Our measure of interest was whether females ate more of the flake items (i.e., cheated less quickly) when the male had perceptual access than when he did not. Consistent with our predictions, females ate fewer flake items in the male not visible condition (Mean = 3.6 items, SD = 3.2) than in the male visible condition (Mean = 4.5, SD = 3.1; Fig. 2A; for flake items eaten by pairs see Supplementary Fig. S1). Additionally, females tended to cheat less over rounds (Supplementary Fig. S2). Indeed, the number of flake items eaten was predicted by both conditions (LRT, X21 = 8.13, p = 0.004; Fig. 2A; Supplementary Table S1) and round (LRT, X21 = 12.46, p < 0.001). In a separate model, we found that the interaction between condition and round was not a significant predictor (LRT, X21 = 0.25, p = 0.6). Thus, like non-human primates6,7, cleaners cheat more when others cannot see them.
Results from Study 1 (N = 12 pairs of fish). A Boxplot showing flake items eaten by females across conditions along with raw data (For flake items eaten by pairs see Supplementary Fig. S1). Box shows the interquartile range, the horizontal line shows the median, the whiskers stretch from the minimum to maximum and raw data are overlaid as dots colored by condition. B Plot showing the relationship between female strategic cheating score and male punishment.
We also explored the role of punishment in this effect. We found that punishment did not vary as a function of the interaction between condition and female cooperation, measured as the number of flake items eaten (LRT, X21 = 0.11, p = 0.7) and indeed the full model containing this term was no better than the null. However, when this interaction term was removed, we found that, consistent with past work24, punishment was negatively related to female cooperation. Regardless of whether or not they could see the female, males were less likely to punish the more flake items females ate (LRT, X21 = 4.18, p = 0.04; Supplementary Table S1; Supplementary Fig. S3). That males punished regardless of whether they had visual access to females’ eating suggests that they may punish not merely in response to seeing a female cheat but in response to some aspect of females’ behavior following cheating.
To further unpack the relationship between males’ punishment and female cheating, we looked for individual differences in females’ propensity to cheat based on male perceptual access. To this end, we created a “strategic cheating score” by subtracting the flake items eaten in the male not visible condition from those eaten in the “male visible” condition. This measure revealed that the majority of females cooperated more in the perceptual access, male visible, condition than in the no perceptual access, male not visible, condition (Fig. 2B), suggesting the condition difference reported above appears at the individual-level as well (see also Supplementary Fig. S1 for individual data). The strategic cheating score also allowed us to test whether females that cheated more often when unobserved were those who benefited most from such adjustment—namely those paired with a more punitive male. Males naturally vary in how much they punish females27 and it would, therefore, benefit individual females to be more strategic (i.e., to cheat more behind the opaque barrier) when paired with more punitive males. As predicted, our female strategic cheating measure correlated with male punishment (N = 12, r = 0.64, p = 0.02; Fig. 2B). With the caveat that one can never be certain of the directionality of a correlational effect, these data hint that females who experienced more punishment from males showed more sensitivity in their cooperative behavior depending on whether or not they were being watched than females that did not experience much punishment.
Like primates (see ref. 6), female cleaners cheat more when their partners lack perceptual access to their behavior. Males apparently responded to female foraging behavior, being less likely to punish if the female ate against preference. Female cleaner fish, therefore, seem to take into account males’ visual access and are more cooperative when they are being observed than when they are hidden behind an opaque partition. Nevertheless, there is one potentially important difference between our experimental design and standard designs for primates: we could not construct a barrier that would allow the females to always see the male body but allow us to vary whether or not the male can see the female. Therefore, the results of the first experiments could be at least in part due to females responding to the presence/absence of the male partner. However, as female strategic behavior is correlated with male punishment—likely because females paired with more punitive male showing a more pronounced response to barrier condition—we propose that females’ performance is not merely an automatic reaction to being behind an opaque or transparent partition.
Study 2 used a more complicated choice task to explore whether female cleaner fish take into account a male’s perceptual access when planning where to feed. Females were given a choice to feed behind a transparent or opaque barrier while their males were separated. In one condition, males were separated by a transparent partition (male visible), and thus the female’s choice mattered for the males’ perceptual access (Fig. 1). In a second condition, males were separated by an opaque barrier (male not visible); note that in this condition females should be indifferent in their choice of the transparent and opaque partitions since the males’ perceptual access was already blocked by the opaque barrier. Unlike Study 1, in Study 2 rounds were blocked such that each condition was presented once across six trials within the first block of 12 trials and then again in the second block of 12 trials—i.e., fish saw both conditions presented in block 1 and then again in block 2.
