The Gap

by Thomas Suddendorf

  I noted that the IQ testing community regards three components as essential to intelligence: to learn from experience, to adapt to the environment, and to reflect on one’s own performance. Many nonhuman animals meet the first two parts of this consensus definition of intelligence: they learn and adapt. Predation, for instance, would have exerted strong selection pressures on these capacities; consider orcas or lions hunting in packs and their prey trying to avoid being caught. On the other hand, the third component, reflection, may point to something uniquely human. Embedded thinking, or to think about thinking, may well set humans apart.

  There is little to suggest that other animals reflect on their minds. One line of research, however, raises the possibility of some degree of metacognition in nonhuman animals. The comparative psychologist J. David Smith and colleagues found that a dolphin could discriminate between a high-pitch and a low-pitch tone but increasingly hesitated when the two tones were similar in frequency. Given an option to decline trials, the dolphin did so preferentially when the stimuli were similar and the risk of error was high, suggesting some awareness of uncertainty. Subsequent studies, primarily with monkeys, have confirmed that following extensive training on simple discrimination tasks, some animals can eventually learn to avoid trials they are likely to fail.

  One way to describe such behavior is to say these animals know when they do not know. Several leaner interpretations have been ruled out through careful experimental study, but this is not to say that a rich meta-representational interpretation is necessarily correct. Smith and colleagues try to claim the middle ground by arguing that some animals are capable of uncertainty monitoring and so deliberately select to decline difficult trials.8 This is more than associative models predict but need not involve deep reflection on one’s inner mental life. We saw in the previous chapter that animals have so far failed to provide compelling evidence for representing others’ mental representations. Absence of nested thought, of meta-representation and recursion, might severely constrain the flexibility of animal intelligence.

  As a hunt illustrates, nonhuman animals can pursue goals in the face of obstacles and so also meet part of Pinker’s definition of intelligence. The complexity of goals may be restricted (especially in light of limits in mental time travel), but they evidently can pursue certain objectives. What is less clear is whether they employ reason in their pursuits. Most cases of apparent reasoning have attracted both rich and lean interpretations; the resulting debates are complex and multifaceted. I will not provide a comprehensive review of this large literature, but I have selected key examples in the hope of conveying the current state of science on the following question: Can animals solve problems rationally?

  Perhaps the most famous case of animal intelligent problem solving comes from Wolfgang Köhler’s classic experiments on chimpanzees during World War I. Köhler, the German gestalt psychologist, was on Tenerife at the time—it has been rumored that he might have been a spy—and conducted experiments he described in his influential book titled The Mentality of Apes. In his studies he presented a group of captive chimpanzees with various puzzles. For instance, he attached bananas to the ceiling of their enclosure and observed how the chimpanzees stacked boxes to reach the treat. He would put bananas out of reach outside the enclosure, and they would connect sticks to rake them in. Köhler’s star pupil was a male chimpanzee called Sultan. Köhler argued that Sultan considered the situation until hitting on a solution, solving the problem through insight rather than associative learning.

  Given these early successes, it is surprising that subsequent research has produced mixed results. Though there are some remarkable cases of problem solving described in the literature, ape behavior frequently appears to be much more haphazard than the word “insight” suggests. Great apes do not typically sit still and then rapidly enact a flawless solution. Instead, there is often a great deal of trial and error. There are reports of astounding and persistent failures to solve simple problems, and there seem to be a lot of individual differences in the ability to find solutions. Still, the basic observation that some chimpanzees, as well as other great apes, can figure out solutions where most other animals are stumped has frequently been reported.

  In one study gorillas and orangutans selected tools of the required length to reach a reward. They picked the correct tool even when the problem was out of sight, suggesting that they mentally represented what was required. In another experiment these same apes also could obtain one tool in order to get at another tool that was subsequently used to reach the food. Such “meta-tool” use may be a precursor to using tools to make tools.

  It used to be thought that only humans make tools. This skill is rare indeed, but research has established that at least a few other species manufacture tools: great apes, elephants, woodpecker finches, and New Caledonian crows. I recently visited the research station on the island of Mare, where Gavin Hunt, Russell Gray, Alex Taylor, and their colleagues were conducting their research on these crows. There I saw a crow inspect a baited hole, fly to a nearby pandanus bush, cut and rip a sliver of the barbed leaf, and then insert it into the hole to retrieve the food. When the treat could not be reached, it flew back to make a longer tool. The researchers showed that these birds can also use a tool to obtain another tool. Crows had to retrieve a short stick tool to reach a longer stick tool, which they in turn deployed to get food.

