Theoretical Assumptions

The main (hypo-)thesis of this paper assumes that the behavior function $ \phi_{rat}$ of a rat is not only realized by a network of simple stimulus-response connections - let us name it $ S-R-Model$, but rather by a cognitive map of the environment. This term 'cognitive map' is not explicitly defined by Tolman (!), thus one has to 'infer' the 'possible meaning' of this term by his examples and remarks. My interpretation of his term goes as follows: the rats have the ability by moving in an environment to generate a space-object representation - let us name it $ CogMap$ - which represents the spatial properties and relations of the viewed space and the viewed objects in a way, that this representation allows the rat to infer a possible path between a point in the $ CogMap$ interpreted as actual position to another point in this representation where the rat has stored information about some rewards.Then is the rat able to follow this cognitively represented path in the real environment.

There are some additional hypotheses which are interesting too, which we will include in our development.

Besides the learning of a path by positive reward there exists a kind of un-learning of before-learned paths in the case of frustration if the food does not appear at that place where it is expected or - even stronger - that an explicit punishment happens if the rat makes a 'wrong' decision. While the terms 'punishment' and their contexts are lacking some clearness seems the 'un-learning' in the case of 'not finding the food at the old position' sufficiently very clear.

Finally Tolman observes the phenomenon of a kind of 'regular behavior' in a series of random choice-situations which he interprets with the hypothesis that the rat uses internal rules which generate preferences which possibility should be selected 'rather than another'. These 'internal rules can change.

The theoretical relationship between these internal rules and the $ CogMap$ representation is not clear. At a first sight one would assume that both are different mechanisms. But one has to consider that in those cases where the rat shows a behavior which one would interpret with the internal rule there seems to be no $ CogMap$ representation available, otherwise the rat would be able to follow a certain path. Otherwise -the possibility of 'latent learning' - even if a $ CogMap$ representation would be available, as long as no clear goal is 'active', the information of a $ CogMap$ is not of much use; every path has the same 'value'.

The fourth hypothesis about Vicarious Trial and Error (VTE) appears to fuzzy than to discuss it in our context; perhaps later where our virtual learning systems are more advanced.

Summing up all these different hypothesis one has to keep in mind that Tolman did nowhere give an explicit formal account of his hypotheses nor of the used theoretical terms. This is a weak point of his position. Nevertheless this paper gained a huge echo in the community and became one of the landmark papers in animal-based learning theory.

The interpretation that his position 'ended' the behavioral approach in psychology and instead opened up e new cognitive approach I would not follow. From the point of philosophy of science (in German more adequately 'Wissenschaftstheorie') he did not(!!!) change the paradigm, he only extended the theoretical model which interprets the empirical data. The 'radical behaviorists' (e.g. Watson, Hull, Skinner...) based their theoretical models on the observable behavior of animals and men. The same did Tolman. The only difference is located in the models: while the radical behaviorists allowed in their theoretical models only networks of stimulus-response connections Tolman allowed more complex models, especially a kind which he called cognitive maps. The change of the theoretical model does not change the paradigm of a behavior-based theory. Moreover one has to state that Tolman never gave an explicit theoretical account of the structure and functions of a 'cognitive map'. Thus it is not excluded that someone will give in the future a theoretical explanation of 'cognitive maps' in terms of 'stimulus-response connections'. If one considers the fact, that today nearly every cognitive ability can be modeled by using artificial neural networks or extended classifier systems, then the assumption of a general difference between the radical behaviorists and the cognitive approach seems to vanish. Neural networks are nothing more than sets of interconnected units which realize nothing more then stimulus-response connections (as well as extended classifier systems). Therefore it seems that the story of the disappearance of radical behaviorism by the appearance of cognitive psychology is only an artefact caused by a severe lack of theory on psychology.

Gerd Doeben-Henisch 2013-01-14