Rose: Consciousness 
Updates
FiguresRose: Consciousness
Chapter 05
An important and interesting attempt to plot the evolution of representations is given in Millikan’s (2004) recent book. She suggests there was originally only one type of representation, which had combined descriptive and directive functions (the ‘pushmi-pullyu’; for more on these see the web update to Chapter 11). Just as a bee’s dance describes where there is nectar and at the same time tells other bees where to go, so the earliest and most primitive mental representations also subserved these two aspects of function. Thus, beginning with internal biological signals such as blood pressure, Millikan reasons that reflexes, instincts, tastes and pains developed that also carry this functional duality. These lead to what Gibson (1979) called ‘affordances’ for action, which are perceived at the same time as they guide behaviour.
Next, however, Millikan argues that single-type representations needed to appear which were uniquely either descriptive or directive. The older dual-function representations can only subserve behaviour which is egocentric and inflexible and the organism cannot predict the outcomes of future courses of action. Purely descriptive representations of facts however can be recombined adaptively to generate new behaviours in new contexts, and are also necessary for dealing with things and events in the world that are not centred on our selves. In addition, purely directive representations are needed to represent future goal states independently of the current environment. Both types can also be stored as memories and retrieved, whereas the older type deals only with the current environmental situation. Both need to use the same code or language, so they can be compared to tell when or whether a goal has been reached. (Their structural difference comes in the uses to which they are put within the global system.) They may include explicit models of the self, whereas the primitive type has the self built-in as an implicit aspect of every representation, giving them their me/now-centred focus.
Millikan proposes mechanisms by which the new representations become separated off from the original type. The complex structure of an evolved pushmi-pullyu representation needs first to be disassembled into its parts. These will include low-level feature detectors that can respond across all the objects and situations in which those features occur (generalisation). These can then be recombined to represent objects which have continuity across time and space (and across the sensory modality of the detectors) and to represent the uses (affordances) of those objects. For vision, this occurs in what becomes the ventral stream of processing. Meanwhile, the dorsal stream retains its combined representation of the immediate external object and the organism’s body to guide the action relationship between the self and the world. This separation allows the independent development not only of recombinable representations of the partial features of object affordances (ventral) but also of skill fragments that can be recombined to create new action plans and schemes (2004, p. 180).
The genesis of a separate type of representation for goal states (directives) is however merely asserted rather than explained (ch. 16). Ontogenetically, it is proposed that through trial and error learning outcomes might be linked to initial pushmi-pullyus to form paths of possible action. The combination of fragments of known paths might lead to ‘insight’ as to how to reach a goal; but Millikan refuses to speculate on the actual mechanisms of such creativity (p. 208). Nevertheless it is noteworthy that, like other types of complex representation, plans too can have hierarchical structure (cf. Consciousness, Side-box 8.1) so can engage in processes of generativity by virtue of their compositionality (Millikan, 2004, ch. 4; Consciousness, section 5.3.4).
Throughout her book, Millikan has much to say on the distality of intentionality (Consciousness, section 5.3.8). Here, she says that in different cases either distal or proximal stimuli (or anything in between) may be the proper target of a representation (pp. 54-58, 81-83, 161-163, 199).1 Millikan adopts the terminology of Gibson in developing a concept of ‘direct perception’ for cases where a distal meaning is the one perceived (ch. 9). But beware, hers differs from Gibson’s original usage in several ways. She rejects ‘indirect’ perception theories on the grounds that ‘logical inference’ is not as a good a metaphor as ‘translation’ or ‘mapping’ for how representations or signs are interpreted.2 The same principle applies generally to signs inside and outside the body.
However, her model of perception implicitly assumes that such processing takes place between serial stages. Although combinations are allowed at each stage, no recurrent processing is considered. But we could argue that top-down knowledge and stored templates (Consciousness, Chapter 9) exist and indeed provide the premises for inference, despite Millikan’s denial of this mechanism (she regards a lack of premises as disproving the use of inference in perception: 2004, p. 119).3
Unlike most works on the philosophy of mind, Millikan’s benefits from a knowledge of ethology and is replete with concrete examples of animal psychology to illustrate how the mind could have arisen in the natural world. Its significance is also discussed in the web update to Chapter 11.
Section 5.3.2.1 Causation
A collection of papers edited by Baltes et al. (2006) strongly supports the existence of continuous interaction between biological and cultural/environmental factors in driving the dynamic development of both the individual person and their surroundings. Simple reductionistic views of causation such as neural determinism are repudiated in favour of a picture in which biology and culture co-evolve and construct each other via their mutual reciprocal modification and shaping. These workers also recognise the existence of multi-level organization with the arising of new emergent properties including intentionality.
New references
Baltes, P., Reuter-Lorenz, P. and Rosler, F. (2006) eds. Lifespan Development and the Brain: The perspective of biocultural co-constructivism. Cambridge University Press, Cambridge.
Howson, C. (2000) Evidence and confirmation. In Newton-Smith, W.H. (ed.) A Companion to the Philosophy of Science, Blackwell, Oxford, pp. 108-116.
Kersten, D., Mamassian, P. and Yuille, A. (2004) Object perception as Bayesian inference. Annual Review of Psychology 55, 271-304.
Knill, D.C., Friedman, W.T. and Geisler, W.S. (2003) eds. Bayesian and statistical approaches to vision. Journal of the Optical Society of America A 20, 1231-1448.
Footnotes
1. See Jacob and Jeannerod (2003, pp. 7-8) for further examples in which it is the proximal sensory stimulus that is represented.
2. Note that in Consciousness, section 2.5 (e.g. Figure 2.4) I gave together examples of ‘computation’ that involved inferences from premises, the application of rules and operators, mapping and filtering. It is by no means clear that ‘computation’ describes only one of these processes — especially at intermediate levels of description. Recall also Helmholtz’s (1866) notion of ‘unconscious inference’ (the origin of modern theories of ‘indirect perception’), by which he meant something that operates via processes akin to associative linking. Yet is association to a hypothesis or conclusion a matter of mapping or of logically inferring? In his day the number of known mechanisms was limited, but today we have more options as to how ‘associations’ could work. Until we have more evidence, terms such as ‘computation’, ‘inference’ and ‘association’ should remain neutral as to the mechanism.
3. For example there is a well-established body of empirical and theoretical support for the idea that recurrent and/or lateral connections in the brain introduce Bayesian priors into perceptual processes (i.e. stored knowledge about the world; for reviews see Knill et al., 2003; Kersten et al., 2004). On this view, ‘Bayesian inference’ is carried out on premises which are probabilistic rather than binary variables; but nevertheless logical inference is not restricted to true/false premises (e.g. Howson, 2000).
View print version