Initially I was told and thought that the computations in the brain are done by the networks of neurons. Reading the O’Reilly and Munakata book Explorations in Cognitive Neuroscience this is achieved by imagining neurons as points (nodes) connected to each other. The connections are simple directed weights, can have positive or negative values and can change. They determine the amount of signal being passed from one node to another. Already such setup can perform complex computations. Good, this works for sure.
Biology is bit more complex, though. Neurons are not points, they differ greatly one from another, there are several types, … Commonly though they all have dendrites. Mostly they have receptors catching the signals from other neurons (this is always the case for interneurons). And they are a complex network themselves! They have several branches, which normally all end up in the soma (neuron body). Based on what rolls down to the soma, the neuron either fires an action potential and sends the signal forward to the next set of neurons.
Now we finally come to the question. Are those dendrites simply collecting the signals coming from other neurons and summing it all up in soma (being integrators), or they actually do some computations on their own? To get deeper into that, check the review by London and Häuser called Dendritic computation (not too hard to guess what they think about it, huh?). In this way, the dendrites could somehow be the neurons inside the neurons and would thus make the whole network extra complex and extra capable of fast computations. But at the same time, complexity creates vulnerability, so we need to make sure the system stays robust.
Anyway, here is the question, let’s get at it!