Monday, June 05, 2006

The Cognitive Map : How Mice and Men learn when in the Mouse Trap

As a continuation of the Mouse Trap theme, would like to share links to a very insightful paper in Psychology that had marked a departure from behavioral to cognitive explanations and provided a very relevant concept of Cognitive Maps. This original article by Tolman is a delightful and easy read, though some background in classical behaviorist theories of Instrumental and Operant Conditioning would help. What delights is that Tolman uses and explains these concepts without the associated jargon.

I find the 5 different cognitive modes of learning, he identifies, quite instructive and intriguing:

1) Latent Learning (the mice are not blind, though they may act as such if not motivated enough to have eyes:-): They learn the maze, though if not rewarded they may not exhibit that learning in their behavior. This type of learning helps to clarify the difference between learning something and behavior in-line with that learning; and it is clear that the appropriate behavior is mediated by motivation. Learning happens automatically, irrespective of mediation by goal-directed or reward-presence, maybe subconscious in nature, and in the form of a Cognitive Map that is formed latently; but if no incentives are there to make use of the Cognitive Map then that learning is not reflected in the behavior. In the presence of motivation, learning may become conscious and manifest in behavior.

2) Vicarious Trial and Error Learning (The million dollar question in the Matrix: Which door leads to the Source:-) and the billion dollar question superseding that : which door would Neo choose? :-): This learning behavior that Mice exhibit on a decision or choice point should not have been called Vicarious Trial and Error Learning. The name somehow misses the point. I also do not agree with the explanations. My own two cents: the mice 'choose' between the options presented on a choice point and this VTE is just an observable external behavior reflecting their internal dilemma : whether to choose food or to choose the locked door and fool the Experimenter:-) As mentioned in the Hitchhikers guide to the galaxy, it may end up that mice were experimenting on us all the way! On a more serious note, if we do assume that there may be reasons for mice to choose non-food at some times, lets say when it is satiated and does not really need food but prefers the thrill of bungee jumping back to the start point, then it makes sense that VTE would be observed more in conditions in which the Mice is able to differentiate easily between the options and thus use discretion /discrimination. These are the easy learning situations of the contrasting Black and White doors. For the dark grey and Black doors situation, the learning task is more difficult and so when the Mice doesn't really know which door would lead to food and which to non-food, there is little point in deliberation and he may as well choose any door (or rather the choice he makes may be factually incorrect, so he need not hesitate to choose deliberately - he may as well choose randomly - as he doesn't really have that much control ). Only when choice is real- that is he knows that he has sufficient information to make an informed choice, would there be observed deliberation and choice and associated VTE. Interestingly VTE starts increasing in difficult tasks too, as learning starts to happen and choice become real. An intriguing observation is that the stupid mice do more VTEing in mazes than intelligent ones: are they the really intelligent mice who are acting just stupid and experimenting on Humans?

3. Searching for Stimulus (Who let the dogs out? ): This seems related to the fact that mice would actually indulge in some behavior (in this case looking around their cage for the preponderant stimuli) that is directed towards identifying the salient features of the environment that are associated with their immediate prior experience. They are interested in finding the cause of the effects that they have just witnessed. This interactivity/connectionless view is limited as the experiments focused mostly on avoidance learning and new insights may be available from the similar behavior observed during escape/approach learning if that too exhibits the learning style typical to this avoidance learning : viz searching for stimulus that can be associated with the experience post facto. Interesting to note that when the stimuli following the responses are random, then as the mice's 'search for stimulus' throws an exception, the mice acquires 'learned helplessness' whereby it stops monitoring/analyzing its environment. Thi sis presumed to be the mechanism behind clinical depression in humans.

4. The 'Hypothesis' experiments (the search for patterns/ pattern recognition : to look for the most apt Cognitive Map relevant to this situation): As per this type of learning the mice presumes or forms a hypothesis of what the desired sequence of steps leading to correct outcome should be, actively indulges in systematic exploration to verify the hypothesis and gives up and tries another hypothesis if the results are not favorable. This is the classical 'scientific method' and it is amazing that the mice use that! It is limited in the sense that the experiment is restricted to approach learning. Interesting to note that this type of learning presupposes the existence of 'concepts' like left, right, light and dark in the mice and presupposes an ability to sequence these in temporal fashion and act accordingly.

