The ambiguous-cue paradigm (ACP) was originally developed to assess mechanisms underlying learning [1
], but can be adapted to assess emotional states in nonhumans (e.g., [3
]). The task is simple in its design, yet presents subjects with a discrimination that many find difficult to learn (e.g., [4
]). The paradigm presents the subject with a simultaneous discrimination task involving two pairs of stimuli. One stimulus (positive or P) is always reinforced when selected, one stimulus (negative or N) is never reinforced, and one stimulus (ambiguous or A) is reinforced depending on whether it has been paired with the positive (PA pairings) or negative stimulus (NA pairings). Although many species have demonstrated better learning of the NA compared to the PA pairing (e.g., children and the mentally disabled [5
]; rhesus monkeys (Macaca mulatta
]; pigeons (Columba livia domestica
]; European starlings (Sturnus vulgaris
]), rhesus monkeys have also displayed superior learning on PA trials [6
], as have chimpanzees (Pan troglodytes
]), and pigeons [9
]. These different patterns of learning may be informative both in terms of learning strategy and cognitive biases, which may relate to affective disposition.
NA trials may be easier to learn compared to PA trials because the PA trials present an “approach-approach conflict” [11
] as P is always reinforced and A is reinforced half of the time, leaving the subject conflicted over which reinforced stimulus to choose. However, one could argue that NA pairings involve an avoid/avoid conflict as they present subjects with one stimulus that is never reinforced and one that is reinforced only half the time. Thus, animals’ responses to the NA and PA pairings may indicate whether they attend more to cues of reward or to non-reward, and thus, might be useful in assessing their affective states. Animals that focus on cues of reward may be influenced more strongly by the approach-approach conflict and so learn NA faster (where only one stimulus has been rewarded), whereas animals that focus on cues of non-reinforcement may experience an “avoid/avoid” conflict and find PA easier to learn. Indications of focus on reward or non-reward cues may align with positive or negative affective states.
In humans, affective states can be indicated by cognitive biases that are demonstrated by the preferential processing of certain types of information, such as threatening stimuli in the case of a negative bias [13
]. The judgments of animals, just like their human counterparts, may be influenced by their emotional states [14
]. An animal that is shown to react to ambiguous stimuli when presented under novel conditions similar to the manner in which they previously responded to rewarded stimuli may be seen as behaving optimistically. Alternatively, an animal that is shown to react to ambiguous stimuli similar to the manner in which they previously responded to non-rewarded or aversive stimuli can be seen as behaving pessimistically [15
]. Pessimistic or optimistic affective states can be evoked by manipulating environmental conditions.
There is evidence that providing animals with larger, more enriched enclosures may elicit positive cognitive biases (in rats [16
]). Captive European starlings exhibited more optimistic interpretations of ambiguous stimuli after they had been housed in larger cages that also contained more enrichment items (such as water baths, perches, and bark chips) compared to when they had been housed in smaller cages lacking any additional enrichment items—although, in this case, space and enrichment were confounded [17
]. The cyclical changes in environment that captive animals encounter may be due to the seasonal changes in climate, seasonal changes in visitor numbers, or seasonal changes in husbandry routines. In Experiment 1 of the present study, three male gorillas experienced changes in the amount of space they were offered, and arguably the types of enrichment available, over the course of a year. It may be that these seasonal changes in the gorilla habitat cause similar patterns of cognitive bias as those described by Matheson et al. [17
], with the gorillas displaying more positive biases when in the outdoor (larger) habitat than when in the indoor (smaller) habitat.
