Neural Net Connectionism

This movement has been fortified by computer scientists' efforts to mimic the brain by constructing artificial intelligence systems. Based on approximation of cortical neurons as linear threshold units, a large number of "neural net" models have been constructed and simulated on computers. A key concept in relating neural network dynamics to the facts of psychology is the cell assembly introduced by Donald Hebb in 1949. In his view, cell assemblies are specially organized reverberatory circuits that constitute elements of thought. An individual neuron may participate in many of them just as an individual member of society participates in many social assemblies. By allowing strengthening or reinforcement of repeatedly used connections ("synaptic plasticity"), recognition, learning and problem solving become manifest in lowered thresholds of specific loops. By assigning energy levels to various patterns ("landscapes") within the net, mathematical solutions can also be imposed. The "intelligence" or capabilities of a given neural net model depends on the richness of its interconnections and nonlinear feedback. Neural net connectionist theory may help to advance robotic and computer systems for artificial intelligence, and may provide significant insight into brain function. These theories have provided evidence that dynamic activities within a given network can at least mimic some aspects of brain activities.

Shortcomings of early neural net models are that they have been based on hypothetical neurons with huge assumptions about neural function. Each neuron has been considered an on/off gate or switch, and interneuron synapses viewed as variable weight interconnections. More recent models incorporate axonal impulses, synaptic delays, dendritic analog functions and spatial coherence. In their most elegant form, neural net theories provide possible representations of mental objects ("consciousness"?) in the transient instantaneous patterns of network activity. "Temporally stable cooperative coupling" among sets of neurons are suggested to manifest thoughts and images by the work of Hebb, Kohonen, Edelman, Thom, von der Malsburg, Hopfield, Pellionisz, Llinas, Changeux, and others. Some of their work suggests the brain forms sets of "prerepresentations" of what is expected from which sensory input induces "selection" of its reality candidate. In this context, Changeux (1985) has defined consciousness as "a kind of global regulatory system dealing with mental objects and computation using those objects." Neural net models and associative memories have significantly advanced understanding of collective neural capabilities. Their essential features (parallel processors with lateral variable connections) may also be operant within neurons in the cytoskeleton.

Figure 2.3: Closer view of neurons with intraneuronal cytoskeleton visible. Dendrites, ascending from below into cell bodies, have numerous dendritic pines. Several synapses are depicted. By Jamie Bowman Hameroff.
0 0

Post a comment