LOGAN: BASIC LAWS OF LEARNING 1/1/2000 SPATIAL LEARNING Abstract This publication provides an extension of the sys- tematic empirical analysis of learning paradigms. This volume concerns the domain of learning to traverse mazes or other environmental obstacles enroute to some goal. This form of learning presents the organism with two or more alternative routes from which to choose, with some alternatives frequently being dead-end cul-d-sacs. But the domain also includes any situation involving finding and remembering the location of some form of positive reinforcement. (Use of aversive motivation is classified as Escape Conditioning, although that distinction is an arbitrary and perhaps questionable one.) Spatial learning is an amalgam of discrimination and differentiation learning. That is to say, on the one hand, there are many potential intra-maze cues (visual, olfactory, tactual) as well as extra-maze cues such as the windows, doors, and animal cages in the room. Insofar as the organism uses these cues as the basis for learning the location of the goal objects, spatial learning is a special case of serial discrimination learning. On the other hand, traversing a maze entails qualitatively different orienting/turning responses, classifying it as a form of serial differentiation learning. Many of the issues that arose in the context of spatial learning were a result of this operationally ambiguous status. For example, one hypothesis was that rats acquire a cognitive map of the maze. One possible implication of this hypothesis is that, if a new direct short-cut to the goal is provided, the organism will choose that path rather than the old, familiar, but longer one. Many studies confirmed that prediction, but some did not. The result depended on whether a distinctive cue, such as a light, was placed over the goal box. Proponents of the cognitive-map construct contended that the light was legitimate and indeed necessary in order for the organism to be able to orient a map appropriately. Opponents, however, contended that the light changed the task from a left-right differentiation to a discrimination based on "approaching the light." As usually turns out to be the case, the supposedly crucial test of a theory was inconclusive. Nevertheless, the issues involved in systematic analyses of spatial learning served to sharpen the lines of cleavage among competing theoretical systems. In doing so, they enriched both the empirical and systematic bases for an understanding of learning processes. Certainly one of the most pervasive forms of learning in natural life is navigating around one's physical environment, be it an elephant returning to a source of water or a college student going to class. Accordingly, no comprehensive theory of learning can be complete without including spatial learning as a basic, fundamental phenomenon. INTRODUCTION Much of this text is derived from Logan & Ferraro, "Systematic Analyses of Learning and Motivation," (Wiley, 1978) To the experimental psychologist concerning primarily with animal learning and motivation during the period following the advent of behaviorism, the apparatus of choice was clearly the maze. This was a quite natural development, because the white rat was also the organism of choice and spatial learning intuitively appears to be a "natural" for a rat. Mazes could readily be constructed to vary in length and presumed difficulty; errors could be arbitrarily defined and objectively counted; and organismic factors, such as drive motivation and cortical damage, could be studied in that context. In retrospect, however, the maze was actually too complicated a situation for the neophyte science of behavior. Nevertheless, it is important to reserve a place for spatial learning in a complete systematic analysis in part because that is the context in which many of the issues arose. The what-is-learned question initially involved maze learning, and the hypothetical construct of a cognitive map is most conducive to thinking in terms of spatial orientation. The studies of latent learning were also developed in the context of multiple-unit T-mazes. Some of the issues that arose are somewhat specific to that situation. One of these, for example, involves centrifugal swing...especially if the choice points are relatively close together, the rat naturally comes out of one turn with physical rather than psychological forces favoring the other turn at the next choice point. However, even in this context, it proved to be useful to reduce the problem to the study of alternation in a simple T-maze: Alternation Behavior. Although alternation in a multiple-unit maze could be attributed to physical dynamics, many theorists were more inclined to think of alternation as a reflection of a basic curiosity or exploratory drive. The alternative approach was sometimes called the "inhibition of reinforcement," or reactive inhibition. This is consistent with the facts that rats tend to alternate the direction of turns from trial to trial, regardless of whether or not the preceding trial was reinforced, and this alternative tendency decreases with the time between trials. However, one can pose the "What-is-alternated?" question--the response or the stimuli? Cues in Spatial Learning. Many of the difficulties in the early studies of spatial learning arose from the use of elevated mazes (simply boards along which the rat crept for fear of falling) that provided abundant extramaze cues. Other researchers used enclosed mazes that restricted the sources of stimulation to proprioceptive and intramaze cues. As was characteristic of early systematic analyses of learning and motivation, the issue was treated as an either-or analysis: the rat is a place learner or a response learner. And as was also characteristic of early research on such issues, the results were ambiguous because a rat may use all available information in the process of learning. It is largely for this reason that the systematic experimental analysis of selective control of behavior has shifted from the complex environment of a maze to compound stimuli whose elements can to a greater extent be controlled by the experimenter in operant and instrumental contexts. Insight. Although the most quoted example of insight in animals is Sultan--Kohler's chimpanzee who joined two sticks together in order to make one that was long enough to reach a distant banana--insightful behavior was systematically studied in spatial learning situations. One of the more interesting is the Umweg problem, in which the organism must make a detour in order to get to a visible goal object. One of the major problems in this context is that simply labeling the occurrence of a sudden solution to a problem, "insight," hardly provides much of an explanation of the event. Correction. The multiple-unit maze inherently involves a correction procedure. The rat may enter a cul-de-sac but can then retrace to correct the error and proceed on toward the goal. This poses a problem for proponents of a reinforcement theory. An entry into a cul-de-sac is not nonreinforced but receives delayed reinforcement, and the time of this delay is not under experimental control because it depends on how far the subject pursues the incorrect route and how long it takes to retrace to the choice point. And then, of course, the cues at the choice point are not the same when approached from the incorrect arm as from the stem. These features make a detailed analysis of learning in a multiple-unit maze inherently difficult and led to a preference for focusing on the simplest form of spatial learning, namely a single-unit T-maze. The Goal-Gradient Hypothesis. Spatial learning may be viewed as learning a chain of responses with terminal reinforcement. So conceived, those responses closest to the goal receive the shortest within-chain delay of reinforcement. Hence, from a behavioristic perspective, there is a gradient of delay-of-reinforcement working backwards from the goal. This implies that errors should be eliminated in backward order, and that speed should increase as the goal is approached. However, a second complication arises from such a perspective, the anticipatory-response tendency. Responses learned near the goal should tend to become anticipatory and aid or hinder earlier choices depending on whether the same turn is appropriate at those choice points. Which introduces yet a third complication: the early choice points are more distinctive simply by virtue of being first. This suggests a reverse goal gradient based on generalization from one point in the maze to another. In any event, it should be clear that simply counting total errors suppresses important information about errors at different points along the way. Genetic/physiological Factors. From the beginning, spatial learning has been a useful context in which to research the heritability of aptitude for learning and memory and for the biological bases of the psychological processes. Just as the multiple-unit maze was largely replaced by the T-maze, the latter has largely been replaced by radial and water mazes. In any event, over fifty years ago Hull agreed with the following statement by Tolman: "I believe that everything important in psychology (except perhaps such matters as the building up of a super-ego, that is, everything save such matters as involve society and words) can be investigated in essence through the continued experimental and theoretical analysis of the determiners of rat behavior at a choice-point." To that, Tolman added the following quote: "To my ratiocinations I hope you will be kind, as you follow the wanderings of my amazed mind."