CHAPTER  12

Places, Paths, and Bearings

Tolman argued that the Skinner box and the straight-alley runway make animals look stupid because they give them only stupid things to do. He designed his experiments to show how intelligent they could be when psychologists give them intelligent things to do. Probably the most productive line of research that Tolman inspired began with the argument about just what a rat learns in a maze. His notion of cognitive maps that guide the behavior of human and nonhuman animals influenced theory and research throughout the 20th century. This chapter introduces modern studies of foraging and orientation that offer us more ethological and more intelligent alternatives to the traditional cognitive map.

PLACE LEARNING VERSUS RESPONSE LEARNING

Chapter 5 described in detail how early reinforcement theorists such as Hull explained maze learning as a series of links in a chain of stimuli and responses ending with reward in the goal box. According to early cognitive theories on the other hand, rats learn a cognitive map of the maze—start box, choice point, end boxes, runways, and (when extramaze cues are available) the room that houses the maze. When rats learn that food is in the goal box at the end of the right arm of a T-maze, then they naturally go to the right at the choice point because that is the shortest route to the food. Rather than strengthening a chain of stimuli and responses from choice point to choice point, experience in the maze teaches cognitive rats where the food is and how to get there.

Rats could find the food in the usual maze either way: by learning to go to the place where the food is or by learning to make correct turns at each choice point. Tolman, Ritchie, and Kalish (1946) wanted to find out whether rats learn to return to the same place regardless of the responses it takes to get there, or instead, learn to repeat the same turns at each choice point. They designed the cross-maze shown in Fig. 12.1 to define the difference between place learning and response learning. This maze has two goal boxes, G1 and G2, as in the usual T-maze, but it also has two starting boxes, S1 and S2. When the experimenter always rewards a rat in the same place, say G1, then the rat must make right turns after starting at S1 and left turns after starting at S2. When the experimenter always rewards a rat for making the same turn, say a right turn, then the rat must run to G1 after starting at S1 and to G2 after starting at S2. If rats naturally learn to return to the same feeding place, as cognitive theorists claimed, then it should be easier for them to learn to return to the same goal box on every trial, even when they must take different paths from different starting places. If rats naturally learn to repeat the same response at the same choice point, as reinforcement theorists claimed, then it should be easier for them to learn to repeat the same turn on every trial, even when the same turn leads them from different starting places to different feeding places.

Each of the rats in this experiment started equally often from S1 and S2 in an alternating sequence, which was S1, S2, S2, S1, S1, S2 on odd days and S2, Sl, Sl, S2, S2, Sl on even days. The place learning group always found food in the same goal box. For half of the place group this was G1, and for the other half this was G2. An animal in the place group that always found food in G1 got there by turning right when starting from Sl and by turning left when starting from S2. An animal in the place group that always found food in G2 got there by turning left when starting from Sl and by turning right when starting from S2.

FIG. 12.1. Plan of the cross-maze used by Tolman et al. (1946) to compare place learning with response learning. Copyright © 1997 by R. Allen Gardner.

The second group, the response learning group, always found food after making the same turn at the choice point. For half of the response group this was a right turn, and for the other half this was a left turn. An animal in the response group that always found food after a right turn found food in G1 when starting from Sl and in G2 when starting from S2. An animal in the response group that always found food after a left turn found food in G2 when starting from Sl and in G1 when starting from S2.

If animals learn where the food is, then place learning should be easier than response learning. If, instead, animals learn to make right or left turns in response to the stimuli at the choice point, then response learning should be easier than place learning. The results of the first experiment by Tolman, Ritchie, and Kalish were decisive. All of the animals in the place group reached the learning criterion of 10 successive correct choices within 18 trials and all but three of them reached this criterion within 12 trials. Meanwhile, only three of the animals in the response group reached criterion within 72 trials.

All of the animals in the place group quickly learned to run to the rewarded place, and most of the animals in the response group soon ran exclusively to one of the goal box places, even though this strategy earned food reward on only 50% of the trials. Chapter 11 reviews the mass of evidence that 50% reinforcement builds up very strong habits. Place learning is such a dominant mode of learning that nearly all of the animals formed a place habit in both groups.