In Experiment 2 of the present study, the American black bear did not experience changes to habitat size, but rather changes to visitor density, with warmer weather during the summer drawing larger crowds than cooler temperatures during the fall and winter. To date, there has been only one study to investigate cognitive biases in bears. Keen et al. [18
] tested grizzly bears (Ursus arctos horribilis
) on a novel cognitive bias task that made use of differential distribution of food rewards. The bears in this study were trained to respond differently (touch with a nose or a paw) to two different stimuli (a light grey cue card or a dark grey cue card) in return for either a large or small amount of food. Following training, the bears were exposed to enrichment items that varied in preference. After the enrichment exposure, the effects were assessed by presenting probe stimuli (intermediate shades of grey) to the bears and observing whether they responded in a manner that corresponded to larger amounts of food (optimism) or a smaller amount of food (pessimism) during training. Keen et al. [18
] did not detect any effect of enrichment type on the bears’ cognitive biases during testing. However, they did observe that when the bears spent more time engaged in anticipatory behaviors (i.e., pacing) prior to testing, they displayed positive cognitive biases (optimism). It may be that a two-hour acute exposure to enrichment items (a cow hide and a parking cone) was not sufficient to induce a lasting change in cognitive bias. It is possible that other environmental factors, such as seasonal changes, including visitor density, may have a more lasting impact on cognitive biases.
We previously used a modified version of the ambiguous cue paradigm to assess affective state in western lowland gorillas (Gorilla gorilla gorilla
) but found the training period necessary to coincide with manipulations of browse foraging enrichment [19
] too brief to allow for adequate learning of the discriminations [3
]. In the current study, we extended training of the ambiguous cue paradigm in the gorillas (Gorilla gorilla gorilla
) and applied the same paradigm to cognitive bias assessments in an American black bear (Ursus americanus
). In both cases, we used the ambiguous cue paradigm as a means to assess affective state in these subjects as part of a long-term welfare assessment. As an alternative to paradigms that present probe stimuli that are intermediate in some stimulus dimension along a continuum between reward and nonreward stimulus properties (e.g., [20
]), we used the ambiguous stimulus presented in the simultaneous discrimination task paired with a novel stimulus as a means to assess optimism and pessimism. If the ambiguous cue is chosen over the novel cue, this indicates an optimistic attitude toward a stimulus to which responses have been reinforced and non-reinforced equally often. This paradigm has the advantage that it does not involve the presentation of intermediate stimuli such that responses may simply reflect a perceptual discrimination of stimuli closer to reward and non-reward contingencies. Furthermore, performance on the training pairs indicate whether animals attend to avoid/avoid or approach/approach conflicts, and thus, can also shed light on potentially stable cognitive biases/affective dispositions in individuals.
The experiments described below present a rare opportunity to compare acquisition and mastery of the ambiguous-cue paradigm in a bear and in gorillas, given previous studies suggesting that bears perform as well as, if not better than, great apes on cognitive tasks, such as the discrimination of natural categories [21
], and quantity estimation [21
]. The capacity of bears to outperform apes in cognitive tests supports recent conjecture that foraging complexity is potentially more important than sociality in driving the evolution of certain aspects of complex cognition [27
] given that bears experience low levels of sociality but varying levels of foraging complexity whereas apes experience high levels of sociality and foraging complexity. Both species can be described as generalists that exploit a patchily distributed diet and engage in extractive foraging [29
], qualifying them as experiencing complex foraging demands. We were interested in cognitive bias at the individual level and all subjects were of interest in this regard given their unique housing situations (a bachelor group of gorillas and a solitary black bear).
