The authors present research on alternative basic elements for neural network modeling. A principle that has emerged from this research, which may have important implications for understanding natural and perhaps artificial intelligence, is examined. A paradigm that deals with what and when may be essential for modeling natural intelligence. The authors call such a paradigm a spatio-temporal neural network paradigm. Such a paradigm, which emphasizes real-time, closed-loop interactions between a learning system and its environment, is emerging from research on alternative models of single neuron function. In particular, it is found that neuronal models of classical conditioning phenomena and neural network models of instrumental conditioning phenomena suggest that, as a general principle, real-time considerations may be fundamental to natural intelligence. More specifically, the authors are investigating the hypothesis that learning in biological systems consists of acquired positive and negative real-time feedback loops built on a foundation of innate positive and negative real-time feedback loops.