Later experimenters varied conditions to discover what stimuli the animals used to find their way back to the same place each time. The basic procedure started with a large experimental room, say ten to twenty times the size of the cross-maze in Fig. 12.1 in which there might be windows in the south wall, a door opening on to a hallway in the north wall, the rack of home cages along the east wall, and sinks and storage shelves on the west wall. Overall, the experimental rooms were unevenly lighted and unevenly heated and ventilated. Thus there were many extramaze stimuli to tell the rats where they were in the room. These experiments demonstrated that rats use cues to place and direction in a room rather than right and left turns. They can use right and left turns, but they only use turns when experimenters block out all other information (see Restle, 1957, for a detailed review).

Cognitive theories are top-down theories. They demand some sort of executive entity in the brain that receives inputs transmitted upward from the sensory systems, studies fresh inputs and stored memory, makes decisions, and then sends commands downward to the motor systems. Reinforcement theories seem less mentalistic but reinforcement theories are top-down theories, also. That is, first some central executive entity must decide that the animal has been reinforced, that the S that followed the R was, indeed, an S* or an S^r. Second, after a period of trial and error, some executive entity must store information about the contingency between S* and R and, if there is a contingency, then how large a contingency. Then the executive entity must command motor systems to make Rs. Place versus response experiments demonstrate conclusively that the dominant mode of the information and the commands is in terms of places and goals rather than specific stimulus-response units. Until the development of modern robotics (chaps. 7 and 9), the only sort of system that could handle such information seemed to be a top-down cognitive system using entities like cognitive maps.

The trouble with the cognitive map has always been that it seems to dodge the question of maze learning. For example, if rats use cognitive maps, where are those maps and who looks at them? If there is a little cognitive rat inside who looks at the cognitive map, then we only have a new set of puzzles about the cognition of the little rat inside of the big rat. Saying that the little cognitive rat inside “naturally” or “obviously” takes the shortest route tells us nothing about how any actual rat chooses any actual route. To the hardheaded theorists of the first half of this century, the only scientific alternative to cognitive maps seemed to be some version of the goal-gradient mechanism of chapter 5 that analyzes a runway or maze into many small segments, each providing its own stimuli and each requiring its own responses.

SPONTANEOUS ALTERNATION

On the first trial in a simple T-maze, about half the rats find food in the first goal box that they enter. What would you expect a rat to do on the second trial if, on the very first trial, it ran through the maze and found food at the end of the right arm? It seems as though both reinforcement and cognitive theories should predict that, while the effect of a single reward on the first trial might be small, the rat should be somewhat more likely to repeat the rewarded choice. A very large number of experiments, however, contradict this implication of both reinforcement and expectancy theories. After finding reward on the first run through a T-maze, rats mostly run to the opposite arm on the second trial. This powerful phenomenon is called spontaneous alternation. Depending on various conditions, 90% or more of the turns on Trial 2 can be in the opposite direction to the turn on Trial 1, whether or not the animal found reward on Trial 1 (see Dember & Richman, 1989, for extensive discussion).

From an ethological and ecological point of view, alternation is a highly appropriate strategy for an animal that lives by foraging for food. This is the first time that the animal explored this particular territory. If the animal can remember that it found food on the right, then it must also remember that it consumed the food that was there. So, on the second trial, it would be a good strategy to strike out in a new direction. If the animal found the first goal box was empty from the start, then it is even more sensible to try the left arm on the next trial.

Montgomery (1952) and Glanzer (1953) compared place alternation with response alternation in cross-mazes like the one used by Tolman et al. (1946) and illustrated in Fig. 12.1 to study place versus response learning. They started the rats alternately at start boxes Sl and S2 and fed them at goal boxes Gl or G2, whichever the rats chose on the first trial. They next gave the rats a second trial starting from the opposite start box. On the second trial, the rats could alternate places or they could alternate responses. They found that rats alternate places rather than turns on Trial 2 just as they learn places rather than responses in maze-learning experiments.

