The following question is quite hypothetical, and just to get an idea.

Assume any model of how positive propositions $p$ are actively (opposed to synaptically) represented in a neural network, e.g. "the cat sat on the mat". Presumably, this will be by some kind of firing of some neurons or neuron assemblies.

I wonder how the corresponding negative proposition $\neg p$ might be represented, e.g. "the cat sat not on the mat". Will there be an extra NOT-neuron firing? Will the same neurons fire in NOT-mode? Will the cat sit on a NON-mat? Or completely different. Or are such propositions not represented at all, i.e. we can only "think positively", and only report verbally, that we do or did not think that $p$ (which would be some kind of a meta-report)?

Which proposals are around for the representation of negated propositions?


1 Answer 1


A plausible candidate for that is interneurons subnetwork.

Let's say some entity may be green and cold. It also may be yellow and hot, but not cold and hot. So, we can say that for this type of entities there are 2 independent variables, describing its state. If right now the variable 'color' is set to the value "red" - that means the negation of all other values of that variable (not green, not blue, ... infinity of them). The question is only how to store the construct "variable" in the neural tissue?

One can gather all possible values of the variable and wire them all with dense lateral inhibition circuit. Activation of one neuron inside that ensemble will kill the activations of the others.

And look - we've got a good instrument for detection of mistakes in our model of the world now! If the network has wired some entities' representations by lateral inhibition - it means that the network has stated a hypothesis that these entities (in some context) are just different values of the same driving factor. If in the next moment strong feedforward signal has activated 2 principal neurons inside the circle - the hypothesis was not correct. The network has detected "paradox".

This explanation is plausible because of 3 facts:

  1. There is a famous pinwheel pattern in the visual cortex:

enter image description here

The figure above depicts reconstructed map of different values of the variable "spatial orientation" in the visual cortex. They are arranged in non-random order.

  1. Superimposition of inhibitory neurons onto that map shows that one interneuron can work with different "values" of spatial orientation.


  1. There is a scientific proposition, that the underlying algorithm is the same in all parts of the cortex (i.e. differences are more likely in hyperparameters, than in essence). We don't know the algorithm (and I don't pretend that my answer is correct), but scientists do not see dramatical differences in the microbiology of tissue in different cortex regions. So, if some "variable - value" relation appears in the most good studied part of the cortex (primary one) than we, probably, have some ground to expect such tendencies in other parts.

source of pictures - Neurobiology: Turning a corner in vision research,1999,Nature

  • $\begingroup$ How beautiful! And how instructive! Never heard of (resp. seen) this before. Does the scientific proposition you mention in item 3 have a name (I can google for)? $\endgroup$ Nov 22, 2017 at 9:17
  • $\begingroup$ I'm not sure about name, but there was an experiment with a newborn kitten: they destroyed his visual cortex (before synaptic prunning has wired it sparsely) and it has been shown that kitten's audio cortex has implemented visual functions functions. So the kitten see the world. Ofcourse this focus works only for new born animals. $\endgroup$
    – nanenaro
    May 7, 2018 at 15:03

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.