добавить свой файл
1 2 3 ... 5 6

Metaphors for mind

As in most sciences the cognitive sciences aim for a theory that unifies its field, from neuro-psychology to everyday mental experience. That ambition is shared with the scholars of antiquity, but we are more aware of the complexity of what we are taking on. Contemporary cognitive scientists deal with this by adopting multiple (partial) theories, metaphors and images to explain different aspects of cognition. This strategy is common in the life sciences. Take the case of "how cells work". Plant and animal cells are understood using many distinct metaphors. These include: the static "architecture" of the cell which is built up from specialised sub-structures and organelles; the cell as a dynamic "signalling network" in which lots of little chemical messages whiz from point to point causing things to happen at different places in the cell; the cell as a "bag of chemicals" whose concentrations vary dynamically over time according to quantitative chemical laws, or even as computational systems in which DNA sequences are "programs" for building proteins or controlling the cycle of cell growth and division.

Cognitive scientists trying to understand "how the mind works" have adopted a similar strategy. Unfortunately this is leading to fragmentation of the subject, with different communities who work on different aspects of the problem using different languages and concepts, and frequently arriving at positions of mutual incomprehension.

In the next few sections I review some of the theoretical frameworks that have emerged in the last half-century, starting with psychology and progressively introducing ideas from Artificial Intelligence and, most recently, formal theories of knowledge and mental states. I shall start with the work of two famous scientists, Donald Broadbent and Alan Newell, who provide a useful springboard for the discussion. This choice is somewhat arbitrary, in that I wish to remember two of my personal mentors, but they are also important and representative figures in complementary fields of cognitive science. At a time when ideas from neuroscience and from computational science are becoming increasingly successful, but also moving apart, it may be useful to remember them and the intellectual continuity they represent.

Statics: minds as information processing architectures

"A nervous systems acts to some extent as a single communication channel, so that it is meaningful to regard it as having a limited capacity. ... incoming information may be held in a temporary store at a stage previous to the limited capacity channel:.... The maximum time of storage possible in this way is of the order of seconds."

D E. Broadbent Perception and Communication (1958) 297-298.

Donald Broadbent (1926-1993) was director of the Medical Research Council's Applied Psychology Unit in Cambridge, England. He made many contributions concerning the organisation and integration of human cognitive functions, like memory, attention and decision-making, drawing on experimental investigations and detailed observations of the complex demands that real-world environments place on human beings. His applied focus gave him a keen awareness that cognitive functions are strongly affected by environmental factors like high noise and other stressors and their interaction with physiological parameters. His engineering training also allowed him to see how the information theory that developed during and after the global conflict of 1939-45 could shed light on these processes. The centrepiece of Perception and Communication (1958), the book that established his reputation, visualised functions like memory, attention and decision-making in an image of the mind as an information processing system made up of a number of interacting physical components (figure 2).

“Box and arrow” diagrams of the sort that Broadbent used have continued to be a popular way of providing an image of mental processes in psychology because, under fairly simple interpretation rules, the organisation of the components often seems to yield unambiguous, and experimentally testable, predictions about human behaviour. Box and arrow conventions can also be used at different scales, from the detailed processes involved in recognising words to the high-level

organisation of cognitive processes thought to be involved in human consciousness (figure 3).

Figure 2. Broadbent's model of selective attention

(reproduction of figure 7 in Perception and Communication)

Such schematic diagrams are routinely used to represent and design complex systems in engineering, but in engineering we often already understand how the bits work, and may even have constructed functioning components. Furthermore, designers use standard engineering conventions to convey information unambiguously, which do not really exist in psychology. In cognitive science, in contrast, the purpose of the diagrams is to represent systems that we don’t understand in any detail, and cannot even be confident of the basic components. Unfortunately despite their simplicity such diagrams are open to the criticism that they are under-specified and therefore we may have limited confidence in predicting the detailed behaviour of a hypothetical cognitive system based purely on informal images of this kind.

Figure 3. Box and arrow models of a detailed information processing model of how single words are processed and recognized in the human auditory system (Morton and Paterson 1980), and stages of four subsystems or processes hypothesised to be involved in human “awareness” (Shallice, 1988, page 402): reproduced in (Shallice, 1988 page 92).

Richard Cooper of Birkbeck College London has observed that it would be valuable to have standard conventions for modelling cognitive "modules", their properties and their interrelationships. His suggestion has been put to practical use in the COGENT cognitive modelling system illustrated in figure 4 (Cooper and Fox, 1998; Cooper, 2002).

Figure 4. Broadbent's model of selective attention modelled in COGENT. The modeller sketches components of each model compartment using standard components, “exploding” these where needed to define internal structure (see text).

Figure 4 shows a different image of Broadbent's (1958) theory of selective attention using the COGENT modelling system. COGENT provides a set of standard “cognitive components” for modelling information processing systems, including various types of processing unit, memory buffers and knowledge bases, communication lines and learning mechanisms. The left panel shows the complete model, representing the cognitive system and its communication with the “task environment”. The cognitive system is shown in an exploded view in the centre panel. It is modelled here as three sub-systems, one of which is shown in the exploded view on the right. The perceptual subsystem is made up of two “processing modules” (shown as hexagons) that are able to operate on information obtained from other parts of the model. In this model the processing modules can retrieve information from and send information to the short-term store that was an explicit element of Broadbent’s model. In the COGENT system processes are viewed as stateless, they do not store information so memory systems and their properties must be modelled separately. (COGENT supports a range of different types of memory, shown as rectangles with round ends.)

COGENT is a flexible tool for describing the static organisation of cognitive systems, and has been used extensively for constructing models of natural systems and designing artificial ones (see Cooper 2000 and also Glasspool’s chapter in this volume). It provides a standard way of visualising cognitive components, a range of predefined “cognitive modules” and a practical visual design environment. Cell biologists use generally accepted conventions for visualising static aspects of cellular processes for explanation and teaching, such as 3D molecular models and metabolic pathways drawn as flow graphs. Cognitive scientists have similar needs but often use diagrams as a basis for predicting behaviour as well. COGENT offers clear conventions for describing static cognitive organisation graphically and, as we see in the next section, a precise and unambiguous language for specifying the properties of the system and hence predicting its behaviour.

<< предыдущая страница   следующая страница >>