  Crows are part of a family of birds known as corvids, which also includes magpies, ravens, and jays. The comparative psychologists Nathan Emery and Nicola Clayton have argued that in a range of different domains corvids have capacities quite similar to those of great apes. Ravens, for example, are capable of solving problems such as pulling up strings to get whatever is attached to them. When faced with parallel slanted strings, they consistently pull up the one that has food on it, and ignore the other. Some researchers argue that they therefore have demonstrated insight into the causal relations of connectivity.

  Yet there remains some lingering skepticism about these cases of animal problem solving. As we have seen repeatedly, behavior that looks smart need not necessarily be the result of intelligent thought. The Clever Hans effect is a persistent problem for rich interpretations of animal behavior. For instance, the string pulling behavior of corvids may mean that they understand the connection of the string and the food, but their behavior can also be explained in terms of simple associative learning, since with each pull the bird is rewarded by the food moving closer. Taylor and colleagues recently conducted experiments that manipulated what the crows can see when they pull up food. When they cannot see the food getting closer with each pull, they stop pulling. When they can see it, even through a mirror, they continue to pull. Disrupting the visual feedback disrupts the action. This suggests that the crows do not have insight into the problem but rather act on immediate reinforcement.

  Although one may question how powerful associative learning really is,9 we always need to ask whether leaner explanations might be responsible for observed behavior. Daniel Povinelli, the major advocate of killjoy interpretations of ape theory of mind–like behaviors, failed to find support for insight and causal understanding even in chimpanzees. He found that, in spite of making and using tools in the wild, apes have limited understanding of the functional properties of tools. In Povinelli’s studies chimpanzees were equally likely to choose to pull on a string that lies on top of a banana as to choose a string that was tied around a banana. They performed poorly when given the option between a rake with a floppy end and one with a solid end in their attempts to rake in food. They made the most elementary mistakes, and Povinelli concluded that they simply cannot reason about abstract causal forces such as gravity or support. Instead, they learn associations between observable events.

  But when presented with choices between natural stick tools, orangutans recently solved some of Povinelli’s connectivity problems. Several other results suggest that his conclusions were premature. In one ingenious study, orangutans us
ed water to solve a problem creatively. Presented with a tube that had a peanut inside that they could not reach directly, the orangutans took water in their mouths and returned repeatedly to spit it into the tube until the peanut floated to the top and became accessible. Several other apes tested subsequently had problems finding this solution. Rooks, corvids that are not known to use tools in the wild, have recently been shown to be able to manipulate water levels in similarly clever ways as described in Aesop’s fable of the crow and the pitcher. Presented with floating worms in a vessel they spontaneously put stones in, thereby raising the water level until the food could be reached. New Caledonian crows can learn to do this as well; they select functional objects among distractors. Perhaps, then, some corvids and great apes can solve problems through insight after all—at least some of the time and under some circumstances.

  The so-called trap tube task is particularly revealing. The animal has to insert a stick into a Plexiglas tube to push out a morsel of food. The trick is that there is a trap on one side of the tube, such that when you push the tool in, say, from left to right, the food falls in the trap; when you push it from right to left, it comes out of the apparatus and can be consumed. After ninety trials only one of four capuchin monkeys learned to push the food away from the trap. Yet when the tube was turned around, even this monkey failed, suggesting that these primates had no understanding of the simple causal relations involved. Chimpanzees fared slightly better. However, they pushed in the same way regardless of whether the trap was at the bottom or top of the tube, even though when the trap is oriented upwards, gravity dictates that food will not fall into it. Follow-up studies suggest that the majority of great apes fail these tasks, although some can learn to solve them. Nevertheless, even the successful ones fail when there are small changes to the setup—such as when they need to push out rather than rake in the food.

  Povinelli and his colleagues Penn and Holyoak therefore maintain that great apes do not comprehend the analogical similarity between perceptually different but functionally equivalent tasks. In fact, these authors propose that this is what essentially separates human from animal minds: only humans form “higher-order relations between relations.” The idea is that only humans construct theories about the underlying causal mechanisms that govern the world. This lean account of animal analogical capacities—you guessed it—has been challenged. When presented with a new version of the trap tube task that did not involve a tool, some chimpanzees can avoid the trap and can transfer competence to analog versions in some circumstances. In fact, even New Caledonian crows were recently shown to pass the standard task and to transfer it to other versions that had distinct perceptual cues (although they had trouble when the trap itself was altered).10 Thus great apes and these crows, at least, appear to have some capacity to reason about causal relations and to transfer their insight. There is other evidence that chimpanzees may, after all, reason by analogy.

  The psychologist David Premack taught chimpanzees to place a plastic symbol “same” between two oranges and “different” between a banana and an apple. The animals could then transfer this to other pairs of objects that were either like each other or not. The chimpanzees could solve analogies such as small triangle is to large triangle as small square is to large square.11 One study also found that a chimpanzee understood functional analogies such as “a can opener is to a can as a key is to a lock.” The link here is not a perceptual equivalence but the sharing of an equivalent goal: opening. This argues strongly against the argument by Povinelli and colleagues about what makes human cognition unique. This result has not yet been replicated, so debates between rich and lean interpretations continue.