Intramaze Stimuli Versus Extramaze Stimuli

Walker, Dember, Earl, and Karoly (1955) pointed out that place in the room in the Montgomery and Glanzer experiments was confounded with place in the maze in an apparatus like that in Fig. 12.1. That is, with mazes made of wood or composition board, as is common in such experiments, the appearance and texture of each arm are bound to be slightly different. If Gl always points east and G2 always points west, then we cannot tell whether the rats are alternating stimuli coming from within the maze, intramaze stimuli, or stimuli coming from outside the maze, extramaze stimuli.

Walker et al. (1955) painted the arm leading to one of the goal boxes black and the arm leading to the other goal box white to increase the stimulus difference between the two arms. The maze had wooden walls and a wire mesh roof through which the rats could see the major landmarks of the experimental room. In their first experimental condition, PI in Fig. 12.2, they used the same floor plan (Fig. 12.1) that Tolman et al. (1946), Montgomery (1952), and Glanzer (1953) had used, and started each rat from SI on Trial 1 and then from S2 on Trial 2. A rat in Condition 1 that ran to B on Trial 1 and D on Trial was repeating left turns but alternating both intramaze stimuli and extramaze stimuli. Similarly, a rat in Condition 1 that ran to D on Trial 1 and B on Trial 2 was repeating right turns but alternating both intramaze stimuli and extramaze stimuli.

FIG. 12.2. Two floor plans used by Walker et al. (1955) to separate intramaze stimuli from extramaze stimuli. Copyright © 1997 by R. Allen Gardner.

In their second experimental condition, Walker et al. used the same floor plan as before for Trial 1, PI in Fig. 12.2, and started each rat from S1, but then they rotated the whole cross-maze by 180°, as shown in PII of Fig. 12.2, before starting each rat from S2 on Trial 2. Thus, B pointed east and D pointed west on Trial 1, but after the rotation B pointed west and D pointed east on Trial 2. Consequently, an animal in the second condition that ran from S1 to B on Trial 1 and then from S2 to D on Trial 2 was alternating intramaze stimuli but repeating left turns and westerly extramaze stimuli. Similarly, an animal in the second condition that ran from S1 to D on Trial 1 and then from S2 to B on Trial 2 was also alternating intramaze stimuli while repeating right turns and easterly extramaze stimuli. In the third and fourth conditions, Walker et al. started each rat from the same point, either S1 or S2, on both Trial 1 and Trial 2. They used floor plan PI for both trials of the third condition and rotated from PI to PII for the second trial of the fourth condition.

In this way, Walker et al. pitted all three alleged sources of alternation against each other in different combinations in the four different conditions so that they could measure the contribution of each to spontaneous alternation. They found 80% alternation of intramaze stimuli and virtually no alternation of extramaze stimuli or responses. Note that Walker et al. deprived their rats of food during the experiment, but they never gave the rats any food in the cross-maze.

Odor Trails

Many mammals, including laboratory rats, have scent glands on the underside of their bodies and mark their trails with a variety of odors. A rat can tell a lot from an odor trail. It can tell whether the trail was laid down by itself, another familiar rat, or a stranger. It can tell the sex of the rat that laid down the trail. It can tell how old the trail is. It can tell whether the rat that laid down the trail was running away frightened or foraging for food. It can even tell whether the rat that laid down the trail had just detected food or stimuli predicting food (Ludvigson & Sytsma, 1967). Scientists have known for a very long time that odor trails have significant effects on behavior, and careful experimenters have always cleaned out the floors of mazes after every trial to eliminate this source of stimulation. Sad to say, more experimenters failed to observe this precaution as if rats cared as little about odor trails as human beings. Fortunately, the effect of odor trails is irrelevant to the objectives of many experiments.

It is scandalous that there were so many experiments on spontaneous alternation before anyone considered the question of the odor trail left behind on Trial 1. Douglas (1966) pointed out that in all previous experiments on spontaneous alternation—like those of Montgomery (1952), Glanzer (1953), and Walker et al. (1955)—odor trails left behind on Trial 1 were completely confounded with intramaze stimuli. That is, when rats alternated goal arms in these experiments they could have responded to the odor trail that they, themselves, had left behind on Trial 1 and ignored all other differences between Gl and G2, such as the fact that the goal arms were painted different colors. On Trial 2, rats can tell which alley they tried on Trial 1 by sniffing. They do not have to remember anything.

