The authors have declared that no competing interests exist.
Conceived and designed the experiments: RB SFB LHS LP LD DP SW AIP. Performed the experiments: LHS LP LD DP JE AIP SW. Analyzed the data: LHS LP LD AIP SFB RB. Wrote the paper: RB SFB LD LP LHS. Coordinated the research and provided the resources, including funding: RB SFB TS.
The insight that animals' cognitive abilities are linked to their evolutionary history, and hence their ecology, provides the framework for the comparative approach. Despite primates renowned dietary complexity and social cognition, including cooperative abilities, we here demonstrate that cleaner wrasse outperform three primate species, capuchin monkeys, chimpanzees and orang-utans, in a foraging task involving a choice between two actions, both of which yield identical immediate rewards, but only one of which yields an additional delayed reward. The foraging task decisions involve partner choice in cleaners: they must service visiting client reef fish before resident clients to access both; otherwise the former switch to a different cleaner. Wild caught adult, but not juvenile, cleaners learned to solve the task quickly and relearned the task when it was reversed. The majority of primates failed to perform above chance after 100 trials, which is in sharp contrast to previous studies showing that primates easily learn to choose an action that yields immediate double rewards compared to an alternative action. In conclusion, the adult cleaners' ability to choose a superior action with initially neutral consequences is likely due to repeated exposure in nature, which leads to specific learned optimal foraging decision rules.
The ecological approach to cognition proposes that a species' ability to solve a particular problem is tightly linked to its evolutionary history and, hence, to the ecological conditions under which it was selected
Here, we provide the first test of the hypothesis that cleaner wrasse foraging decisions are the result of specific cognitive abilities. Our laboratory experiment involved two identical food sources – two plates differing in colour and patterns to allow discrimination, but providing exactly the same food - where one source (plate) was ephemeral and the other one permanent. This mimicked the simultaneous visit of a resident and a visitor to the cleaning station. Accordingly, the food maximizing solution involved eating from the ephemeral food source first and only then from the permanent one. The potential difficulty of the task is due to the fact that no matter which plate an individual chooses first, it will receive exactly the same immediate reward, and only then will it (possibly) have the chance to perform a second act that would yield an additional reward. Thus, the initial decision may not lead to reinforcement learning unless an animal is somehow able to integrate the future consequences into its immediate decision. Despite theoretical considerations indicating that the task is not trivial to solve, a previous study suggested that cleaners could quickly solve it, though individual learning was not investigated
An important aspect of the ecological approach is to test whether other species that do not engage in cleaning interactions are less able to solve the task. We decided to use primates – capuchin monkeys, chimpanzees and orang-utans – for the comparison for several reasons. First, the general circumstances of the cleaners' decisions involve social interactions and foraging, which matches the two contexts that have been proposed to select for large brains in primates
Although evidence suggests that the primates will excel in tasks that involve future consequences in the context of cooperation and foraging, the specifics of our task may favor cleaners. For example, cooperation and foraging are intertwined in cleaners in a way that is absent in primates; most importantly, cleaners cooperate with their food sources. In addition, primates encounter ephemeral food sources (e.g., insects, small vertebrates) unpredictably and opportunistically, and thus the ecological constraints are quite different from those of the fish, for whom the interaction with ephemeral sources is predictable. Based on this, we predicted that unlike the cleaner wrasse, the primates would not perceive the task as a social interaction but just as an optimal foraging task. Thus, our experiment offered us the opportunity to test the ecological intelligence hypothesis in a quite specific way. We expect that if ecology is the driving force that helps to solve the problem, then cleaners should individually learn to solve the tasks faster than any of the primate species. Conversely, if the general context and brain size (relative or absolute) prepare better for the task than rather specific ecological conditions do, then the primates should learn to solve the task faster than the cleaner wrasse. We also considered an additional way to test the role of learning for the cleaners' decision making process, reversing the role of the two plates once an individual reached the learning criterion. The former permanent plate now became the ephemeral plate and vice versa. Although cleaners are able to discriminate between different client categories, including resident and visitor, and can even individually recognize clients
All six adult cleaner fish individuals learned to eat first from the ephemeral plate, which was smoothly withdrawn if the cleaner were to forage on the permanent plate first. Individuals took 3–10 sessions (of 10 trials each) to reach the criterion of significance with a median of 4.5 sessions. In contrast to the adult cleaners, only one juvenile cleaner and two out of four chimpanzees solved the task within 10 sessions, and all other subjects failed (
Dots represent an individual. The y-axis indicates the number of trials required to learn the task.