5. 'Spatial Orientation' learning (Let there be light!): This particular dimension of learning was the most instrumental in Tolman coming up with the Cognitive Map concept. In this, the mice while exploring the maze and learning the sequence of steps that lead to the food box (or goal, which significantly is paired with light in this case), also apparently learnt the precise spatial location of the food box, so that if the maze was replaced by a radiating spokes of alleys, then after some exploration of each alley, the Mice would finally 'choose' one of the alleys and run all the way down the alley (without looking back till it escapes from the alley to find wither the food box or the alley ends into an opening ) and in one particular experimental setup when no other intervening variables were present, the alley that was chosen was significantly related to the presence of food box (light source). Few points to note: Light (or an adequate point of reference like the Sun), seems essential is this sort of learning, maybe the different light/shadow interactions/ intensities are integrally required to from concepts of location. Also, in the first experiment only one light source ( and possibly diffractions effects maybe relevant, while in the second there are 2 sources of light - leaving scope for things like interferences effects to have been instrumental in the learning process. Haven't yet figured out an explanation for the behavior exhibited in the second experiment, but one thing is evident - the presence of a third light (L3 labeled in the diagram) , would have indicated to the mice, that their starting positions and orientations have been changed, and thus they focused on the right direction (left and right ...And thus perpendicular to the walls of the room/table) but missed out on the location. Interestingly this second experiment too exhibits 'escape' behavior, whereby once a mice has chosen an alley, it goes all the way down that alley. I would sum this up mostly as Insight learning regarding the 2-dimensional nature of table-space, wherein the mice learn the 2-dimensional spatial location of an object (food box) and either use the r, theta co-ordinates (angular geometry) to guess its location and behave accordingly, or use the other 'Cartesian' X,Y co-ordinate system to guess the location of the object and take an alley which is perpendicular to the walls of the table (which serve as reference co-ordinates) in its search of the object. In both cases, if my hypothesis is correct, the Rats should ideally not go all along the way down the alley chosen and exhibit the strong escape behavior; but in Radius, theta case should stop/ hesitate when the radius they have covered is sufficiently greater than the actual radial distance; and in the X,Y case, should stop/ hesitate and look for a turn when the reasonable X, or Y distance has been covered, and if the alley does have a perpendicular turn / choice point after some time, than they should turn in the right direction of where the food is located). Interesting hypothesis, but I am afraid I cannot verify these. Isf someone can conducts these experiments for me and inform me of outcome I will be really grateful. In any case, in the present circumstances, this learning mechanism seems to mostly mediate escape behaviors and that too in the presence of light source and is restricted to learning about spatial locations and the nature of Space (2- D for rats)

Interestingly in another article on the web in TIP , these mechanisms are represented in the reverse order and that too with one important transposition wherein the order of Escape and Approach is reversed. : 1) approach 2) escape, 3) avoidance 4) choice point and 5) latent. This is described under 'sign learning' and I find that fascinating as I am currently hooked to things like Da Vince Code ( haven't read the book or movie, jut documentaries on the same) or tarot and looking for subtle signs that would help in uncracking the code/ breaking the matrix. A tidbit from this page that I find intriguing is that Tolman was also investigating motivation for war.

Resuming discussion on the original article, it also mentions 3 mechanism related to Cognitive Maps that are observable in Humans ( I presume this is over and above the 5 Learning mechanism that are definitely present in both Mice and Men). These, of a different dimension, are briefly mentioned below, but seem to be based on Freudian defense mechanism and are generally speaking unhealthy (defense) mechanism that we may use by referring to Cognitive Maps that are not in touch with reality ( and the learning and behavior instead of being based on Reality principle are based on Pleasure principal). These are summarized below (though haven't though about them in details and will follow up in a later mail with more elaborations) and this sort of learning ideally needs to be unlearned in order for effective behavior.

6) Regression : Reverting back to earlier learned cognitive maps that are no more relevant in the present situations when dealing with things like emotional loss ( the exact example given is of a loss of spouse...And this concrete example may have much to do with the situations in which this 'learning' becomes relevant)

7) Fixation : Using one particular Cognitive Map in all situations as it may have served well in one particular situation in which it was associated with high reward/ motivation.

8) Displacement of aggression onto outgroup : This, on first look, appears to be the classical displacement (of anger) that is touted as a defense mechanism when a man that returns home angry on the boss, shouts on his wife; but this is made contingent to the fact that aggression is directed on outgroup and is thus qualitatively different. This assumes concepts like group and belonging to group as well as ability to differentiate and discriminate between group and outgroup. It also involves deliberate hate ( and not just prejudice of the outgroup) towards the outgroup, who is made a willing target of aggression in case of frustrations within the group.

All the above 3 negative learned mechanisms Tolman maintains is a result of narrow cognitive Maps. The article ends on a very philosophical note!!

Before concluding, I would just like to briefly summarize my position on the experiments done in the mouse traps:-). I believe the correct order ( an this does matter in the theory I am trying to build up) should be

1) Latent 2) Choice point 3) Avoidance 4) Escape and 5) Approach

6) regression 7) fixation 8) Displacement to outgroup

the only difference I propose in ordering is that 4) Escape (learning related to Space) comes before 5) Learning related to time ( sequence of steps / patterns in time that lead to correct outcome)

Also, to briefly give you an idea from where I come from

I believe in the eight stage developmental theory of Erikson with the 5th step of ego formation being a definitive step in development of self, whereby the adolescent tries many roles in different situations, and achieves ego integrity (or finds the role that he is supposed to play) on successful completion of this step. The rest 3 development tasks of Intimacy, efficacy and Integrity are of qualitatively different nature that the preceding five tasks. Of course I don't take the Freudian background or explanations of these stages/ tasks but am more into the cognitive developmental theories fir eg of Piaget.

Do comment and let us rock!!

Sphere: Related Content

No comments: