Animal Tool Use

Question 1

11. 4 Using and understanding tools

Answer 1

Tool use has been defined as ‘‘the use of an external object as a functional extension of mouth or beak, hand or claw, in the attainment of an immediate goal’’ (van Lawick-Goodall 1970). Making and using tools is often seen as a landmark in human evolution, but in fact all sorts of animals use tools (Beck 1980). Some crabs attach anenomes to their claws, where the anenomes’ stings repel the crab’s enemies. Sea otters break mollusk shells on stone ‘‘anvils’’ that they hold on their chests.

Egyptian vultures crack ostrich eggs by throwing stones at them, and chimpanzees use bunches of leaves as sponges to collect water from crevices. Some animals make the tools they use. New Caledonian crows nibble strips off the stiff edges of pandanus leaves and use them to extract insects from holes (Hunt 1996; Bluff et al. 2007; see the cover of this book for a photograph of one of these crows with a twig tool). Chimpanzees make a variety of tools from sticks, leaves, and grass (McGrew 1992; Whiten et al. 2001).

As a functional category of behavior, tool use is hazy around the edges (St Amant and Horton 2008). For instance, only from an anthropocentric viewpoint
does it make sense to distinguish one gull’s dropping stones onto mussels (tool use) from another’s dropping mussels onto stones (not tool use, see Beck 1980, 1982). Both are performing a food-reinforced chain of behavior involving stones. And to take another avian example, New Caledonian crows’ ability to choose appropriate materials for tools and manipulate them to extract food from holes is arguably less impressive than the discriminative and motor skills shown by myriads of other bird species in building their nests (Hansell 2000).

But in humans, using tools is also thought to entail at least an implicit understanding of how and why tools work, a simple ‘‘folk physics.’’ Furthermore, like early hominids, chimpanzees in different geographic areas have distinctive types and uses of tools (Whiten et al. 2001) suggesting that tool use is transmitted socially and, more controversially, that apes have a primitive form of culture (Chapter 13). Provocative terms like ‘‘folk
physics for apes’’ (Povinelli 2000) and ‘‘chimpanzee cultures’’ (Whiten et al. 2001) have attracted attention to animal tool use, and new findings about tool use by
birds have increased it.

Alongside of fieldwork, clever laboratory analogues of
situations observed in the field have been devised to allow a critical, controlled, look at two complementary questions;
(1) How is tool use acquired? Is it, as might appear to the skeptic, entirely instrumental learning by trial and error or is some form of social influence required? Is insight ever involved?
(2) What do animals understand about tools? Do they, for example, immediately recognize—in a way not explicable as stimulus generalization from past experience—what tool is needed in a given situation?

We begin, however, with a brief look at the evolution
and function of tool use.

Question 2

11.4.1 What kinds of animals use tools?

Answer 2

Hundreds of observations of primates (Reader and Laland 2002) and birds (Lefebvre, Nicolakakis, and Boire 2002) using tools have been reported, not to mention candidates from other taxa, but routine use of genuine tools by one or more populations is established for only two species of birds and a few primates.

A ‘‘genuine’’ tool is one the animal manipulates, like a stick used to extract prey from holes. ‘‘Borderline tools’’ (Lefebvre., Nicolakakis, and Boire 2002) are objects that are used but not actually manipulated, like the balls of mammal dung that burrowing owls place around their nests to attract the beetles which they eat (Levy, Duncan, and Levins 2004).

One obvious ecological prerequisite for using genuine tools to get food is that the animal rely to some extent on extractive foraging, that is, eating things that have
to be extracted from a hard shell or a hiding place inside a tree trunk or the like. Thus the fact that gorillas feed primarily on leaves probably accounts for the apparent rarity of tool use in this species compared to chimpanzees and orangutans (but see Breuer, Ndoundou-Hockemba, and Fishlock 2005).

Among birds, it is noteworthy that the two species showing most widespread use of tools for extractive foraging, New Caledonian crows and woodpecker finches, are island species. For example, the woodpecker finch of the Galapagos does with twigs and cactus spines what a woodpecker does with its bill. Presumably in a mainland habitat it would be outcompeted by species equipped with stout bills for extracting the same kinds of prey more efficiently.

The absence of competitors may open up a niche that can be exploited by evolving a behavioral specialization. Differences among populations of tool-using species also provide some clues to ecological conditions favoring the evolution of tool use. A prime example comes from capuchin monkeys (Cebus species). Although capuchins readily make and use tools in captivity, researchers in the forests of South America had seldom seen them using tools in the wild (Visalberghi and Fragaszy 2006).