For this component of the experiment, the previously ephemeral plate/tray became the permanent plate/tray, and vice versa. All the adult fish developed a significant preference for the new ephemeral plate within 10 sessions (median: 7; ranging: 6–9;
Again, dots represent an individual, and the y-axis indicates the number of trials required to reverse the preference.
A key conclusion from our experiment is that the sophisticated foraging decisions which cleaner wrasses demonstrate during interactions with client reef fish are not easily achieved by other species with larger and more complexly organized brains. The ability to choose between an ephemeral and a more permanent food source of otherwise identical quality is apparently far from simple as the vast majority of individuals from three primate species that otherwise excel in cognitive tasks failed to learn the task within 100 trials, as did juvenile cleaners. However adult cleaners consistently solved the task. Thus, our task differs from experiments that demonstrate extremely fast learning of solutions if individuals are placed into a key stimulus-response context, in which even invertebrates like bees may outperform primates, including humans
When confronted with a choice that directly yields two different amounts of food primates can easily discriminate outcomes with one reward from those with two
Second, it is possible that the fish experienced the removal of the plate as a stronger punishment than did the primates. Both the fish and the primates presumably reacted to the removal of the second plate, containing food, as a negative reinforcer (e.g., punishment). However, fish may have additionally experienced it as a social punishment; one indication that they indeed perceive the task as a cleaning situation is that they respond with tactile stimulation when the plate returns, a behaviour cleaners use to reconcile and to make clients stay longer under natural conditions
Finally, a more cognitive mechanism than associative learning that would allow subjects to solve the task is insight based on backwards induction. In backwards induction, one has to start with the desired endpoint and then figure out which steps lead to that endpoint. Evidence for backwards induction has been demonstrated in a chimpanzee, Julia, who had to open up to 10 Plexiglas boxes with specific tools inside in the right sequence to finally obtain food in the last box
We finally note that the apes' unexpectedly low performance on the reversal task was likely due to frustration with the procedure. Apes – including some of these subjects – are typically very good at reversal learning tasks
We propose two non-mutually exclusive explanations for why adult cleaners learned to solve the task. First, the cleaners may have developed the decision rule to preferentially approach ephemeral food under natural conditions and then applied the same rule to this task. In contrast, the primates were born in captivity, where sufficient food is provided multiple times per day (at all facilities) and they rarely catch ephemeral food like invertebrates. Second, as discussed above, the cleaners may have perceived the task as a social interaction. In that case they would have perceived the removal of the ephemeral plate as the loss of a cooperation partner and hence as a negative reinforcer that reduced the likelihood that the subject would choose the permanent plate again on future trials. The aversion to losing any client would make the ephemeral plate more attractive to cleaners, whereas primates are not selected to experience either the negative reinforcement of a missed opportunity or social reinforcement for interacting with their foraging substrate. Thus, we consider it likely that cleaners, but not the primates, simultaneously experienced a positive and a negative reinforcer, which would explain why they learned to solve the task rather quickly as compared to the primates. If that was the case, a change in protocol for the primates that let them perceive the interaction as social (for example by replacing the trays with human partner) should yield much faster learning.
If our hypotheses are correct then one would also predict that even individuals of the closely related cleaner wrasse species
There are various potential explanations for why juveniles failed to solve the task while adults managed. One possibility is that maturation processes in the brain preclude juveniles from solving the problem at hand. Second, there were small differences in the experimental protocol due to different research sites and in turn testing possibilities: juveniles – kept on Lizard Island for the period of the experiment - experienced longer time intervals between subsequent trials as compared to the adults which where housed and tested in Neuchâtel, Switzerland. However, in an earlier experiment adult cleaner fish that were trained on a similar task (i.e. “one plate remains until inspected while the other does not”, p. 132), but with 30 min intervals between trials, significantly chose to first clean the plate that would not wait until being inspected
While maturation and (to a lesser extent if at all) experimental design may have affected the results, we consider it likely that individual experience plays a major role; juveniles have fewer visiting clients and are therefore rarely in a situation that calls for this discrimination. The situation changes for adults; in a field study, adults had to make choices between a visitor and a resident client more than twice per hour (120 times in 52 hours of observation
We note that the primates consistently performed poorly despite the fact that we ultimately adapted the methods to be as appropriate as possible for each species (within the constraints of using trays with different ‘behaviors’ to present identical foods). In particular, the capuchin monkeys received several different methodologies as we attempted to optimize a protocol which allowed them to eventually solve the initial task. Variables that seemed to have helped them included a barrier between the plates and much shorter time intervals between trials (
Adult wrasse | Juv. wrasse | Chimpanzees | Orang-utans | Capuchins | |
|
|||||
N individuals | 6 | 7 | 4 | 4 | 8 |
Date | 3–4/09 | 7–8/10 | 8/10–4/11 | 8/10–4/11 | 8–12/09 |
Location | Neuchâtel, CH | Lizard Island, AUS | LRC GSU, USA | Zoo Atlanta, USA | LRC GSU, USA |
|
|||||
Time between trials | 15 min | 30 min | 90 sec | 1. 90 sec | 1. 15 min |
2. 30 sec | 2. 5 min | ||||
Subject order | varied | varied | varied | varied | varied |
Plate color | red-yellow | green-grey | blue | blue | green-blue violet-yellow blue-yellow |
green-white | pink-grey | yellow | yellow | ||
Plate side counterbalanced | yes | yes | yes | yes | 1. no |
2.yes | |||||
Food type | mashed prawn | mashed prawn | banana | cheerios cereal | apple |
Plate preference test | yes | no | yes | yes | no |
Food already on plate/tray | yes | yes | yes | no | yes |
Removed plate/tray | out of view | out of view | out of view | out of view | 1. visible |
2.out of view | |||||
|
|||||
Maximum N sessions | 10 | 10 | 10 | 10 | 10 |
Trials per session | 10 | 10 | 10 | 10 | 10 |
N sessions per day | 2 | 2 | 1 | 1 | 1 |
N test days per week | 7 | 7 | 3 | 5 | 3 |
|
|||||
Maximum N sessions | 10 | - | 10 | 10 | 10 |
Trials per session | 10 | - | 10 | 10 | 10 |
|
|||||
reload ephemeral tray 10× per trial | Counterbalance, tray shape/color | ||||
shorter time intervals | |||||
Cardboard barrier between plates |
The capuchins were all tested prior to the tests with the apes, and as much as possible we used the final capuchin protocol for both ape species. The choice tray featured a divider, and non-chosen options were immediately removed from sight. One thing that we could not do similarly was the five minute ITI. In pilot testing, the chimpanzees and orang-utans reacted with extreme frustration to a 5 minute delay, leaving the testing area and refusing to return. Thus we shortened the ITI to 90 seconds to encourage subjects to participate. If, as we found, the capuchins' behaviour was positively influenced by the shorter ITI, then a shorter one yet for the apes should have made the task easier. Additionally, while this might have resulted in less cost per choice, subjects still only receive 10 trials per session, so there were very few chances to receive treats during testing (in most cognitive tests, subjects receive at least two to three times this many trials in a session). Overall, this meant that the details of the procedure were optimized during the course of the study for the primates, but not the fish. Therefore, we consider it unlikely that cleaners outperformed primates due to advantages with respect to methodological details like the color or shape of trays, the food, or the inter-trial interval.
Finally, note that the primates acquired food by reaching out and grasping it, while fish swam to different foods and took them directly into their mouths. This was due to differences in body plan between fish and primates. Fish have to move between compartments with their whole body, but from where they were located could easily see both rewards simultaneously. Due to the size of the primates and caging constraints, it was impossible to house them such that they could simultaneously see both rewards and be housed in a third room. This would be particularly problematic for our study if they could not immediately view the ephemeral reward being removed when they chose the permanent reward first. Moreover, this procedure would have meant that the primates had far longer time intervals to both access the first reward and between the first and second. Additionally, primates typically make choices by grasping with their hands. Of course, when comparing species with very different body plans and abilities, identical procedures may be difficult or impossible, both for practical reasons (e.g., the presence or absence of hands) and differences in experience or ways of interacting with the world. In particular in cases such as ours, in which a species performs differently than expected, we encourage the use of multiple procedures in an effort to optimize the design for the species, even if this results in some methodological differences.
In conclusion, our results provide the first evidence that cleaners' sophisticated behavior in cleaning interactions is due to selection for specific rule learning that require experience and/or maturation. All three primate species have a complex diet and are known to cooperate, but still they were outclassed by adult cleaners in this foraging task. Although we cannot entirely rule out differences in procedure that resulted from the comparison between very different species, a possible mechanism underlying the fishes' response is that they perceive the leaving of a food source as a negative reinforcer, and therefore choose the ephemeral food source first before approaching the permanent one. This implies that the specificity of the cleaners' ability to give priority to ephemeral food sources lies not in a sophisticated cognitive process but in the ability to identify relevant stimuli. Nevertheless, recent research on fishes has yielded evidence for various supposedly more complex cognitive abilities (reviewed by
Experiments on adult cleaner wrasse were carried out in March and April 2009 at the University of Neuchâtel, Switzerland, while juvenile cleaner wrasse were tested at Lizard Island Research Station, Australia, in July and August 2010. Experiments on capuchins (August to December 2009) and chimpanzees (August to December 2010) were carried out at the Language Research Center, Georgia State University, USA, and orang-utans were tested at Zoo Atlanta, USA (August to December 2010).
