Predictive animal behaviour
Finally, after cell biology, taxonomy, evolutionary development and comparative anatomy I have reached a section in the crash course I am following that I think I can have an educated perspective on: animal behaviour.
The crash course is…a crash course and intended as a very brief and speedy introduction to a vast array of deep and complex subjects, so what follows is by no means of criticism of the course content.
It still might be true though that the way animal behaviour is introduced and taught to most students, is still very much of the ‘stimulus and response’ variety. Of course crudely this is true. However, coming from a background in cognitive science it is clear that the perception/action cycle, or system, is very different to how it was originally conceived by Pavlov, Skinner and other early behaviourists. The way it all works in humans appears to rest firmly on the notion of prediction: your brain (or perhaps rather, mind) is a prediction machine.
It seems likely that this would not be a human specific cognitive trait, and would at least be shared by other mammals — and depending on how deeply rooted predictive processing is in our neural biology, or in how successful an algorithm it is, and thus how likely it might be to have evolved multiple times — as well as by other non-mammalian species too.
So in fact animal behaviour is a complex system, or process, in which internally generated predictions about both the external world (available to an animal brain and nervous system by way of raw sense data) and the internal world (e.g. interoceptive and proprioceptive sensations) are constantly being met with either confirmatory or disconfirmatory experiences, and in which the animal’s own behaviours are both simultaneously seeking to confirm predictions, and providing input data for future predictions. That might be somewhat difficult to parse.
By way of an example, consider a rodent that ambushes an insect for dinner. In the traditional model of stimulus and response, the rodent lays in wait until a particular pattern of let’s say visual stimulus plays across its retinas (signifying the prey being desirable and in range of attack), and is processed by its brain. This then results in a partly learned, partly evolved behaviour of ambushing arising, and the rodent is rewarded with a tasty dinner, and the whole behaviour is reinforced. Of course this is as I said before, crudely true, but in fact not entirely correct.
If the predictive model is correct, then a more accurate description would be that the way in which the rodent lays in wait is already contributing to the generation of predictions in its brain/mind. When the insect starts contributing sense data to the rodent, the predictions in the rodent’s brain/mind are being both confirmed and disconfirmed in myriad ways (these predictions will be about every minor detail, such as edges, colours, shadows, movement etc.). Simultaneously and as part of the same process, the rodent is actively making changes to itself, by (e.g.) moving its eyes, changing the position of its body and altering the tension in its muscles and soft tissues. These changes are done to minimise prediction error — this means they are done to constantly attempt to make the incoming sense data fit the current predictions. Don’t forget also that in some ways all the data (both external from the senes and internal from the muscle fibres etc.) is homogenous. Even as the rodent starts its ambush, this whole process is continuing, of predictions being made and updated in the face of confirmation and error, and the rodent altering its behaviour to try and reduce error in its predictions.
It’s like a batter hitting a baseball — if the batter saw the ball leave the pitcher’s hand, then next the batter’s brain did some calculation, then next the batter swung, it would never work out, there just isn’t enough time. Instead the batter’s brain is already making predictions before the ball even leaves the pitcher’s hands, and the batter is making micro adjustments to her whole system (body, eyes etc.) throughout the whole swing, to minimise the errors in her brain’s predictions. The predictions are fluid, the stimuli are fluid and the actions are fluid, and these elements are all constantly influencing each other from beginning to end. If you still think there is a microsecond of stimulus that comes first, well, consider why it is that the rodent or the batter are where they are in space and time to be able to receive such a stimulus? Because their behaviour was in a sense trying to receive that stimulus, in an attempt to minimise error on higher level predictions about what should be done to satiate hunger, or earn a paycheque or whatever.
The predictions are hierarchical, or tiered, ranging from the lowest level of basic sense data all the way up to ones serving organisms’ higher order needs to survive and reproduce. A very interesting question to ponder is — how does the first prediction get off the ground? What comes first? But these are questions for another time.
