCE 162
Figure 7.12: Hill’s young woman / old woman, circa 1915: it is possible to see one or the other, but not both at
the same time. (Photo downloaded from Wikimedia Commons: http://commons.wikimedia.org/.)
the young woman / old woman (see Figure 7.12). When we are discussing representations, the
observer is (or should be, on my account) very strongly in the foreground.
7.4.2 The Wider Debate Over Cognition
Earlier I noted that, in my concept acquisition story, concepts looked more like (non-representational)
abilities; while in my concept application story, they looked more like (what some insist on calling
mental) representations. I introduced the debate between concepts as abilities and concepts as
representations in Chapter Two. That debate found its clearest expression in the table toward
the end of Chapter Five, where I suggested that most contemporary theories of concepts, and
most debates about them, line up on one side or the other of the divide. The idea of these two
competing, never-truly-resolvable perspectives has been a running theme throughout this work.
It is time to put the debate between concepts-as-abilities and concepts-as-representations into
the context of a wider debate over the nature of cognition, and in doing so bring it to a close.
That wider debate has been lurking in the background through all the preceding chapters. It
is one that is longstanding, often bitter, and so far unresolved; taking many forms and couched
in various terms: pitting representationalists against anti-representationalists, cognitivists against
connectionists, symbolists against associationists, and rationalists against empiricists (cf . (Brooks,
1991a,b; Chalmers, 1990; Fodor and Pylyshyn, 1988; Perry, 1986)). One way to frame the debate
is this: is mental content of any sort best understood in terms of symbols or symbol manipulation
(where the brain is thought of as e.g. a physical instantiation of an abstract Turing machine), or
in terms of explicitly non-representational associations between various observed regularities (on
CHAPTER 7. THE CO-EMERGENCE OF CONCEPTS AND EXPERIENCE 163
different levels of concreteness and abstractness)?
Of course, those on the one side of the aisle do not deny that associations play a role. But the
meaning is in the things being associated, not in the associations; and context is deemed, at least
most of the time, largely irrelevant. Likewise, those on the other side of the aisle do not deny
that we employ symbols, nor (necessarily) insist that “representation” is an empty term; but for
them, the perspective is turned around: meaning is in the associations, not in the things being
associated. Context is key, and local details of structure are largely irrelevant. Push either side
hard enough, and some ground will be ceded to the other; but on both sides, the tendency is to
think that one side or the other must be right.
7.4.3 The Death of Representations
If I may borrow a line from Mark Twain, the death of representations has been greatly exaggerated.
Concepts are, indeed, more than representations, logically; but, at the same time, representations
are unavoidably, for us, part of what it is for something to be a concept: for when an agent reflects
upon some concept (or on concepts in general), or otherwise employs a concept reflectively, she is using it to represent something to someone, be it herself or another agent. Representations, properly
understood, are both necessary for understanding cognition, even at the most basic sensorimotor
levels, due to our inability as reflective conceptual agents to step aside from our representational
perspective; and not sufficient for understanding even the most abstract levels of symbolically
structured thought, given the logically ubiquitous involvement of (non-representational) sensorimotor engagements throughout cognition. The representations need not, in any sense, be in the
mind of the agent being observed, but rather are bound to the perspective of the agent doing the
observation (who nonetheless may project it into the mind of the agent observed!).
At the same time, representations are not just some awkward necessity; they have real value. In
responding to Rodney Brooks (1991b), David Kirsh talks of concepts, but it is representations, I
believe, that are his ultimate target:
There is a limit. . . to how far a creature without concepts can go. . . . Concepts are
either necessary for certain types of perception, learning, and control, or they make
those processes computationally simpler. Once a creature has concepts its capacities
are vastly multiplied (1991, p. 191).
What representations gain us, argues Richard Shusterman in response to anti-representationalist
Merleau-Ponty, is the capacity for explicit reflection and the consequent ability to recognize and
modify habits, including bad ones:
. . . In order to effect. . . improvement, the unreflective action or habit must be brought
into conscious critical reflection (if only for a limited time) so that it can be grasped
and worked on more precisely (2008, p. 63).
He could as well be responding to Rodney Brooks when he says, “the claim that we can do something
effectively without explicit or representational consciousness does not imply that we cannot also
do it with such consciousness and that such consciousness cannot improve our performance” (2008,
p. 68).
CHAPTER 7. THE CO-EMERGENCE OF CONCEPTS AND EXPERIENCE 164
So which takes precedence: concepts or experience, acquisition or application, reason or empirical
discovery, action or thought? It all depends on where you stand, and how you look. Partly it is a
matter of which questions you are asking, partly on the context in which you intend to apply the
answers; but the tension and the oscillation between competing views is, I believe – in contrast