But it turns out that several groups of capuchin monkeys in more arid, open, areas in northeastern Brazil routinely use stones to crack nuts (Waga et al. 2006). Some groups carry hard palm nuts to habitual ‘‘anvils,’’ where
heavy stones are left lying around; a monkey may stand up bipedally holding a stone a quarter of its own weight and drop the stone onto a nut (Fragaszy et al. 2004).

The reasons why these populations use tools so much more than the forest monkeys probably include both the scarcity of foods other than palm nuts and the fact that the low density of trees means the monkeys spend a lot of time on the ground, where a stone, a nut, and a hard surface are more likely to be encountered together and to remain together through repeated nut-cracking attempts than in a tree (Waga et al. 2006).

Mainly because tool making and using was traditionally assumed to be uniquely human, it is widely assumed to require some kind of exceptional cognitive ability. In
any case, activities such as appropriately bringing together nut, anvil, and stone tool or selecting and modifying sticks to make a tool of a required length and thickness seem unusually cognitively demanding.

These considerations suggest that tool use should be associated with enlargement of the brain or some part of it. Comparative surveys of both birds (Lefebvre, Nicolakakis, and Boire 2002) and primates (Reader
and Laland 2002) have indeed found evidence for such associations. For example, correcting for such confounds as overall frequencies of observation (see Box 2.2), genuine tool use in birds has been reported most often in corvids, other passerines, and parrots, groups among those with the largest relative neostriatum and whole brain.

‘‘Borderline’’ tool use is related more to overall innovation rate but not to brain measures. In primates, tool use, innovation (see Box 2.2), and social learning are all related to size of the ‘‘executive brain,’’ that is, neocortex and striatum. Neither of these surveys provides much insight into precisely what neural specialization, if any, is associated with tool using nor do they look at differences among closely related
species that differ in their propensity to use tools. In any case, whether or not it requires a specialized conceptual ability, using tools might not be expected to be associated with a single localized neural specialization because it can involve perceptual, motor, and/or learning abilities.

Question 3

11.4.2 What do tool users understand?

Answer 3

Understanding what? The trap tube as a case study
A foot-long horizontal transparent tube with a peanut in the middle was placed in the cage of a group of four capuchins. When sticks were provided, monkeys used them to obtain peanuts by inserting a stick into one end of the tube and pushing the peanut out (Visalberghi and Trinca 1989). To use sticks effectively, the monkeys need not have understood anything about the requirements of the situation, such as needing a stick
neither too thick nor too short and an unbroken surface between the peanut and the exit from the tube.

On the face of it, poking a stick into the tube is instrumental behavior reinforced with food and acquired through trial and error. To test whether
the animals understood anything other than ‘‘pushing the tool into the tube causes food,’’ Visalberghi and her colleagues gave the capuchins clever modifications of the tube task. In what has become a benchmark test for understanding of tools, a trap was introduced in the middle of the tube (Figure 11.16, Visalberghi and Limongelli 1994).

Now inserting the stick at the end closer to the reward (a candy in this experiment) pushed the reward into the trap. Three out of four capuchins given the trap tube never got the candy more than half the time in 140 trials. The fourth began to succeed almost every time after 90 trials. This individual was then given further tests designed to probe what it had learned. For instance, the tube was rotated so the trap was on
top.

Now the stick could be inserted on either end, but the monkey persisted in carefully selecting the end farther from the candy and frequently monitoring the
movement of the candy as she slowly slid the stick into the tube. Thus this successful animal was using a distance-based associative rule. Five captive chimpanzees were tested similarly to the capuchins (Limongelli, Boysen, and Visalberghi 1995; also see Povinelli 2000). These animals, experienced in a variety of laboratory tasks, all used sticks to get rewards from the plain tube right away.

However, in 140 trials with the trap tube, only two of them ever performed above chance, and that not until after 70–80 trials. To see whether they were using a distancebased rule, these two animals were tested with a new trap tube that had the hole displaced from the center so that inserting the stick on the end nearest the reward could push it into the trap. Both animals were successful in this task almost from the beginning, showing that they took into account the position of the reward relative to the trap.

