The backpropagation algorithm: Implications for the biological bases of individual differences in intelligence

Variations in brain structure and function may explain individual variation in human intelligence. However, it is not currently possible to directly examine this hypothesis. As an indirect examination, a neural network employing the backpropagation algorithm to solve the exclusive-or function was manipulated to possess different numbers of processing elements (neurons), connections (synapses), and conduction failure (synaptic failure). The effect of the variations on network accuracy and energy utilization were compared to human reaction time and cerebral metabolic data to evaluate which variations most reliably reproduced the human results. Varying the synaptic failure rare appears essential for mimicking the human reaction time data and increasing network connectivity is the most efficient way to improve network accuracy for a given degree of neuronal activation. The results suggest that variations in the physiologic events of synaptic neurotransmission and variations in the structural interconnectivity of the neurons in the brain will be found to underlie an important portion of the variation in human intelligence.