Communications

I have been puzzling about how one bit of the brain, one module perhaps, talks to another module. At the level of the cell, the signal is carried on somewhat randomised bursts of action potentials on the axons of the carrier neurons. But how do we turn these thousands of more or less identical action potentials into the sort of information that we talk about?

That is to say, excluding those signals which are command-like rather than data-like: turn this module on, turn that module off. Turn these synapses on, turn those synapses off. I am interested in the signals which carry information about something, perhaps the scene outside the window. The birds and the bees, or in the example above, the employees and their payroll.

We might imagine that the brain has somehow coded that something up, in such a way that the output neurons can be thought of as a one or two dimensional array, in which case the module receiving the signal needs to have wired up its neurons so as to match those of the module sending. Maybe the neurons are implementing, in some more or less direct way, so many neurons for every character, the sort of data description statements from a computer program which are illustrated above. Maybe we are talking about hundreds of neurons here, maybe even thousands.

If a BT electrician were making a connection of this sort in a junction box, he might take the first wire from the sender and attach it to the first point on the receiver. Then take the second wire and attach it to the second point. Or, more tediously, he might get his mate to fire a signal down his end of the first wire. First electrician then checks his his end of every wire not so far accounted for to see which one the signal was on and then, having found it, attach it to the next, in this case, the first point. And so on, through however many wires it was – with the point being that we have matched the output neurons of the sender with the input neurons of the receiver – without which matching the junction box will be pretty much useless. The sweet old lady down the road will be getting red hot porn from the Netherlands and the sweaty old gent across the street will be getting gardeners’ question time. With, in all probability, neither of them being very happy about it.

Which suggests to me that the brain might have a problem here too. I associate to an observation made somewhere by one C. M. Bishop to the effect that the world is far too complicated to be analysed in detail by hand. Having got the general idea, you just have to connect the world to a computer which can learn for itself – perhaps a computer incorporating a network of electrical (rather than organic) neurons – and let it get on with it. Trying to understand, to get a grip on the world in all its detail yourself is not the way forward.

So perhaps the brain needs to be able to use the signal coming down the (say) hundred neurons without needing to bother about which neuron was which. Now if we add up all the contributions from all the neurons, we get what can be thought of as a non-negative real valued function of time. And the more neurons the better the signal. Perhaps, just to be on the safe side the sending module keeps sending the signal, over and over, so that our real valued function is periodic. And once the receiving module detects the periodicity, it can analyse that function and pull out, perhaps, its Fourier coefficients, the real content of the signal.

The point of all this being, that in order to do this, the brain does not have to do what the electrician with the junction box had to do. It just adds everything up without regard to order or anything else. Much simpler. All one needs to evolve is the machinery for doing the analysis – the general purpose machinery which does not need to know about employees and their details.

Another advantage might be that an arrangement of this sort might be a bit noisy, there might well be a bit of noise with the signal, but it will also be robust. There is no single point of failure.

By way of a postscript, I mention the ability of growing bodies to lay things down in orderly sequences, perhaps along chemical or electrical gradients. Things like the segments  of caterpillars or vertebrae. Perhaps arrays of neurons are created in this sort of way. But given that our two arrays of neurons are separated by a bundle of axons, I don’t see how this helps.

With thanks to the Oracle Corporation for the picture at reference 1. Who, I am pleased to say, since I took Window 10, have stopped popping up every day to tell me about their latest update on the Java front.

Reference 1: http://www.oracle.com/technetwork/developer-tools/warehouse/cobol.pdf.