One of the successful chimpanzees anticipated the effects of the stick on the reward, as she rarely even began by inserting it on the wrong side. The other animal was more likely to begin with the stick on the wrong side, then withdraw and reinsert it. Still, these
behaviors can be seen as reflecting learned rules based on the position of the reward relative to the hole, something like ‘‘Insert the stick on the side of the trap away from the candy’’ or ‘‘Push the stick only if it is moving the reward away from the trap.’’

By now, in addition to capuchins and chimpanzees, other great apes (Visalberghi, Fragaszy, and Savage-Rumbaugh 1995; Mulcahy, Call, and Dunbar 2005), several other primates (Santos et al. 2006), human children (e.g., Horner and Whiten 2007), woodpecker finches (Tebbich and Bshary 2004), a New Caledonian crow (Bluff et al. 2007), and—for good measure—human adults (Silva, Page, and Silva 2005; Silva and
Silva 2006) have been tested on the trap tube task or variants of it.

For example, in the ‘‘trap table task’’ (Povinelli 2000; Santos et al. 2006) an animal chooses between two
tools for pulling food across a table toward itself, one of which is positioned to draw the food into a hole (Figure 11.17). Only human adults and children above the age of
4 or 5 immediately avoid the trap (Silva and Silva 2006; Horner and Whiten 2007). Individuals of other ages and species eventually learn to avoid it but take about as
many trials as the capuchins.

As suggested in the description of the successful capuchins and chimpanzees in the original studies, animals can learn to use any of a number of cues to avoid the trap. Nothing requires that they ‘‘understand gravity’’ or even the necessity to avoid holes or make the reward slide over an unbroken surface. Beyond the trap tube. A serious problem here is that presenting successful animals with a tube with the trap
rotated to the top is a test with very limited power.

Understanding that the tube is no longer functional would be evidenced by inserting the tool at random on either end, but the same behavior would result from generalization decrement, that is, treating the altered tube as part of a new problem. More importantly, there is no cost to choosing a particular side when the trap is on top because the reward comes out regardless (Machado and Silva 2003). Indeed, unless it brings the reward sooner, human adults avoid using a tool near a nonfunctional trap, as if unthinkingly applying an algorithm ‘‘avoid traps’’ (Figure 11.17; Silva, Page, and Silva 2005; Silva and Silva 2006).

Just as with other sorts of concept learning (Chapter 6), what is needed is a test in which conceptual understanding (here of the physical causal structure of the situation) predicts an outcome opposite to that expected from reliance on familiar cues (Machado and Silva 2003). A design which does this was pioneered by Seed, Tebbich, Emery, and Clayton (2006), who trained rooks to avoid traps in the setups labeled A and
B in Figure 11.18. Because rooks do not naturally use tools, the stick ‘‘tool’’ was preinserted into a tube and the birds had only to pull on the correct end to obtain the reward.

Animals trained to a criterion of 80% correct on tube A or tube B immediately performed at a similarly high level on the other tube of the pair. However, because the trap looks the same in both cases this finding tells us only that the birds had learned to pull on the side away from the trap. Whether they had also learned something about the characteristics of traps was tested
with tubes C and D. Each of them incorporated the ‘‘nontrapping’’ sides of tubes A and B, but arranged in such a way that success required taking into account
how the reward would move when the stick was pulled.

Six of 7 birds performed at chance on 20 trials with each of these tubes, but one performed almost perfectly from the outset. What the successful rook had learned was not probed further, but a similarly
designed task was used with chimpanzees (Seed et al. 2009). Initially these animals didn’t use a tool but learned to slide a reward toward the exit from a transparent ‘‘trap box’’ by pushing it along with a finger.

Training on two tasks was followed by tests in which responding based on cues predictive of success in the initial tasks were opposed to responding based on understanding traps. The animals learned both initial tasks much more quickly than previous animals required to use tools in tasks with traps, but as with the rooks only one subject (here, out of six) showed
immediate transfer to the probe tasks. Then seven experienced animals along with eight naive animals were given a new version of the trap box to be used with a stick tool or a finger.

This turned out to be difficult for naive animals: only one, using a finger, performed above chance in 150 trials. All experienced animals using a finger solved it within 30 trials, and two who had to use a tool also solved it within the time allowed. This experiment supports two primary conclusions. One is that under some conditions chimpanzees apparently learn more than responses to arbitrary cues, as evidenced by the superior performance of the experienced animals in the final task.

