You’re looking for “connectionists”. They don’t really exist as a type of scientist / advocate anymore as that dates back to the early 1990s. AI / neural networking became a dirty word in the 1990s and they started referring to it by other names in order to get funding. There was a long AI winter from around 1991 – 2000 or so. Instead you saw the words “Machine Learning” as one of many replacement terms. They knowingly used an outdated model of the neuron because it was computationally easy to follow but they knew at least at the time that it wasn’t a mimic of the brain. Where you see that kind of fluff is in marketing “brain on a chip” from IBM an stuff – and unfortunately, some machine learning researchers actually believe they’re modeling the brain, although I think most of that is what’s in the science advocacy press to keep people enthusiastic about science and to sell books on THE SINGULARITY and stuff. CogSci has been well aware that the neuron is far more complicated than the leaky integrate-and-fire model used in neural networking; it’s just for many purposes, when working with the electric model of the brain, it remains a useful metaphor and easy to work with. But generally I think the current modeling is focused on chemical gating including the behavior of microtubules, white matter, etc.

5.4 Relation to Biological Neural Networks

Connectionist networks are often called ‘neural networks’ and described in terms of (artificial) neurons connected by (artificial) synapses, but is this more than a metaphor? Generally, connectionist models have reflected the contemporary understanding of neurons. For example, McCulloch and Pitts focused on the ‘all or nothing’ character of neuron firing, and modeled neurons as digital logic gates. Newer connectionist models have had a more analog focus, and so the activity level of a unit is often identified with the instantaneous firing rate of a neuron. However, these models still ignore many important properties of real neurons, which may be relevant to neural information processing (Rumelhart et al., 1986′, vol. 2, Chap. 20). As a consequence neuroscientists have stressed the differences between biological neurons and the simple units in connectionist networks; the relation between the two remains an open problem. Nevertheless, it is much easier to envision neural implementations of connectionist networks than of symbol-processing architectures.

See Churchland (1986) and Quinlan (1991) for an introduction to connectionist approaches in philosophy and psychology. See Connectionist Models of Concept Learning; Connectionist Models of Development.

==