Douglas eliminated odor trails by spreading fresh paper on the floor of the maze between Trial 1 and Trial 2. With the additional control for odor trails, Douglas repeated all of the alternation tests of previous experiments and added a few new ones. He tested for response alternation in isolation by using one T-maze on Trial 1 and then using a different T-maze in a different experimental room with fresh paper on Trial 2. He tested for the effects of extramaze stimuli in isolation by using one T-maze on Trial 1 and a different T-maze with fresh floor paper on Trial 2 while placing the second T-maze in the same location in the same experimental room as Trial 1. He tested for intramaze stimuli in isolation by keeping the T-maze used on Trial 1 the same, including the paper floor and moving the whole business to a different experimental room. As in Walker et al. (1955), Douglas never gave his animals any food in the T-mazes. This is an easier procedure for the experimenter and may also increase the amount of alternation.

In his first series of tests, Douglas found that the effect of extramaze stimuli alone was 75% alternation, the effect of intramaze stimuli alone was 61.5% alternation, and the effect of response alone was exactly 50% or chance-level alternation. Douglas next performed a series of tests to isolate the stimuli responsible for intramaze alternation. He varied the composition of the walls of the goal arms as well as the color, but the only stimulus that led to alternation was an odor trail on the floor paper. No matter how Douglas varied other intramaze stimuli, whenever he put down fresh paper on Trial 2, alternation dropped to chance level. Yet, even when Douglas used a different T-maze in a different room on Trial 2, he got 65.5% alternation when he used the same floor paper on Trial 1 and Trial 2.

An animal can save itself the trouble of storing a cognitive map and studying it later if it can leave behind a trail of markers as Hansel and Gretel did when they were kidnapped in the forest. It is embarrassing that experimental psychologists required so many decades to discover that nocturnal animals, like rats, are much more interested in patterns of odor and much less interested in patterns of light than human beings. With a little more interest in the ethology and ecology of their experimental subjects, experimental psychologists could have discovered long ago the well-known fact that rats, like many mammals, have scent glands that they use to mark places.

Compass Bearing

Douglas then made an unexpected discovery. His next series of experiments aimed at isolating the extramaze stimuli responsible for alternation. He tried moving all sorts of visual, auditory, and olfactory sources of stimuli from the Trial 1 room to the Trial 2 room, but alternation always dropped back to the 50% chance level in the second room. Douglas only got extramaze alternation when he retested the animals in the same room in the same maze location on Trial 1 and Trial 2.

In one of these futile searches for the source of extramaze alternation, he made a slight change in procedure. Up to that point, Douglas had rotated the T-maze 90° (see Fig. 12.3) whenever he changed rooms, so that the goal arms were running from east to west in one room and from north to south in the other room. As he became more and more discouraged, he neglected to rotate the maze when he changed rooms in one set of tests. The effect was an immediate recovery of the roughly 75% alternation level found when the maze was kept in the same location in the same room.

Douglas next showed that when he changed everything—T-maze, room, and floor paper—but kept the goal arms in the same east-west or north-south orientation, he always got roughly 75% alternation. If he rotated the maze 90°, however, alternation dropped to chance in every condition except the one in which he reused the same paper from Trial 1 to Trial 2. He also found the usual 75% alternation when he moved the T-maze around in the same room, but kept the east-west or north-south orientation of the goal arms constant. Soon, Sherrick and Dember (1966a, 1966b) confirmed Douglas’ findings in their laboratory.

How can rats tell the difference between an east-west orientation and a north-south orientation? Douglas guessed that they did this with the semicircular canals in their inner ears. Mammals have an interesting set of organs in their inner ears consisting of three semicircular canals at right angles to each other so that they respond to rotation in all three major planes of the body. Rotation in any direction stimulates the sensors in the semicircular canals. This is the rotation that makes children dizzy when they spin themselves around. Douglas guessed that the rats could tell which way they had turned last and which way they were pointed next if they could sense horizontal rotation.

FIG. 12.3. T-maze with 90° rotation. Copyright © 1997 by R. Allen Gardner.

Douglas found that he could reduce the effect of compass bearing on alternation by holding a rat at chest level, horizontally, head pointing outward, and then pirouetting rapidly around for a full 360°. He could eliminate alternation by compass bearing, entirely, by making eight rapid turns in succession. He worried that the pirouetting of the big man might disturb the little rats so he controlled for the disturbance. When he held a rat at chest level, horizontally pointing outward, but now rotated the rat vertically head over heels, the animals continued to alternate compass direction.

Recently, Etienne and her associates (see Etienne, 1992, for a summary) studied the stimuli that hamsters use to find their way back to a home nest after foraging. The experimental apparatus consisted of a circular arena, two meters in diameter, with a nest box attached to the outside of the boundary wall. The hamsters entered the arena from the nest box through an opening in the wall of the arena. When hamsters find a source of food (Etienne et al. provided little piles of filbert nuts), they seldom eat on the spot. Instead, they stuff as much as they can into their cheek pouches and bring the lot back to hoard in their home nests. Next time a hamster is away on a foraging trip, the experimenter can empty the nest and reuse the nuts. In this way, hamsters never get satiated for nuts and go on foraging, indefinitely.

In the course of foraging, a hamster may travel various distances and make many turns. Nevertheless, a hamster returns directly to its home nest by a fairly accurate linear path, as shown in Fig. 12.4. They can use various kinds of visual landmarks to find their way home, but they can also find their way in total darkness. In total darkness, the experimenters monitored the movements of the hamsters with infrared light. Human beings need special optical aids to see in infrared light and without such special aids hamsters are in the equivalent of total darkness. Etienne et al. used a variety of devices to eliminate other kinds of differential stimuli that might arise from the experimental room and the experimental apparatus.

Etienne et al. could lead a hamster around the arena by baiting a spoon with food and holding it close to the hamster’s nose but just beyond its reach. In this way they could lead a hamster by an experimentally controlled path to a pile of nuts. The hamster then promptly filled its cheek pouches and returned home on a direct line. If the devious experimental path included as many as five 360° turns in the same clockwise or counterclockwise direction, the hamsters started to make errors. If the experimental path out had six complete turns in the same direction, the hamsters made serious errors in their return path. After eight 360° turns in the same direction, they acted quite lost. If the turns in the foraging path were alternately clockwise and counterclockwise, the hamsters acted as if turns in opposite directions cancel each other out.

FIG. 12.4. A hamster forages in an arena and returns to its nest box. (After Etienne, 1992) Copyright © 1997 by R. Allen Gardner.

Another way that Etienne et al. confused a hamster was by letting it find the pile of nuts on a circular dish-shaped platform after foraging on its own. The platform had a hook on its edge and the experimenter rotated the platform with a long pole while the hamster was busy stuffing its cheek pouches with nuts. One or two passive 360° clockwise or counterclockwise turns on the platform and hamsters began to make significant errors on the return path. Three or more complete passive turns and hamsters acted as though they were completely lost. Like rats, hamsters must use information about rotation from the semicircular canals in their inner ears to guide them back to their home nests.

Rats and hamsters seem to navigate in the dark by first noting the direction that they take when leaving the home nest. With each turn and each excursion in a particular direction, they can judge the angle between themselves and the home nest. It is as if they have a horizontal dial in their heads with a pointer that always points toward home. As long as they stay on the first heading, the pointer indicates that a return path 180° from the present heading is the best path home. If the animal makes a 90° turn to the right and proceeds on this new heading, the pointer keeps pointing home (at first 90° to the right of the new heading) and indicates a larger and larger angle for the correct path home. If the animal next turns 120° to the left, then the pointer begins to indicate a smaller and smaller angle for the correct path home. Equipped with an internal device such as this, an animal never has to form a cognitive map of the territory and never has to store more than one bit of information in memory. A forager only has to follow the angle that currently indicates the way home. If there is still food left at the last foraging sight, the hamster can find its way back by reversing the last return direction when it starts out the next time.

Navigation

Hamsters and rats must rely on odor trails and semicircular canals when they are in total darkness. Etienne and her associates and other modern experimenters (see Poucet, 1993) have also shown that animals can use visual landmarks to guide themselves home and also to guide themselves outward to return to a good foraging site. Etienne and her associates have placed landmarks such as a small light or a silhouette at the edge of the arena or at some distance beyond the arena. Hamsters usually orient to relatively distant landmarks corresponding to the relatively stable landmarks that they would see on the horizon in a natural habitat (Etienne, Lambert, Reverdin, & Teroni, 1993). Under natural conditions they can use visual landmarks together with their semicircular canals. That way they avoid confusion when important visual landmarks are obscured by obstacles or after they have made many turns while foraging. Navigation is for getting home and getting back to a good foraging site when there is a distance to travel. Once an animal gets near home base or back to the foraging neighborhood, it can use distinctive local cues such as particular plants or rocks (Poucet, 1993).

Navigation is based on compass bearing and distance traveled rather than map location. That way a hamster in a natural habitat can get home even though it has to travel a zigzag path to get around bushes and rocks. All it has to do is to keep returning along the same compass bearing. It is unnecessary to hold the actual zigzag path in memory, because following the compass bearing and actual distance traveled will get them back to the neighborhood of the forage site. When they get close, odor and other local cues can take over. It is also quite unnecessary to store in memory an elaborate map of the foraging territory and the obstacles that define the many alternative routes between home base and all possible foraging sites. Nor is it necessary for the little hamster inside the big hamster to pour over such a map in order to tell the big hamster how to get from place to place.

Honeybees forage over even larger territories than hamsters. When bees find a good foraging site they use the sun to guide themselves back to the hive on a beeline. Exploring for food in the usual haphazard path can take them to some places where they cannot return to the hive on a beeline. If there is an obstacle, for example a hill, the scouts interrupt their beeline to the hive in order to fly around the obstacle, leaving the obstacle at the beeline angle (see Fig. 12.5). Back at the hive, their dances communicate the total amount of flight as well as the compass bearing. Followers can retrace this path by reckoning the detour into their flight path. It is so easy that they can do it with their tiny bee brains. Robotics engineers should take note.

In an ingenious experiment, Kirchner and Braun (1994) showed how bees navigate by means of compass bearing and flight distance. The experimenters managed to attach tiny magnets to the backs of foraging bees. They placed the bees in a bee-sized wind tunnel that was about 10 meters from the hive. When the bees flew against the artificial wind, the magnets tethered them so that they flew in place for varying amounts of time. The top half of the wind tunnel was open to the sky so the bees could maintain their bearing with respect to the sun. The bottom half of the wind tunnel had a pattern of lines on it that appeared to move beneath the bees much as the ground moves in normal flight. The direction of flight in the wind tunnel was perpendicular to the angle between the hive and the tunnel. When released, the bees found a rich source of food just outside the exit.

When Kirchner and Braun allowed some of the bees to return to the hive without further intervention, the bees flew at right angles to the correct bearing of the hive as if returning along the path they had flown in the wind tunnel. Some of these bees flew more than 200 meters in the wrong direction before giving up and circling back to find the hive. That these lost bees found their way home agrees with experimental evidence (Dyer, 1993; Dyer, Berry, & Richard, 1993) that bees, like hamsters, can also use distant landmarks on the horizon for navigation. Since the wind tunnel and the food source were only 10 meters from the hive, the bees could have used local landmarks to return to the hive. That they relied on flight distance, instead, indicates the flexibility of a navigational strategy. They navigate until they are near the hive and then they use local landmarks to find the hive, precisely. Dead reckoning only has to get them to an area near the hive. While they are navigating they can ignore local landmarks along the flight path, which they might otherwise confuse with similar landmarks near the hive (see also Chittka, Geiger, & Kunze, 1995).

FIG. 12.5. A steep hill between hive and food requires a detour. The bearing is the line from Hive to Food and the distance is the detour line, Hive to D plus D to Food (after von Frisch, 1953, pp. 124–125). Copyright © 1997 by R. Allen Gardner.

When most of the bees finished feeding outside the tunnel, Kirchner and Braun replaced them in the wind tunnel and retethered them for a return flight of the same length as the outward-bound flight, but in the opposite direction. When these bees left the tunnel at the original entrance, they found their way directly back to the hive. Only 10% of them danced in the hive after this experience, but those dances signaled a food site in the direction of the wind tunnel flight and at a distance proportional to the amount of time that they had flown in the tunnel. Once again, dead reckoning only has to get a bee to the general area of a good feeding site. Pinpointing a precise place such as a single flower is quite unnecessary.

Notice another advantage of remembering only the direction and distance of each flight and then only one flight at a time. A foraging animal must exploit adventitious discoveries. The forager should return to a site only if it continues to find food there. The first time that a forager fails to find food at a site should be the last time that it returns. The forager should immediately start exploring for a new site. When a good forager finds a new site, the new distance and direction should immediately govern navigation. A Hullian forager that had to build up reinforcement, or a Tolmanian map reader that had to build up cognitive expectancy after repeatedly finding food at a site would have to return many times to the empty food site before the reinforcement or expectancy could extinguish and a fresh search could begin. Both Hullian reinforcement and Tolmanian expectancy would have negative survival value in the natural world of foraging animals.

Radial Arm Maze

In the 1970s experiments with an apparatus called a radial arm maze revived interest in the traditional concept of a cognitive map. In Fig. 12.6, the eight arms leading away from a central platform are open without walls or with very low walls to expose rat subjects to a rich array of extramaze stimuli. Typically, experimenters bait the end of each arm with a small amount of food, allowing each rat to explore the arms in any sequence. Experimenters only refill the food wells at the ends of the arms after a complete trial which lasts until a rat has made eight excursions out to the ends of the arms or when a fixed amount of time has passed, whichever happens first.

A rat forages most efficiently by exploring each arm exactly once. This is much more like natural foraging than the usual maze-learning task. Typically, rats get five trials per day in this apparatus and they soon learn to explore on average about 7.5 out of the possible 8 arms per trial. Cognitive psychologists claim that rats accomplish this by forming a cognitive map of the apparatus and the room around it.

In a critical experiment, Olton and Collison (1979) separated intramaze stimuli from extramaze stimuli by an ingenious device. They modified the apparatus so that the arms could be rotated independently of the central platform. For half of the rats Olton and Collison put the food in small wells at the end of the arms, which is the usual procedure. For the other half of the subjects Olton and Collison put the food in identical wells placed on small platforms just beyond the ends of the arms, but in easy reach of the rats. When the arms rotated, the outer platforms and the food wells on the outer platforms remained in place.

FIG. 12.6. Diagram of a typical eight-arm radial maze.

In Phase I of the experiment, which lasted for 15 daily sessions of 1 trial each, Olson and Collison allowed each rat to explore the arms and consume the food while the arms remained in place as in the usual procedure. Soon, all of the rats reached criterion performance, visiting on average about 7.5 out of the possible 8 arms on each trial. In Phase II, each time that a rat returned from an excursion to the end of an arm, the experimenters trapped it on the central platform with a system of gates and rotated the arms haphazardly one, two, three, or four times 45°, half clockwise and half counterclockwise. After each rotation the experimenter opened all gates and the rat was free to make its next excursion.

The animals that found food at the end of the arms could use intramaze stimuli to avoid repeating excursions to previously visited arms. The animals that found food on the outer platforms could use extramaze stimuli to avoid repeating excursions to previously visited places in the room. The results were decisive. The rats that found food on the outer platforms were only slightly disturbed by the rotations and quickly recovered their Phase I scores of about 7.5 out of the possible 8 radial directions. The rats that found food at the end of the arms dropped to chance performance as soon as the rotations started and remained at chance for 30 days at one trial per day after which the experiment ended. Clearly, the rats in this experiment avoided repeating radial directions and ignored intramaze stimuli such as odor cues.

Animals avoid repeating excursions to the same place in the room rather than the same arm of a radial arm maze. The most popular view of this finding attributes efficient foraging in the radial arm maze to a cognitive map of the room that forms in the brains of the animals and which they have to consult when they make choices (M. F. Brown, Rish, VonCulin, & Edberg, 1993; Poucet, 1993). In Brown et al. Experiment 5, however, rats foraged efficiently in a radial arm maze when their only extramaze cue was a small gap in a curtain hanging well beyond the arms of the maze and blocking all other extramaze cues. They could navigate with this single directional cue, just as the hamsters in Etienne et al. (1993) could navigate with a single directional light on the horizon of their experimental room. Later, M. F. Brown and Moore (1997) showed that rats could also use cues from their semicircular canals to forage in a radial arm maze completing the pattern of agreement with the results of Douglas (1966) and Etienne (1992).