They may represent something about the functional properties of such things as a solid shelf or a barrier which allows them to transfer to new tasks with the similar elements. Notice this is not the same as understanding why these properties are important in terms of gravity or the tube. Here transfer to the second set of tasks was probably facilitated by the fact that the experienced chimpanzees had already learned four versions of the trap box. As we saw in Chapter 6, a concept or category is taught by exposing subjects to multiple exemplars, and this has rarely been done in
other studies of tool use (Machado and Silva 2003).

Indeed, a study with just two tasks, a trap tube and a ‘‘trap platform,’’ found no evidence of transfer (MartinOrdas, Call, and Colmenares 2008). The second conclusion is that whether apes can succeed in avoiding traps while getting food may depend on how the task is presented. Tasks that test the same
conceptual ability from a human viewpoint may make different cognitive demands on a chimpanzee. A finger may be easier to use than a stick because attending to
movements of a stick leaves fewer resources for other aspects of the task (Seed et al. 2009).

Similarly, choosing to rake in food from the solid side of a table with a trap is very much easier when a single tool is positioned in the middle of the table and
subjects choose only where to direct it than when there is a tool on each side (as in Figure 11.17) and subjects choose which to pull (Girndt, Meier, and Call 2008).
Indeed, with a single rake apes that failed a two-rake version chose the side without the trap on about 80% of trials from the very beginning.

Again, a difference in attentional demands may be involved. When two rakes are positioned around food, the animal must inhibit its tendency to grab one before noting the relationship of the food to the trap (Girndt, Meier, and Call 2008). Finally, avoiding a trap may be easier when raking food than when pushing it out because pulling food toward oneself is more natural (Mulcahy and Call 2006b). Demonstrations that such conceptually irrelevant factors can be crucial have two
interpretations.

On the one hand, under the ‘‘right’’ conditions apes can solve tasks with traps much more readily than first appeared (e.g., in Povinelli 2000). On the other, if apes understood the task, these details should not matter so much. There are, however, some reasonable mechanistic explanations of why one versus two
tools or the need to use any tool affects performance. The next step will be to put these explanations to the test.

What makes a good tool?

Shape, size, and contact,
Less demanding and arguably more central to tool use in general than avoiding traps is to discriminate between objects that are good and bad tools in the first place (Fujita, Kuroshima, and Asai 2003), for example matching the shape, thickness and/or length of tools to task requirements. In one test of this ability, capuchins, chimpanzees, bonobos, and an orangutan were given food in a tube and sticks tied into a thick bundle or a stick with smaller sticks inserted into its ends, in an H shape (Visalberghi, Fragaszy, and Savage-Rumbaugh 1995).

All the animals untied the bundle or removed one of the small sticks from the H but only the apes appeared to do so out of an ability to anticipate the results of their actions. For example, the capuchins sometimes tried to push the whole bundle of sticks into the tube, but the apes never did. Two New Caledonian crows chose a stick from a ‘‘toolbox’’ to use for extracting meat from a transparent tube. They often took the longest stick available, which worked every time, but on trials when they did not, the length taken matched the length required pretty well.

And when accessing meat through a small hole in a tube, the crows removed twigs from a branch to make a tool of the appropriate width, only rarely trying one that was too thick (review in Bluff et al. 2007). Woodpecker finches, however, often tried a stick that was too short before taking a better one from a
‘‘toolbox,’’ but this is not so different from what they do in the wild (Tebbich and Bshary 2004). In any case, visually matching length of tool to depth of hole may have no function in the wild because the prey is usually concealed in the dark hole.

This may explain why two wild New Caledonian crows offered grubs at different depths in experimenter-made visible holes behaved much like the woodpecker finches (Hunt, Rutledge, and Gray 2006). Another test of the ability to choose good tools on the basis of their immediately perceptible characteristics is the support problem, a classic test of physical understanding first used by Piaget (see also Box 11.2). Very young children recognize that an out-of-reach object resting on (i.e., supported by) a cloth can be obtained by pulling on the cloth.

In the version for monkeys and apes with options like those depicted in Figure 11.19 the animal chooses which of two cloths to pull to obtain an apple. An effective cloth can be perceived immediately as one with an unbroken surface, however irregular, between the working end and the treat. Cotton-top tamarins learn to discriminate between broken and unbroken cloths, even when the differences between them are quite subtle (Hauser, Kralik, and Botto-Mahan 1999).

