I have just read Lakoff & Johnson's “Metaphors we live by” (see this summary), which is a classic brilliant book about the use of metaphor in both language and the way we think. The thesis is that we constantly use metaphors when we speak, especially about abstract concepts, and it actually structures the way we think about those concepts, even though we normally don't even notice we speak in metaphors. Typically we tend to speak of abstract concepts by using metaphors that are more concrete. For example, a theory is a building (it has foundations, it requires support and it can fall apart), understanding is seeing (a clear argument; a point of view; an eye opener). Time is an abstract concept that is very difficult to speak of without using metaphors, for example time as space and duration as distance (an hour long, a point in time, etc).
Metaphors are also widespread in science. Because they structure our thought and often go unnoticed, it is quite interesting to identify them: what ways of thinking do scientific metaphors convey, and other ways are possible? For example, neurons code and neurons compute. These are two quite distinct metaphors. In the coding metaphor, the neuron hears about the world and speaks to other neurons, in the language of electrical impulses. In the computing metaphor, the neuron is a sort of factory that processes inputs. But you could also say that neurons push each other, where axonal terminals are hands and spikes are mechanical energy. This is just as true as the metaphor neurons speak to each other (coding metaphor). Each of these metaphors conveys a different way of seeing the activity of neurons (bonus: check the metaphors in this sentence!).
In this series, I want to explore some of the metaphors in neuroscience. I start with two concepts: “integration” (this post) and “firing” (next post). Those two words are found in the integrate-and-fire model, but clearly they appear throughout the neurophysiological literature. Neuron function is generally described as a two-stage process: synaptic integration followed by firing.
The neuron integrates its synaptic inputs: the neuron is a container, and the inputs come into the container. The inputs are objects, they accumulate within the neuron, so after integration they are in the neuron – possibly in a “processed” form (neuron = factory). So in integration there is a concept of summation: inputs add up. Of course there can be qualifications: the integration could be “nonlinear”. But the very fact that “nonlinear” is a qualification means that the basic concept is that of summing up inputs that come in. What is important to realize here is that it is a metaphor, that is, you could imagine other metaphors to think about the propagation of electricity in neurons. For example, you could think of neurotransmitters as pushing dominoes placed all along the neuron; specific things could happen at branch points depending on when the dominoes get pushed on each branch. In both metaphors we are saying that the output depends on the inputs, but each metaphor emphasizes or hides certain aspects of that relation. For example, in the first metaphor, time is deemphasized and is replaced by counting. In the second metaphor, the notion of “summing up” doesn't even make sense because activity is transient.
Importantly, the two metaphors convey very different models of neural function. The integration metaphor entails a model in which the neuron's membrane is gradually charged up and down by synaptic currents. It deemphasizes space by hiding the notion of electrical propagation; it deemphasizes time by seeing inputs as objects (which can be counted) rather than activity (which is transient). In terms of mathematical models, the integration metaphor corresponds to the “perfect integrator” (i.e., a capacitor plus input currents). Of course there are variations around that model, but the prototype is the integrator. The domino model cannot be understood as a variation of an integrator. The domino metaphor views neural activity as intrinsically transient, and there is a clear relation between the timing of inputs and the timing of outputs. A coincidence detection model, in which an output spike is generated when synchronous inputs arrive at the cell, might fit the domino metaphor better than the integration metaphor.
Thus, to talk about synaptic integration or to say that the neuron integrates inputs is not a neutral statement. It entails a particular way of seeing neural activity, specifically a way that deemphasizes the notion of activity and views inputs and outputs as objects. This metaphor is consistent with other dominant metaphors in neuroscience, in particular the notion of representations, which sees neural activity as objects that can be manipulated (specifically, as pictures). Thus, the integration metaphor refers to the more general metaphor neural activity = object. A weakness of this dominant metaphor is that there is a strong disonance between the notion of activity, which is transient, and the notion of object, which is persistent. This disonance appears with the next metaphor: neurons fire.