by Corry Shores
[Search Blog Here. Index-tags are found on the bottom of the left column.]
[Central Entry Directory]
[Emergentism, Entry Directory] [Computation Entry Directory]
[The following is largely citation. Text emphasis is mine.]
Connectionism and Neuronal Emergence
Connectionism
The entire approach depends, then, on introducing the appropriate connections, which is usually done through a rule for the gradual change of connections starting from a fairly arbitrary initial state. The most thoroughly explored learning rule is "Hebb's Rule." In 1949 Donald Hebb suggested that learning could be based in changes in the brain that stem from the degree of correlated activity between neurons: if two neurons tend to be active together, their connection is strengthened; otherwise it is diminished. Therefore, the system's connectivity becomes inseparable from its history of transformation and related to the kind of task defined for the system. Since the real action happens at the level of the connections, the name connectionism (often called neoconnectionism) has been proposed for this direction of research. (87bc)
Let us consider an example. Take a total number (say N) of simple neuronnlike elements and connect them reciprocally. Next present this system with a succession of patterns by treating some of its nodes as sensory ends (a retina if you wish). After each presentation let the system reorganize itself by rearranging its connections following a Hebbian principle, that is, by increasing the links between those neurons that happen to be active together for the item presented. The presentation of an entire list of patterns constitutes the system's learning phase. (87-88)
After the learning phase, when the system is presented again with one of these patterns, it recognizes it, in the sense that it falls into a unique global state or internal configuration that is said to represent the learned item. This recognition is possible provided the number of patterns presented is not larger than a fraction of the total number of participating neurons (about 0.15 N). Furthermore, the system performs a correct recognition even if the pattern is presented with added noise or the system is partially mutilated. (88a)
Emergence and Self-Organization
The strategy ... is to build a cognitive system not by starting with symbols and rules but by starting with simple components that would dynamically connect to each other in dense ways. In this approach, each component operates only in its local environment, so that there is no external agent that, as it were, turns the system's axle. But because of the system's network constitution, there is a global cooperation that spontaneously emerges when the states of all participating "neurons" reach a mutually satisfactory state. In such a system, then, there is no need for a central processing unit to guide the entire operation. The passage from local rules to global coherence is the heart of what used to be called self-organization during the cybernetic years. Today people prefer to speak of emergent or global properties, network dynamics, nonlinear networks, complex systems, or even synergetics. (88b-c)
Neuronal emergence
Recent work has produced some detailed evidence that emergent properties are fundamental to the operation of the brain itself. (93b)
Information-processing metaphors are, however, of limited use. For example, although neurons in the visual cortex do have distinct responses to specific features of the visual stimuli, these responses occur only in an anesthetized animal with a highly simplified internal and external environment. When more normal sensory surroundings are allowed and the animal is studied awake and behaving, it has become increasingly clear that stereotyped neuronal responses become highly context sensitive. There are, for example, distinct effects produced by bodily tilt or auditory stimulation. Furthermore, the neuronal response characteristics depend directly on neurons localized far from their receptive fields. Even a change in posture, while preserving the same identical sensorial stimulation, alters the neuronal responses in the primary visual cortex, demonstrating that even the seemingly remote motorium is in resonance with the sensorium. A symbolic, stage-by-stage description for a system with this type of constitution seems to go against the grain. (93-94)
It has, therefore, become increasingly clear to neuroscientists that one needs to study neurons as members of large ensembles that are constantly disappearing and arising through their cooperative interactions and in which every neuron has multiple and changing responses in a context-dependent manner. A rule for the constitution of the brain is that if a region (nucleus, layer) A connects to B, then B connects reciprocally back to A. This law of reciprocity has only two or three minor exceptions. The brain is thus a highly cooperative system: the dense interconnections among its components entail that eventually everything going on will be a function of what all the components are doing. (94a-b)
This kind of cooperativeness holds both locally and globally: it functions within subsystems of the brain and at the level of the connections among those subsystems. One can take the entire brain and divide it into subsections, depending on the kinds of cells and areas, such as the thalamus, hypocampus, cortical gyri, etc. These subsections are made up of complex networks of cells, but they also relate to each other in a network fashion. As a result the entire system acquires an internal coherence in intricate patterns, even if we cannot say exactly how this occurs. For example, if one artificially mobilizes the reticular system, an organism will change behaviorally from, say, being awake to being asleep. This change does not indicate, however, that the reticular system is the controller of wakefulness and sleep. It is the animal that is asleep or awake, not the reticular neurons. In fact, there are many levels of resolution at which such neuronal emergences can be studies, from the level of cellular properties to entire brain regions, each level of detail requiring a different methodology. (94b-d)
Varela et. al. have us consider visual perception in its peripheral stages.
The standard information-processing description (still found in textbooks and popular accounts) is that information enters through the eyes and is relayed sequentially through the thalamus to the cortex where "further processing" is carried out. But if one looks closely at the way the whole system is put together, one finds little to support this view of sequentially. (94-95)
What we find is that the vision processing part of the brain only "listens" to 20 percent of the information that comes from the retina. The rest is obtained from "the dense interconnectedness of other regions of the brain." (95d)
As well, more information flows out of the vision-processing part of the brain than flows in. (95d)
Thus even at the most peripheral end of the visual system, the influences that the brain receives from the eye are met by more activity that flows out from the cortex. The encounter of these two ensembles of neuronal activity is one moment in the emergence of a new coherent configuration, depending on a sort of resonance or active match-mismatch between the sensory activity and the internal setting at the primary cortex. (ft.21. For a detailed examination of this for the case of binocular rivalry see Varela and Singer, Neuronal dynamics in the cortico-thalamic pathway as revealed through binocular rivalry. Experimental Brain Research 66:10-20). The primary visual cortex is, however, but one of the partners in this particular neuronal local circuit at the LGN [lateral geniculate nucleus] level. Other partners, such as the reticular formation, the fibers coming from the superior colliculus, or the corollary discharge of neurons that control eye movements, play an equally active role. (ft22, Singer, W. 1980. Extraretinal influences in the geniculate. Physiology Reviews 57:386-420.) Thus the behavior of the whole system resembles a cocktail party conversation much more than a chain of command. (96a)
What we have described for the LGN and vision is, of course, a uniform principle throughout the brain.
An individual neuron participates in many such global patterns and bears little significance when taken individually. In this sense, the basic mechanism of recognition of a visual object or a visual attribute could be said to be the emergence of a global state among resonating neuronal ensembles. (96b)
Varela, Francisco J, Evan Thompson, & Eleanor Rosch. The Embodied Mind: Cognitive Science and Human Experience. Cambridge, Massachusetts: The MIT Press, 1991.