However, in tests focused on the relationship between the goal object and the cloth, chimpanzees do not immediately discriminate between cloths that actually support an object and those which simply surround or touch it, as in the examples in Figure 11.19 (Povinelli 2000). Perception of surrounding or containment is also important for discriminating between effective and ineffective hook tools or canes (Figure 11.20), a test used with chimpanzees (Povinelli 2000) and five species of monkeys and lemurs (Hauser and
Santos 2007), as well as with young children (Povinelli 2000; Cox and Smithsman 2006).

Interestingly, once animals learn to use a cane of a particular color, thickness, and material, they transfer readily to other tools with the same functional properties even if different in color and texture. For example, tamarins trained with thin blue canes choose a novel thin red cane over an ineffective blue cane (Hauser 1997). Using an object as a tool may focus attention on its functionally relevant features and/or tool users may be predisposed to attend to such features.

Normally a tool must be placed in the correct relationship with the goal object, so one might wonder whether tool-using species are better at bringing this relationship about, a skill that seems to require planning a sequence of motor acts (Cox and Smithsman 2006). This issue was addressed by testing capuchins with canes and related tools in a similar way to Hauser’s (1997) tamarins (Cummins-Sebree and
Fragaszy 2005).

Like the tamarins, the capuchins seemed indifferent to the sheer familiarity of a tool’s irrelevant features such as color and attended to its functional properties. But unlike the tamarins, which do not use tools in the wild, the capuchins sometimes chose a tool that had not been prepositioned around the food and moved it
into position. Their success with such tools improved with practice. These findings are consistent with the capuchins having some sort of specialization for tool-using, perhaps not so much a perceptual or conceptual one as a tendency to engage in certain kinds of exploratory behavior (Cummins-Sebree and Fragaszy 2005).

However, as the authors of this study recognized, more thorough comparative work is needed to control for possible differences in, for example, the animals’ past experience and their sizes relative to the tools.

Concluding remarks:

Shape, size, and orientation do not exhaust the functionally relevant features of tools. Sensitivity to a tool’s material has been examined by offering a choice between a floppy and a rigid pulling tool (Povinelli 2000; Santos et al. 2006). But unlike continuity or containment, rigidity cannot be perceived before the tool is chosen. The animal has to recall past experiences with the material, and this requirement may account for failures in such tests. Indeed, it is a mystery why the vervets and tamarins tested by Santos and colleagues (2006) generally rejected a flimsy rope for pulling in food unless some aspect of their past experience predisposed them against it.

A cotton-topped tamarin choosing a hook tool (cane) that surrounds the desired treat over an ineffective tool. After a photograph in Hauser and Santos (2007) with permission. Information about what materials animals choose for making tools and how they make those tools also potentially sheds light on the features they regard as important, though again such information cannot be interpreted without knowing the animals’ past history. For example, videos of wild chimpanzees arriving at termite nests carrying stout sticks for excavating and/or thin wands for extracting termites as the case requires (Sanz, Morgan, and Gulick 2004) are wonderfully compelling evidence
for flexible use of multiple tools and possibly even planning, but tell us nothing about how the behavior develops.

Question 4

11.4.3 How does tool use develop?

Answer 4

A provisional conclusion from the foregoing section is that using tools involves perceptual and motor skills which tool using species may be predisposed for, but
there is no evidence that it involves understanding the unseen causes by which tools work. In this section we look briefly at the development of tool using, in part to see whether this information sheds any new light on what tool users know. Three kinds of learning have been proposed to contribute to the acquisition of tool use: instrumental learning, imitation or some other form of social learning, and insight.

Of course insight implies suddenly arriving at understanding without apparent prior practice so solid evidence for it might be thought to settle the question of understanding. However, even if—as does seem to be the case—either or both of the other two mechanisms plays the major role in the development of skilled tool use, this does not rule out the possibility that animals acquire some physical understanding once they begin engaging with tools (Bluff et al. 2007).

Instrumental learning must play some role in the development of skilled tool use—it would be surprising if it did not. Animals should acquire the ways of
choosing and manipulating tools that give reward the fastest and most efficiently. An example from the laboratory is the observation that capuchins improved over sessions in placing a hook tool to rake in food (CumminsSebree and Fragaszy 2005).