  Note that even those researchers who argue for analogical reasoning in great apes report stark individual differences and limits in capacity. The problem of inconsistent performance also emerges in other research on reasoning. Consider one more series of recent experiments. Imagine I had one treat and put it in either my left or my right hand. If I show you that my left hand is empty, you can infer, assuming there is no trickery, that the treat therefore is in the right hand. There is some evidence that great apes can make such spontaneous inferences. Josep Call put food in one of two tubes. When the tested apes looked in one tube and discovered it empty, they sometimes instantly retrieved the food from the second tube without looking into it (see Figure 7.1.). But they did not do it often, possibly because there is little cost in peeking into both tubes. A better test, therefore, may be to give animals a forced choice.

  Call placed food in one of two cups. He then shook both cups in turn, with the baited cup making the telltale sound. Great apes, when subsequently given the choice, tend to select the cup with the food. But surprisingly, only a minority of the tested apes (nine of twenty-four) did so reliably. The others often selected the cup that did not make a sound. The successful animals were then given follow-up tests to examine if they can reason by exclusion. The first test was again very simple. One of two cups is baited, but this time only one cup is shaken. If it makes a sound, it obviously contains the food. If it does not make a sound, reasoning by exclusion indicates that the other cup must have the treat inside. Three of the remaining nine apes reliably selected the other cup when the empty cup was shaken.

  FIGURE 7.1.

  Andrew Hill playing the tubes task with the female orangutan Punya (photo Emma Collier-Baker). One orangutan and two chimpanzees showed signs of spontaneous inferences by exclusions in this replication of Josep Call’s study.

  Did at least these three apes understand, or did they simply use the sound as an associative cue without understanding? To rule out such a simple explanation, Call introduced a series of clever tests. He held a tape recorder over the cups and pressed play to produce the sound when held over the baited cup. In this study, most apes did not select the baited cup more often than expected by chance. Their choice was therefore not driven by a simple association between sound and food. This suggests that in the previous study, they were not simply acting on the basis of such an association. Nonetheless, of all of the animals tested, only one ape, a gorilla, performed across the board in a way consistent with inferential reasoning; the others did not. So again, while there is evidence for more than trial-and-error learning, performance is inconsistent.

  My PhD student Andrew Hill followed up this finding by testing twenty chimpanzees, orangutans, and small apes. Again, most of them did not perform well. Yet two chimpanzees selected in ways that were entirely consistent with an inferential reasoning explanation.12 On current evidence, then, reasoning by exclusion is not a uniquely human trait. Still, the difficulties most apes have with these simplest of inferences highlights substantial differences between humans and our closest relatives. It remains unclear what exactly the nature of their capacities are.

  The classic debate between rich and lean interpretations of animal behavior typically boils down to the two options of either associative learning without insight or insightful logic and reasoning like humans. In reality, as these examples strongly suggest, this is a misleading simplification. Showing that a species behaves in ways that cannot be explained by trial and error learning does not mean that they necessarily reason like a human being. We have seen how limited and inconsistent their solutions are. Conversely, if animal behavior is not driven by humanlike reasoning, it does not immediately follow that it must be the result of some “mindless” associative learning alone. Species differ in their problem-solving capacities, and these accounts fail to explain why this is so. In fact, the same animals are often good at learning one thing but not at all good at learning another.13 Such findings suggest that animals do not simply share an all-purpose learning machinery, as was once imagined by behaviorists.

  Different animal species have evolved a variety of cognitive means to solve problems. Even if they do not reason exactly like humans, they may still have a range of solution mechanisms over and above simple trial-and-error learning. They may be prepared to learn some causal rel
ationships but not others. They may be able to attend to critical information while doing one thing but not another. And so forth. The challenge for comparative psychologists is to move beyond the simple dichotomy of rational agent versus associative machine, and try to map the diversity of cognitive abilities that exist in the natural world. My task here is to look at what sets human minds apart. But that does not mean that all animals have the same capacities—they do not.

  On current evidence we can conclude that across all the areas of problem solving discussed, the blanket statement that animals cannot reason can be rejected. Some animals can reason sometimes, under some circumstances. Nonetheless, there appear to be profound limits to their reasoning abilities. Even in the most convincing demonstrations, their performances are inconsistent. There is no sign yet of the construction of explicit theories that describe the relationship of forces. Some animals make tools, but none so far seem to design and refine tools by assembling various components and with various functions. Without embedded scenario building, without the benefits of human mental time travel, theory of mind, and language, it would not be surprising that their capacities for reasoning are limited even in the simplest of tasks. A key potential constraining factor is working-memory capacity.