O V E R V I E W 59
3.1 Cognitive systems as functional systems 60
3.2 The anatomy of the brain and the primary visual pathway 62 The two visual systems hypothesis:
Ungerleider and Mishkin, “Two cortical visual systems” (1982) 65 3.3 Extending computational modeling to
the brain 70
A new set of algorithms: Rumelhart, McClelland, and the PDP Research
Group, Parallel Distributed Processing: Explorations in the Microstructure of Cognition (1986) 72
Pattern recognition in neural networks: Gorman and Sejnowski’s mine/rock detector 74
3.4 Mapping the stages of lexical processing 76
Functional neuroimaging 77 Petersen et al., “Positron emission
tomographic studies of the cortical anatomy of single-word processing” (1988) 78
Overview
One of the most striking features of contemporary cognitive science, as compared with cognitive science in the 1970s for example, is the fundamental role now played by neuroscience and the study of the brain. This chapter reviews some landmarks in cognitive science’s turn to the brain.
For both theoretical and practical reasons neuroscience was fairly peripheral to cognitive science until the 1980s. We begin insection 3.1by looking at the theoretical reasons. The key idea here is the widely held view that cognitive systems are functional systems. Functional systems have to be analyzed in terms of their function – what they do and how they do it. Many cognitive scientists held (and some continue to hold) that this type of functional analysis should be carried out at a very abstract level, without going at all into the details of the physical machinery that actually performs that function.
This conception of cognitive systems goes hand in hand with a top-down approach to thinking about cognition. Marr’s study of the visual system is a very clear example of this. For Marr, the key to understanding the early visual system is identifying the algorithms by which the visual system solves the basic information-processing task that it confronts – the task of specifying the distribution and basic characteristics of objects in the immediate environment. As we saw, these 59
algorithms are specifiable in abstract information-processing terms that have nothing to do with the brain. The brain enters the picture only at the implementational level.
Insection 3.2, in contrast, we will look at an influential study that approaches vision from a fundamentally different direction. The two visual systems hypothesis, originally proposed by the neuroscientists Leslie Ungerleider and Mortimer Mishkin, draws conclusions about the structure and organization of vision from data about the pathways in the brain that carry visual information. The direction of explanation is bottom-up, rather than top-down.
As in most branches of science, experiment and models are intimately linked in cognitive science. A very important factor in the turn towards the brain was the development of ways of modeling cognitive abilities that seem to reflect certain very general properties of brains. As sketched out insection 3.3, so-called connectionist networks, or artificial neural networks, involve large populations of neuron-like units. Although the individual units are not biologically plausible in any detailed sense, the network as a whole behaves in ways that reflect certain high-level properties of brain functioning.
Moreover, artificial neural networks behave in certain ways rather like real neural networks. Because they can be trained, they can be used to model how cognitive abilities are acquired. And, like human brains, they are not “all-or-nothing” – even when damaged they can continue to perform, albeit in a limited way (unlike digital computers, which function either optimally or not at all).
One reason for cognitive science’s neglect of the brain is that until the 1980s techniques for studying human brains while cognitive tasks were actually being carried out were relatively unsophisticated and not widely known among cognitive scientists. This changed with the emergence of functional neuroimaging in the 1980s. Functional neuroimaging was seen by many as providing a powerful tool for studying what goes on in the brain when subjects are actually performing different types of cognitive task. Insection 3.4we look at an early and very influential application of positron emission tomography (PET) scanning technology to the study of visual word processing. This study shows how functional neuroimaging can be used to generate information-processing models of how cognitive tasks are carried out – information-processing models that are derived, not from abstract task analysis, but rather from detailed study of neural activity.
3.1
Cognitive systems as functional systems
Many cognitive scientists have argued that cognitive processes can be studied independ- ently of their physical realization. Just as we can understand a piece of software without knowing anything about the machine on which it runs, so too (many people have thought) we can understand cognitive processes without knowing anything about the neural machinery that runs them. In fact, for many cognitive scientists the software/ hardware analogy is more than an analogy. It is often taken literally and the mind is viewed as the software that runs on the hardware of the brain. What cognitive scientists are doing, on this view, is a form of reverse engineering. They are looking at the human organism; treating it as a highly complex piece of computing machinery; and trying to
work out the software that the machine is running. Details of neurons, nerve fibers, and so on are no more relevant to this project than details of digital circuitry are relevant to the project of trying to reverse engineer a computer game.
In fact, for many cognitive scientists it is not just that cognitive processes can be studied independently of the neural machinery on which they run. They have to be studied that way. This is because they think of cognitive systems as functional systems. The important point is, as the word suggests, that functional systems are to be under- stood primarily in terms of their function– what they do and how they do it. And, these cognitive scientists emphasize, this type of analysis can be given without going into details about the particular physical structure implementing that function.
An analogy will help. Consider a heart. What makes something a heart? The most important thing is what it does. Hearts are organs that pump blood around the body– in particular, they collect deoxygenated blood and pump it towards the lungs where it becomes reoxygenated. The actual physical structure of the heart is not particularly important. An artificial heart will do the job just as well (although not perhaps for as long) and so still counts as a heart. Crocodiles and humans have hearts with four chambers, while most reptiles have hearts with three chambers. What matters is the job the heart does, not how it does it. A grey whale’s heart is no less a heart than a hummingbird’s heart just because the first beats 9 times per minute while the second beats 1,200 times per minute. One way of putting this is to say that functional systems are multiply realizable. The heart function can be realized by multiple different physical structures.