But reinforcement works on species-specific predispositions to pick up potential tools like sticks and engage in other behaviors that seem to be precursors of tool use (Schiller 1957; Tebbich et al. 2001; Bluff
et al. 2007). This has been best documented in captive young woodpecker finches and New Caledonian crows. They do not need to see adults using tools to start using tools themselves, although social influence may have a role in attracting crows’ attention to particular kinds of tools. New Caledonian crows make tools by biting strips off stiff Pandanus leaves, and naive socially isolated birds show the rudiments of this behavior.

However, the presence of differently shaped tools in different parts of New Caledonia suggests that social learning plays a role in tool manufacture (see Bluff et al. 2007), a possibility we return to in Chapter 13. The role of social factors in the development of tool manufacture and use by apes and other primates is more difficult to disentangle because of their protracted development and the impossibility of ethically raising them in social
isolation. Social learning and the possibility that population-specific forms of tool using in apes are ‘‘cultural,’’ that is, socially transmitted, are discussed in
Chapter 13.

Many of the experiments on imitation and other forms of social learning in apes and monkeys have involved tool-using tasks. As we learn there, true imitation—copying an act from seeing it done—is rare in any species, but numerous other forms of social influence can help to get tool using started, after which the physical requirements of the situation can shape the behavior through individual learning.

Insight:

‘‘Aha, I’ve got it!’’ In people, this experience accompanies insight, sudden solution of a problem without apparent previous trial and error. The most famous cases of apparent insight in another species were described by Wolfgang Ko¨ hler (1959) in
the chimpanzees he studied on the island of Tenerife during World War I. They used sticks, strings, and boxes in novel ways to obtain food placed out of reach. For
instance, two sticks were joined together to rake bananas into the cage; a box was moved across the cage and used to reach fruit suspended high on the wall.

When first confronted with such a problem, animals would usually try the direct solution, jumping up and down under a suspended banana or fruitlessly reaching arms and legs between cage bars. These attempts might be abandoned and the animal might start doing something else when suddenly it would jump up, grab the necessary tool, and immediately solve the problem, as if having experienced an insight into what was
required.

Subsequent researchers have emphasized that ‘‘insightful’’ behavior does not appear immediately but may follow many ridiculously (from a human viewpoint)
incompetent failures (Povinelli 2000; Machado and Silva 2003). Moreover, experience builds it from species-typical motor patterns (Schiller 1957; review in Beck 1980). Chimpanzees spontaneously carry and climb on boxes, pull strings, play with sticks and put two sticks together. 

Such experience contributes to solving problems like Ko¨ hler’s, as does perceptual and motor maturation. Similarly, observations of birds pulling up a dangling string with food on the end reveal a central role for species-typical feeding motor patterns (Vince 1961).
Some insight into precisely how experience with the elements of a solution contributes to ‘‘insightful’’ behavior is provided by a not entirely tongue-in-cheek demonstration that pigeons can solve the banana-and-box problem (Epstein et al. 1984; see also Nakajima and Sato 1993).

The pigeons were first trained in two separate parts of
the problem. In some sessions they were reinforced with grain for climbing onto a small stationary box and pecking a facsimile of a banana, wherever it was in the
testing chamber. Jumping toward the banana was extinguished. In separate interleaved sessions the birds were trained to push the box toward a spot on the wall of the same chamber, with spot and box in varying initial locations. Control birds were trained to climb and peck the banana but did not learn to push the box toward a
target.

In the critical session, the banana was placed out of reach and the box was available in the chamber, but no spot was present. The birds trained to push directionally all behaved like Ko¨ hler’s chimpanzees: at first they stretched beneath the banana and looked back and forth between banana and box, but within a minute or
so they began to push the box into place under the banana. When the box was in place they climbed onto it and pecked the banana (Figure 11.21).

Films of the pigeons reportedly gave viewers a strong impression of humanlike thoughts and emotions (Epstein. et al. 1984), but a step-by-step analysis of the
contingencies involved shows that the behavior can be explained otherwise. Looking back and forth between banana and box at first resulted from their eliciting
conflicting responses (Epstein 1985). But because flying and jumping at the banana had already been extinguished, pushing the box quickly became the dominant response.

Because both banana and spot had been associated with grain, mediated generalization (Chapter 6) could account for the banana becoming the target of pushing when the spot was absent. Finally, the birds climbed onto the box when it was under the banana thanks to what Epstein and colleagues call automatic chaining.
By chance, pushing the box had reinstated a situation (banana within reach) supporting the previously reinforced pecking response.

