A critique of simplistic assumptions in the explanation of behavior

Jaime E. Alcalá*

e-mail: jealcalat@gmail.com

CEIC, Universidad de Guadalajara
Diego Torres-Marruffo**

*Autor del trabajo

**Autor del dibujo representando a Rick perteneciente a la serie "Rick and Morty" 

The body and the behavior

It has been said that when reinforcement occurs, is the response what is reinforced, not the body or the person. Which response is this? A past response at t1 that is the same response at t2? To make sense of talking about reinforcement as an operation on something, it requires at least operate on something that persists through time. Learning by reinforcement makes the animal behaves at time t2 in a similar way than at time t1 (that’s a basic property, similarity seems to be a definitory characteristic of regularity). Something must persist and the only constant in different times is the body. We must put the body back again in the picture, but not the whole body but only the plastic parts of ir. Thus, how is the body related with reinforcement? Killeen (2013) said the relation might captured by a variant of the Hebbian rule (perhaps Hebbian LTP, but also non-Hebbian and anti-Hebbian forms).

When an axon of cell A is near enough to excite a cell B and repeatedly or persistently takes part in firing it, some growth process or metabolic change takes place in one or both cells such that A's efficiency, as one of the cells firing B, is increased (Hebb, 1949).

The obvious implication is that reinforcement occurs in a body dynamically functioning (the body is not a storage device but a changing system, and of course there’s no environment without stimuli). The events with which the body has contact make changes on it that persist until some other event makes another contact. The synaptic efficiency of the neural circuit related with the event A changes accordingly to some other neural circuit related with event B. The stronger synaptic efficiency at time t (i.e., a circuit activated by a biologically relevant event like a reinforcer) strengthens those circuits (i.e., those activated by a stimulus) firing at time t-1 or even time t or time t+1 (like in backward conditioning). The Hebbian rule makes a probabilistic dependence of the two events and related circuits: as the pairing of those two events repeats more frequently, the efficiency of the less strong resembles to the efficiency of the other until the initially less efficient gain some control over the process controlled by the stronger one.

This resembles the operation of pavlovian conditioning in some way: exteroceptive event CS acquires the function of elicit response CR by the conditional presentation of the biologically relevant event US.

It is somehow strange why some behaviorists still talk about reinforcement as a selection of the response, a response that has occurred in the past (cf., Catania, 2013). Some of them persistently avoid considering biological systems in their explanations of behavior, as if any kind of behavior were multiple realized and so, the biological constraints irrelevant for the psychological functioning (cf., Bechtel & Mundale, 1999).

About the conditional structure of science

Killeen said in his article that science makes use of material implication. He presented two rules of inference used by science: modus ponens and modus tollens. The two, I think, are not of the same value:

 

Often, an explanation is confused with an argument. The big difference is that the starting point in explanation is some uncontroversial fact B, for which scientists offer a sufficient condition A that accounts for B (Fig.1), i.e., A says why B happens. In an argument, A doesn’t explain B, but only says that B is true if A is true.

In the explanations, the truth of A, the antecedent (not necessarily in time), is not a given fact, but as a statement with empirical content, it must be tested. That’s, I think, the most important question in scientific reasoning: how could A be tested? No evidence can establish the truth of A, as the critics of inductivism has often argued, but, deductively, it just can be falsified, i.e., proven to be false. That’s why I said modus tollens is more useful in scientific reasoning: given A, we deduce a set x of possible results that, in principle, must be part of B ( ; if it’s observed that x doesn’t happen (¬x), then A can’t account for B.

That’s the big picture of falsificationism. Details are not so easy. Hypotheses are not simple and isolated conditions, and the mapping from it to the data can be an oversimplification. For example, which is the relation of the initial conditions with the phenomena? Some deterministic model can considerate them just as a reference point from which the process to be explained can evolve to a posterior point, but the conditions are causally irrelevant. This is not always the case.

Consider the following phenomenon in neurobiology, the stochastic resonance. In the traditional communication systems, noise is always the villain whose influence should be reduced, for it have detriment effects on information transmission. A model whose aim is to explain, say, the tactile sensation, may consider only how an impulse on the mechanoreceptors of the skin exceeds certain value of activation and triggers an action potential in the sensory neurons that, ultimately, conveys information to the spinal cord and to the somatosensory cortex. The model may consider how the animal can respond to a single sensory event if it is immersed in a continuous flow of sensory inputs (namely, noise) dynamically changing. So, the intuitive way of thinking is how the sensory system avoids this noise and the signal is correctly detected (i.e., how the signal-to-noise ratio is enhanced), even at very low values for which the sensory system is sensitive (less than 200 μm in tactile perception). One can think that, in the absence or very low values of noise, the signal-to-noise ratio will be optimal. But that’s not what happens. The random noise, indeed, improves signal detection, especially weak signals, a naturally occurring phenomenon (i.e., the noise is not just in the environment, but produced by the biological system) called stochastic resonance (Moss et al., 2004). So, we can reconsider our position: is noise causally relevant to our explanation of tactile sensation?

What about the effects on behavior? It seems a background assumption that, in the description or explanation of behavior, the body it’s considered a static device that responds to the environment in a reliable and regular manner. Behavior analysts test this by manipulating those environmental variables they think affects behavior, devising ways to avoid variability as much as they can but, often, that variability is in their data. For example, Generalized Matching Law can account up to 90 % of the variance in choice behavior (Kyonka & Grace, 2016) but, what about the rest? We can’t exclude the possibility that behavior is affected not just by the environmental variables we manipulate, but also by those that, to us, seems unchanged and, much more interesting, by intrinsic biological variables dynamically changing. Biological systems are not static devices, but dynamical systems and, as the stochastic resonance suggests, it can add to the variance not accounted by our model.

Can we, thus, hold the simplistic assumption that motivates our research? We must consider the material implication much more than a mapping rule from A to B. Contingent events could be causally relevant, our system can be nonlinear and dynamic in nature and, of course, we must put the body back again in our explanations or descriptions (it’s not sufficient to say the body is a necessary condition) for it can be dangerous to think it is causally irrelevant to the behavior.

Bibliography:

Bechtel, W. & Mundale, J. (1999). Multiple Realizability Revisited: Linking Cognitive and Neural States. Philosophy of Science. 66: 175–207

Catania, C.A. (2013). A natural science of behavior. Review of General Psychology. 17:133-139.

Hebb, D.O. (1949). The Organization of Behavior. New York: Wiley & Sons.

Killeen, P.R. (2013). The structure of scientific evolution. Behavior Analyst. 36: 325-344

Moss, F., Ward, L. M. & Sannita, W. G. (2004). Stochastic resonance and sensory information processing: A tutorial and review of application. Clinical Neurophysiology. 115(2): 267-281