4. General Discussion
We used the ACP to test cognitive bias in gorillas and an American black bear. In general, all four subjects struggled to pass criterion for training, making testing with probe trials to assess cognitive bias difficult. However, we suggest that learning patterns during training may indicate long-term affective states. The individuals tested in these experiments displayed differences in their learning of the NA and PA pairs similar individual differences displayed in other species. Chip displayed superior learning for the PA pair. In contrast, Migwan and Pende’s performance on the NA pairing was better than their performance on the PA pairing. Pende’s performance on the PA pairing was poor enough that it actually prevented him from reaching criterion for testing. Migwan and Pende displayed a pattern of learning similar to that displayed by other species in previous studies [5
]. Interestingly, like the gorillas in this experiment, rhesus macaques have shown different patterns of learning, both across studies and even within studies. The macaques in Fletcher and Garske’s [8
] experiment learned the NA pairing faster, whereas the macaques in Boyer and Polidora’s [6
] and Boyer et al.’s [7
] experiments displayed the opposite pattern. Also interesting about these earlier experiments is that experimenters were able to reverse the pattern of learning displayed by the monkeys by changing aspects of the stimuli. In Boyer et al. [7
], the macaques displayed the more typical pattern of learning NA faster than PA when using objects (three dimensional) stimuli instead of two-dimensional plaque stimuli. This aligns with a previous experiment involving the use of real objects [8
]. Boyer and Polidora [6
] were able to reverse the initial pattern of PA > NA by pretraining the monkeys using plaque stimuli with distinctive cues and then testing using stimuli with less distinctive cues. A similar result was obtained in pigeons [9
]. When considering these studies together, it becomes clear that individuals show unique patterns of learning that are not species-specific and that the methods used in these experiments may greatly influence learning patterns within individuals. The distinctiveness of the A item may control whether learning is faster in PA compared to NA trials [9
It is possible that differences in learning between the NA and PA pairings may shed some light on the affective state of these subjects. For instance, it is possible that Chip’s superior performance on the PA pairings compared to NA pairings could stem from the fact that he attended more to instances of non-reinforcement than reinforcement. If this were the case, the PA pairing would be easier to learn because A is at times non-reinforced whereas P is always reinforced (i.e., touches to P are never not rewarded). For an animal attending to non-reinforcement cues, PA is easier as there is only one cue present that is (at times) non-reinforced (A). During NA pairings, there are two cues present that, at times, are non-reinforced, making the discrimination more difficult. Typically, it is thought that individuals have an easier time learning the NA pairings as they are attending to instances of reinforcement and one stimulus in the NA pair is never rewarded (N), whereas the other is partially rewarded (A); the motivation to avoid touching N would align with the motivation to touch A. On the other hand, the PA pairing consists of two stimuli, both of which have been associated with rewards (P all of the time, A partially); the subject is faced with conflicting motivations to touch both stimuli [6
]. Therefore, it is interesting that Chip learned the PA discrimination better than the NA discrimination. It may be that animals that are pessimistic attend more to non-reinforced stimuli whereas those that are optimistic attend more to reinforced stimuli. However, if Chip’s learning pattern does lend insight into his cognitive biases, the fact that he learned PA faster (attending to non-reinforced stimuli) indicating pessimism, would contradict his performance on the test phase of this study in which he selected A more often than the novel stimulus (indicating optimism).
In contrast to Chip’s performance, Pende displayed superior learning for the NA trials compared to the PA trials. This pattern may indicate that he was attending more to the stimulus that was rewarded, and hence found the NA pairing easier to learn as N is never rewarded and A is at times rewarded. Again, for the NA pair, the motivation to avoid N aligns with the motivation to touch A, making this an easier discrimination compared to PA, in which the animal is faced with conflicting motivations to touch both stimuli. Migwan also displayed this same pattern of learning (NA > PA) suggesting that she, like Pende, attended more to cues of reinforcement than to cues of non-reinforcement. When tested with the ambiguous-novel probe pairings, Migwan displayed innate preferences for the stimuli—making interpretation of her test results problematic.
Aside from the methodological issues, another challenge with assessing changes in cognitive bias across seasons is the difficulty in determining which factor might be responsible for any observed changes. That is, effects of visitor density, weather, and metabolic state tend to be confounded. Testing Migwan at the height of visitor season also required that she be tested in the middle of the summer, whereas testing during low visitor density required testing at the beginning of the fall season. American black bears undergo seasonal changes in metabolic rate and hormone levels associated with preparing for and enduring hibernation [32
]. Including male or altered female bears in testing may be useful as they may experience less fluctuations in hormones associated with a naturally cycling female bear (although some changes are seen across the sexes as they prepare for hibernation). Testing could also be conducted under artificial seasonal patterns, where presentation of exhibit space or high visitor density could be balanced across seasons. In the future, it may also be useful to test for innate shape preferences before starting the training phase and to train to criterion on both trial types before proceeding to the testing phase. It would also be most beneficial to obtain other concurrent assessments of emotional state, such as hormone levels.