Exercise 3.1 Give another example of a multiply realizable system.
If cognitive systems are functional systems that are multiply realizable in the way that the heart is multiply realizable, then, the argument goes, it is a mistake to concentrate on the details of how the brain works. In fact, according to cognitive scientists opposed to looking at the brain, focusing on how the brain works is likely to lead to a misleading picture of how cognition works. It might lead us to take as essential to memory, say, things that are really just contingent properties of how our brains have evolved. We would be making the same mistake as if we were to conclude that hearts have to have four chambers because the human heart does, or if we decided that Microsoft Word has to run on a 2.33 GHz Intel Core 2 Duo processor just because that is the processor in my Apple Macintosh.
Exercise 3.2 How convincing do you find this analogy between studying the mind, on the one hand, and studying hearts and computer programs, on the other?
Some of the things that we know about brains actually support this way of thinking about the mind. One of the things neuroscientists have learnt from studying the brain is that it is highly flexible (or, as neuroscientists often say, plastic). Specific areas of the brain and neuronal circuits can change their function, perhaps as a way of dealing with traumatic damage to one part of the brain, or perhaps simply as a result of learning
and other forms of natural rewiring. But this is just another way of saying that certain types of mental activity are multiply realizable – they can be carried out by different neural structures. Similarly, there are many differences between human brains and the brains of non-human animals. But there are also many cognitive abilities that we share with non-human animals – perceptual abilities, for example; certain types of memory; the capacity to feel pain; and the capacity to reason in certain very basic ways. These abilities are multiply realizable. They are not tied to particular types of brain structure.
The theoretical issues in this area have been much debated by philosophers and cognitive scientists. It is fair to say, though, that in the last twenty or so years this way of thinking about cognitive science has become less dominant and the pendulum has swung towards seeing the study of the brain as an integral part of cognitive science. There are many reasons for this change. Some of them have to do with the development of new techniques and machinery for studying the brain. Cognitive scientists have also been influenced by the development of powerful tools for modeling and simulating brain processes. In this chapter we will look at three major events in cognitive science’s turn towards the brain.
3.2
The anatomy of the brain and the primary visual pathway
In order to understand the significance of the two visual systems hypothesis we need a little information about the large-scale anatomy of the brain. Sketching with very broad strokes of the brush, anatomists distinguish three different parts of the mammalian brain – the forebrain, the midbrain, and the hindbrain. This structure is illustrated for the human brain inFigure 3.1
As the figure shows, the forebrain is the largest of the three regions. Most of the forebrain is taken up by the cerebrum (seeFigure 3.2), which is the main portion of the brain and the most important for cognitive and motor processing. The cerebrum is divided into two hemispheres – left and right. The outer layer of each hemisphere comprises what is known as the cerebral cortex (popularly known as “grey matter”). Moving inwards from the outer, cortical layer we find the sub-cortex (the so-called “white matter”). In the human brain the cerebral cortex is about 2–4 mm thick.
Each cerebral hemisphere is divided into four main regions, called lobes. Each lobe is believed to be responsible for carrying out different cognitive tasks.Figure 3.3illustrates the organization of the left hemisphere into four lobes, whileBox 3.1summarizes what each lobe is believed to be specialized for.
There is further organization within each lobe. In 1909 the German neurologist Korbinian Brodmann proposed a mapping of the cerebral cortex into fifty-two areas. These Brodmann areas are still in use today. An example particularly relevant to us now is Brodmann area 17, which is also known as the primary visual cortex, the striate cortex, or area V1. Brodmann area 17 is located in the occipital lobe and (as the name“primary visual cortex” suggests) it is the point of arrival in the cortex for information from the retina.
The information pathway leading from the retina to the primary visual cortex is rela- tively well understood. It is clearly represented inFigure 3.4, which shows how visual infor- mation from each eye is transmitted by the optic nerve to the lateral geniculate nucleus (a sub-cortical area of the forebrain) and thence to the primary visual cortex. The diagram clearly shows the contralateral organization of the brain. Each hemisphere processes infor- mation deriving from the opposite side of space. So, visual information from the right half of the visual field is processed by the left hemisphere (irrespective of which eye it comes from). Much more complicated than the question of how information from the retina gets to the primary visual cortex is the question of what happens to that information when it leaves the primary visual cortex. This is where we come to the two visual systems hypothesis and to the work of Ungerleider and Mishkin.
Forebrain Forebrain Midbrain Midbrain Hindbrain Hindbrain
Figure 3.1 The large-scale anatomy of the brain, showing the forebrain, the midbrain, and the hindbrain.
Cerebrum Corpus callosum Pons Medulla Cerebellum
Figure 3.2 A vertical slice of the human brain, showing the cerebrum.© TISSUEPIX/SCIENCE PHOTO LIBRARY
Central sulcus
Frontal Parietal
Temporal
Lateral sulcus
Occipital