Whether the comparable behavior of Ko¨ hler’s chimpanzees can be accounted for in a similar way is impossible to say, since their histories were not as thoroughly known. Epstein and colleagues’ account of the pigeons’ behavior shows how knowledge
of animals’ past history and a careful analysis of the stimuli present and the responses they elicit can account for apparently novel or insightful behavior.
One mechanism that may have played a role, resurgence, has subsequently been well documented (Lieving and Lattal 2003; Reed and Morgan 2007).

Resurgence refers to the observation that when two responses are trained in sequence such that the first is extinguished before or during training of the second, the first response reappears during extinction of the second. It is not as well recognized as a source of
flexible behavior as it deserves to be. Awareness of resurgence and automatic chaining together with a dispassionate description of actual behavior might take
some of the mystery out of other examples of apparently purposeful tool manufacture or use, including examples of using a tool to get another tool (‘‘metatools’’ Mulcahy, Call, and Dunbar 2005; Taylor et al. 2007).

One much-cited case is the observation that Betty the New Caledonian crow bent a straight wire into a hook and used it as a tool to pull a miniature bucket of meat out of a little well (Weir, Chappell, and Kacelnik 2002). The first time she did this, Betty had already used hooked wires on the bucket, but she was left with only a straight one, and she initially tried to use it. Having failed, she eventually thrust the wire at the base
of the transparent well, wedging it in in such a way that it bent when she pulled on it.

Now it apparently looked sufficiently similar to hooks she had used in the past to serve as a stimulus for lowering into the well, a successful response for getting the food. Betty was subsequently provided with straight wire on a number of occasions, on most of which she made some sort of bend and got the food, but generally not before trying with the straight wire. However, as Weir, Chappell, and Kacelnik (2002; see also Bluff et al. 2007) acknowledge, only the first trial is relevant as
possible evidence of insight or purposeful tool manufacture. Subsequent bends can all be accounted for by reinforcement history. Clearly more subjects are needed, both here and in a further test (Weir and Kacelnik 2006) in which Betty made effective tools by bending and unbending strips of aluminum.

Question 5

11.4.4 Tool use and causal understanding: Conclusions

Answer 5

Although it has been widely recognized for nearly half a century that making and using tools is not a uniquely human activity, research on nonhuman animal tool use
has not yet completely broken free of the snares of anthropomorphism (Wynne 2007a, b). The area lacks a well-developed theory of the abilities required to
recognize, use and/or make tools, leaving researchers struggling to grasp what animals understand when they use tools (as in Bluff et al. 2007).

In terms of the topic of this chapter, there is no good evidence that anything other than mechanisms of associative instrumental learning discussed in Section 11.3 underlies tool using by any nonhuman species. The role of understanding in humans can be questioned, too (Silva and Silva 2006; but see Penn and Povinelli 2007a). As discussed at length by Povinelli and various colleagues (e.g., Povinelli 2000; Vonk and Povinelli 2006; Penn and Povinelli 2007a), interpreting the world in terms of unseen causes maybe uniquely human.

But that is not to say that tool use involves no cognitive specializations. As we have seen, some animals seem quite good at recognizing the functionally relevant features of tools, and this could reflect a predisposition to perceive the affordances of certain classes of objects, if not a preexisting category of tools as a kind of object distinct from foods, landmarks, and other things (Hauser and Santos 2007).

Currently the study of animal tool using includes a rich mix of wild and captive animals, natural and contrived tests, birds and primates, species that do and do not naturally use tools. More well-controlled comparative studies could address the question of possible perceptual and representational specializations in tool-using species as well as possible convergence
between birds and primates. An example is the parallel studies of apes and corvids by Helme and colleagues (Helme et al. 2006; Helme, Clayton, and Emery 2006).

And as in the study of numerical cognition, progress might be made by better contact with theory and data from child development and an attempt to break
tool use down into components that may be shared among species to different degrees. Finally, even though making, using, and culturally transmitting information about tools may be a key component of human civilization, much tool use may not require
all that much cognitive complexity.

Even people probably learn to use most everyday tools by copying others initially and then perfecting their technique through trial and error. Any folk physics involved is mostly implicit and likely developed though
experience with complexes of related tools and tasks—using or seeing others use a hammer, a stone, the heel of a shoe to drive a nail, secure a tent peg, crack a nut, and so on. Poking a stick into a hole and getting out a grub or a candy may not require any more or different understanding than pressing a lever for a food pellet.