It’s time to inhibit Pavlovian conditioning


John Limber

Department of Psychology

University of New Hampshire, Durham NH 03824


Abstract: Despite a promising introduction, the article fails to capitalize on the concept of information intrinsic to control theory. The authors limit their application of feedforward models to simple non-dynamic cases. Their applications to social behavior are stimulus-occasioned responses. Agents might as well be dogfood! The notion of "conditioning" is generalized without warrant to explain virtually any acquired predictive capability.


On first glance, this appeared to be an interesting and potentially significant paper. The authors suggest analyzing social behaviors using a synthesis of evolutionary theory, control theory (CT), and Pavlovian conditioning. They provide an informative review of recent research into Pavlovian processes in relatively natural settings — with much of that work carried out by the authors. Surely social/sexual behavior is fundamental to any understanding of behavior. Moreover, learning and control systems have an obvious connection. And God knows, it is time for an integrated framework that unites the various fragments of psychological inquiry scattered over dozens of journals. Unfortunately, this initial promise was not fulfilled.

Control theory and psychology

Concepts of CT are hardly new to psychology. Pavlov himself seems to have understood well the role of feedback in maintaining equilibrium, e.g. in selecting just the right mixture of salivary secretions for a given substance introduced into the mouth. Pavlov also -- no surprise -- saw how a conditioned stimulus might adaptively signal the appropriate mixture for a given substance based on prior experience. Nor did he limit such preparatory signals to food availability.

Many psychological conceptions over the years implicitly employed CT concepts — as Miller, Galanter, and Pribram (1960) pointed out in their seminal book Plans and the Structure of Behavior. Domjan et al. claim that CT has not been "extended to social behavior (sect.1, para. 2)". This ignores Bowlby’s (1969) widely cited analysis of maternal attachment in terms of CT. (Also see McPhail and Tucker, 1990.)

Prior uses of CT concepts would not matter if the authors had taken hold of these ideas and made full use of them. However I didn't find the expected discussion of information, models of interactive anticipatory social processes, and control — particularly in light of recent behavioral and physiological interpretations of Pavlovian conditioning e.g. Kamin (1968), Kim, Krupa, and Thompson (1998) in terms of information. The virtue of an explicit CT framework is that it offers an explanatory vocabulary for purposive behavior that transcends diverse methodologies, specific neurophysiological implementations, and parochial terminology.

Feedforward "working models" of social behavior

The article presents a rather simplistic application of feedforward models -— particularly in connection with social behavior. Where is the dynamic element? What can it mean to say, "in principle feed-forward mechanisms are more useful than feed-back mechanisms…(sect. 3.3)?" Are reflexes really more "useful" than learning? Even Pavlov saw conditioned reflexes as just one of several mechanisms, including inhibition, regulating behavior. Milk leaking from a mother’s breast at the thought of feeding is no more or less "useful" than inhibition of lactation by embarrassment. It is the overall adaptive balance of the behavioral system that is important.

Several times I wondered whether the authors oversimplified "social" behavior in order to save conditioned reflexology? Or were they tacitly denying the necessity of treating social "objects" as agents despite their CT framework? Agency requires feedforward models of behavior of considerable computational complexity, involving a hierarchy of goals, intentionality and perhaps "theory of mind" —- maybe a bit much for Pavlovian conditioning? Perhaps I am asking more from the paper than is being offered?

Nevertheless, everyone agrees that feedforward models are necessary for complex interactive behaviors. These "working models," as Bowlby called them, must be adequate for the tasks at hand and indeed in practice reflect the interactive complexities of those tasks. For example, speaking requires a compression of a multilevel hierarchy into a stream of speech in which movements several hundreds of milliseconds in the future must be anticipated by the model. Conversation requires this speaking system embedded within another control system. Consideration of such movement problems led Nikolai Bernstein (1967) and Karl Lashley (1951) to their notorious criticism of the reflexology of Soviet and American behaviorism.

So what does the article offer? As far as I can tell, its feedforward models are little more than a "black box" list of conditioned reflexes. Now this might work for those elements of social behaviors that are not intrinsically social -- behaviors that are directed toward objects that just happen to be agents. What about most interactive mammalian social behaviors including reciprocity, mating, communication, social learning, and parenting? What do the Pavlovian feedforward models for these activities look like?

What is "Pavlovian conditioning" anyway?"

At one time, American psychologists struggled to shoehorn all learning, regardless of species or behavior, into some sort of conditioning model. The subsequent "cognitive revolution" was as much a revulsion at the vague overgeneralizing of conditioning paradigms as anything. Yet the authors continually talk about "Pavlovian mechanisms" as if these are a known causal quantity. There may be a place for the expression "Pavlovian conditioning" in the 21st century — perhaps as a method for studying types of associative learning, maybe a physiologically unique subset of associative mechanisms? Certainly it must be something other than a synonym for predictive association.


Bernstein, N. (1967). The coordination and regulation of movements. London: Pergamon Press.

Bowlby, J. (1969). Attachment and loss: Volume 1 Attachment. New York: Basic Books.

Kim, J. J., Krupa, D. J., & Thompson, R. F. (1998). Inhibitory cerebello-olivary projections and blocking effect in classical conditioning. Science, 279, 570-573.

Lashley, K. (1951). The problem of serial order in behavior. In L. A. Jeffress (Ed.), Cerebral Mechanisms in Behavior . New York: Wiley.

McPhail, C., & Tucker, C. W. (1990). Purposive collective action. American Behavioral Scientist, 34, 81-94.

Miller, G. A., Galanter, E., & Pribram, K. (1960). Plans and the structure of behavior. New York: Henry Holt and Company.