to a similar oscillation McDowell addresses and believes himself to resolve in Mind and World
(1996) – eternal11.
7.5 Conclusions
How do concepts (as abstract and structured entities) and experience (seemingly immediate, and
largely unstructured) inter-relate? That has been the focusing question of this chapter, as I have
sought to move the discussion out of the largely static view of concepts from the earlier chapters
toward a dynamic view of continuous acquisition and application. In the process I have borrowed
a page from the classical definitionists, for whom definitions were neutral as to whether they were
defining new concepts or verifying existing ones. Indeed, the final (though still very provisional!)
definition of concepts that I offer owes quite a lot to those early traditions, even while the spirit
of my approach often sides with the prototype and similarity-space-based accounts, which derive
more from the imagist traditions.
Concepts do have something peculiarly definition-like about them, once the requirements that
definitions be strictly static and strictly public entities are relaxed. The appropriate metaphor here
might be that of a “dynamic dictionary”, where the words on the page are constantly in motion:
look at a definition, look away, look again, and the definition has subtly changed. Offering a
definition in the more usual, dictionary-type sense of the word becomes an intellectualized attempt
to fix the concept, to take a snapshot. Something of what the concept is gets captured (hopefully),
but something more gets lost.
Concepts without the capacity for motion, without the capacity for change, are dead. The same
might be said of any theory about them.
The other primary motive of this chapter has been to put the ideas about conceptual spaces and
the unified conceptual space to practical work. I showed how, in my account of concept acquisition
through ontogenetic development, an initially unstructured conceptual space becomes increasingly
partitioned and topologically complex. At the same time as proximal points in the unified space
take on increasing significance, so do distal points, as parts of that space are mapped onto each
other.
Turning the acquisition process on its head, concepts become the expectations that drive experience.
Conceptual space is mapped back onto the perceptual space from which it has been derived, layer by
layer, until matching succeeds or breakdown occurs. Breakdown can prompt a variety of responses,
from the conservative (tweaking the concept’s logical structure) to the radical (re-partitioning a
particular conceptual space).
11McDowell’s oscillation is between a kind of anti-realist coherentism – beliefs are justified because they all support
and are consistent with each other – and justificatory appeals to a non-conceptual Given (what he terms the Myth
of the Given). See in particular Lecture One, pp. 3-23. My own approach in this work has been to deny, with
McDowell, that myth; while also denying that either experience or world are fully conceptualized.
CHAPTER 7. THE CO-EMERGENCE OF CONCEPTS AND EXPERIENCE 165
Central concepts in this chapter have been:
• Dynamically coupled systems, which are, logically speaking, impossible to separate, even
while it may be conceptually useful or even necessary to (attempt to) do so.
• Circular causality, by which effects can be seen, from some perspective, to be their own
causes, like the dragon eating its own tail (Figure 7.2).
• Co-emergence, a more conceptually abstract way of talking about certain dynamically coupled
systems, whereby two opposite or opposing forces can be seen each to give rise to the other,
like yin and yang.
• Sensorimotor engagement, whereby the most abstract levels of cognition are grounded in basic
agent-environment interactions, in which the sensory and motor systems are dynamically
coupled, while sensory experience and motor action are co-emergent.
The chapter’s principle conclusion is that concepts and experience are co-emergent within a causally
circular loop of sensorimotor-grounded conceptual acquisition and conceptual-expectation-driven
application, where neither has priority over the other.
As a coda, I tied the debate over concepts-as-abilities versus concepts-as-representations back into
a wider, often acrimonious debate over the nature of cognition: is cognition itself to be understood
primarily in representational or non-representational terms? The “correct” answer depends, once
again, on your vantage point.

KM Frontiers: Self signifying knowledge

Susan Sontag was asking for a new language to describe the forms of communications flow. Unlike entrained thought which closes off possibilities and questions in its quest for truth through interpretation, the magic of language is that it can describe and classify instances, while at the same time being open so that it can describe new unthought of possibilities. Even highly simplified, artificial languages, such as computer programming languages, can create vast numbers of applications that were not even conceived by the language creators.
So self signifying knowledge would require new artificial languages which describe the form of the content, rather than the content, itself.

...

According to its founder, Dave Snowden, Cognitive Edge conceives of language, not as a deep structure, nor a mental model, nor even a semantic network, but as another example of a co-evolving, complex adaptive system. People use language to make up and tell each other narratives of their experience using a wide variety of media. For Cognitive Edge, these narratives, or stories, along with how the tellers feel about them, are the source of the self signifying indicators about a group’s thinking

Subcategories

Pages

Pages in Other Languages

Categories

• Grounded Theory Applications