Six adult wild caught
All primates were captive born. The eight brown adult capuchin monkeys (5 males, 3 females, age range 5–20 years, median age of 10 years) were from two separately housed social groups at the Language Research Center of Georgia State University, USA. The four chimpanzees (2 males, 2 females, age range 25–40 years) were also from the Language Research Center, whereas the four orang-utans (3 males, 1 female, age range 7–33 years) were from Zoo Atlanta, USA. All primates lived in stable social groups consisting of adult male(s) and female(s) and any attendant offspring. They were separated from these groups only for behavioral and cognitive testing. Details regarding housing conditions are provided in Information S1. Subjects were fed a diet according to their species-specific needs, but generally consisting of primate chow and fresh fruits and vegetables. They also received enrichment-foods several times per day; consequently, animals were never food or water deprived for testing purposes. Running water was available
The experimental design was based on a study by Bshary & Grutter
All cleaner wrasse were tested in their aquarium. A separation with a central sliding door was introduced at approximately two-thirds of the aquarium length to create an ‘experimental’ compartment and a ‘resting’ compartment. For cleaners, a given trial started by confining the subject to the smaller ‘resting’ compartment of the aquarium. After approximately 60 s, the client plates were placed at equal heights at the opposite end of the aquarium, i.e. the experimental compartment. After about 10 s, the door was opened and the cleaner could enter the experimental compartment at will.
For capuchin monkeys, members of each social group, consisting of four subjects, were simultaneously tested in separated test chambers attached to their home enclosure. Monkeys had previously been trained to be separated from their social group and to individually enter these chambers, where virtually all testing was done. Dependent offspring were always allowed into the testing area with their mothers. Testing chambers measured 61×44.5×33 cm3 in size and were separated from each other by approximately 40 cm. The test chamber was backed by an opaque panel, allowing vocal, but no visual or tactile, access to their group. This allowed us to interact with subjects in a controlled manner with minimal distractions from the group. The sessions for the apes were organized in a similar way: subjects were all tested in a subsection of the indoor section of their home enclosures, while still in auditory and visible contact with the other group members (this is how all testing is done at these facilities). The order in which subjects were tested varied from day to day. As with the capuchins, dependent offspring were always allowed into the testing area with their mothers. Note that for all species, acquiring the food required accessing a separate area from where the subjects were initially located. For fish, this required swimming, while for the primates this required reaching outside of the compartment.
The position of the two plates was randomized, but with an equal number of presentations on each side within each 10 trial sessions. Randomization was constrained such that the same tray was never presented more than three consecutive times on the same side. (Note that capuchins were initially tested with the plates altering sides between sessions; see
We based our significance criterion on Sign-Tests-Table (two-tailed). Significance was reached when a subject made correct choices on ≥9/10 trials on one session or ≥8/10 on two or ≥7/10 trials on three consecutive sessions. For capuchins in the initial sessions, the criterion for learning was ≥16/20 trials (e.g., over 2 sessions) because plate positions were constant during a session and hence a side bias would have led to the inaccurate assessment of significant “learning” in half of sessions (and significant “anti-learning” in the others). Once an individual had reached criterion, we ran the reversal trials. We used the same criterion for the reversal test. We were primarily interested in relative performance rather than the question whether all subjects can learn to develop a food maximizing preference eventually if given sufficient opportunity. Adult cleaner wrasses were the first to be tested out of all the experimental groups. They formed the baseline for the others with respect to the questions we attempted to answer. As all of them learned to solve both the initial and reversal tasks within 100 trials, we fixed 100 trials as an upper limit for the other experimental groups. Because the reversal learning task required learning of the initial task, we decided to expose any primate that failed to learn the initial task within 100 trials to modified versions of the task. The modifications were designed to facilitate learning (see
(DOC)
We kindly thank all of the individuals who helped to make this research possible. This includes first the staff at Lizard Island Research Station, Karen Cheney and Lexa Grutter for logistical support; J. Welsh, C. Lefevre and A. Vail for field support; second the staff at the Language Research Center of GSU; and third the orang-utan care staff at Zoo Atlanta.