Morris Water Maze

In 1981, Morris introduced a simple apparatus that virtually eliminates intramaze cues. A Morris water maze consists of a large tank filled with a milky opaque fluid. Rats must swim in the tank until they find safety on a small submerged platform. They cannot see the platform or any other cues in the tank. The milky fluid also prevents them from leaving odor trails. They must use extramaze stimuli to find the safe platform. The measures are crude, but a Morris water maze is inexpensive to build and to use and rats master the task very quickly. Consequently, it is a popular apparatus with experimenters who are interested in the effects of physiological factors such as brain lesions, drugs, and age (Poucet, 1993). It should be easy to see that rats could use both internal cues from their middle ears and distant cues on the horizon to guide themselves to the hidden platform as many animals do in the other spatial tasks discussed in this chapter. In an outdoor version of a Morris maze Kavaliers and Galea (1994) showed that meadow voles performed poorly under an overcast sky, indicating that they used celestial cues to guide them to the hidden platform.

Caching

Honeybees and hamsters store their food in large home nests. This is called larder hoarding. Many other animals store their food in small scattered caches, and this is called scatter hoarding (Vander Wall, 1990). While larder hoarders find their way between their home nests and new foraging sites, scatter hoarders have to find their way back to many small caches. Vander Wall (1991) allowed individual chipmunks to cache food at different times in the same laboratory enclosure. Later, individual chipmunks found significantly more of their own caches than the caches made by other chipmunks. Similar results were found, for example, by Vander Wall (1982) with a species of bird called Clark’s nutcrackers and by Jacobs and Liman (1991) with squirrels. These findings rule out the possibility that scatter hoarders cache food at random and then find it by searching at random and they also rule out the possibility that scatter hoarders cache food in species-specific places and then find it by searching at random in similar species-specific places.

Both in laboratory (Herz, Zanette, & Sherry, 1994; Vander Wall, 1982, 1991) and in field conditions (Barnea & Nottebohm, 1995) scatter hoarders use landmarks on the horizon to orient to their caches. Under laboratory conditions Clark’s nutcrackers also used local landmarks such as stones and pieces of wood to find particular caches (Vander Wall, 1982). Scatter hoarders, like larder hoarders, can use a combination of distant and local landmarks to find specific places without having to construct, carry with them, and then consult detailed cognitive maps.

SUMMARY

Modern ethologically based research has taken us a long way from S-R-S* reinforcement theories such as Hull’s and cognitive map theories such as Tolman’s. Both of the great schools that were so influential in the early 20th century seem backward now. Clearly, the error of both schools was their disregard for the ethology of the animals in their experiments.

Foraging animals evolved in a very different world from the world of mazes. They may have evolved to explore in a haphazard, zigzag path until they find food, but they certainly did not evolve to wait at the food site for a kindly human to take them back to home base. In nature, their problem is to get directly home from an arbitrary site. If the site is worth a return trip, they need to get directly back when they return. They certainly did not evolve to wait for a kindly human to take them back to the beginning of the exploratory path so they could try again. Under natural foraging conditions, a map of an arbitrary exploratory route is hardly worth memorizing. That animals have learned to repeat an arbitrary route after many trials is a tribute to the adaptability of these amazing creatures, but it is hardly a tribute to the biological insight of the experimenters and theorists.

The practical navigational systems that foraging animals actually use to get from place to place in the natural world are elegant in their simplicity. They are also much more economical than the cognitive maps imagined by cognitive psychologists early in the 20th century. When engineers design robots for exploring sites in space, such as the moon, they have practical tasks in mind. For example, a likely task for a robot would be to prospect for valuable minerals. A robot prospector would have to explore in a haphazard way until it found “pay dirt.” Then it would have to get back to the space ship with a load, and remember how to return to the source, perhaps recruiting other prospectors, if it found a rich source. A bottom-up design that took advantage of cheap navigational devices and local landmarks would be a practical solution, significantly more practical than a design that depended on detailed cognitive maps built up trial by trial by extensive exploration packaged together with a sophisticated cognitive map reader (see discussion of one-trial, bottom-up learning in chap. 8).