We first examined females’ choices across condition and block. Condition was not a significant predictor of choices (LRT, X21 = 0.003, p = 0.954) whereas block was (LRT, X21 = 3.66, p = 0.056; Supplementary Table S2). However, this model was no better than the null (Supplementary Table S2). In a second model, we found that the interaction between condition and block was a significant predictor of females’ choices (LRT, X21 = 22.13, p < 0.001). As shown in Fig. 3, when females first encountered the experimental conditions in block 1, their behavior conformed to our predictions: they were more likely to feed behind the opaque barrier when the male was visible. However, this pattern of behavior reversed in the second block, in which females were more likely to choose to feed behind the transparent barrier when the male was visible (see Supplementary Fig. S4 for data separated by pair). To better understand females’ behavior in this task, we next turn to analyses of male punishment behavior and female feeding behavior.
Scatter plot of females’ choices to feed behind the opaque barrier as a function of round with plotted lines from a linear model separated by condition (N = 11 pairs of fish). Gray shaded area shows confidence bands.
We first examined whether male punishment varied as a function of condition, females’ choices and the interaction between the two terms in a GLMM. We found no effect of the interaction (LRT, X21 = 0.51, p = 0.477). When this term was removed, we found that males were more likely to punish—regardless of condition—when females chose to feed behind the opaque barrier (LRT, X21 = 4.48, p = 0.034; Supplementary Table S2). This result represents an interesting convergence with the male punishment results from Study 1. It is possible that females behave differently after they make a choice (e.g., give off a behavioral cue depending on whether they chose to feed in private), and that males adjust their level of punishment based on those cues. The fact that males punished more when females chose to feed behind the opaque barrier could explain why some females shifted their choices shifted towards feeding behind the transparent barrier between blocks 1 and 2.
Female cooperation, measured in terms of the number of flake items eaten prior to eating a prawn item, varied marginally as a function of their barrier choice and condition (LRT, 2-way interaction, X21 = 3.51, p = 0.061; Supplementary Table S2). This marginal effect is likely spurious, as the model including this interaction term was not a better fit to the data than the null. A significant two-way interaction would have been consistent with the prediction that females assumed that they could get away with cheating behind barriers in the male visible condition. However, when we examine this relationship in the first block alone—where females showed a preference for the opaque barrier—we no longer see an effect (X21 = 0.99, p = 0.32; Supplementary Table S2). Nevertheless, this negative result does not necessarily support the hypothesis that females lack an understanding that cheating would be more “permissible” behind barriers when their male partner was visible. In contrast to Study 1—where females cheated more when the male partner could not see—male partners counteracted that logic by being more likely to punish if the female ate behind the opaque barrier, independently of condition. Thus, while females may initially base their decisions on what males can or cannot see, they may still adjust their behavior based on males’ actual responses.
Taken together, these results suggest that cleaner fish show several of the hallmarks of primate ToM capacities. Like primates6,10,11, female cleaners strategically use visual barriers to hide their actions. In addition, females use this strategic cheating strategy more often at the individual level when paired with more punitive males. Finally, first block performance on Study 2 showed that females preemptively sought opaque barriers more often than transparent barriers in cases where males could see their actions. However, this effect was fragile and disappeared after repeated trials.
One intriguing finding from both Study 1 and Study 2 was that male punishment did not depend on whether they could see females: In Study 1, male punishment was related to female cheating, regardless of their visual access. In Study 2, males punished more whenever females chose to feed behind the opaque barrier. This surprising set of findings hints that male punishment may be based on factors other than female’s cheating behavior. It is possible, for example, that females may behave differently after making different kinds of choices (e.g., when they chose to feed in private vs public), and that males adjust their level of punishment based on those cues. It would be interesting to explore whether this is the case, and also to investigate whether females attend to similar male behavioral cues that may guide their decisions about how much and when to cheat. Understanding whether such behavioral cues exist and how they influence male punishment and females’ decisions is an exciting avenue for future work.
If females provide cues about their cheating, then why would our wild-caught females show any sensitivity to our experimental manipulations? One potential explanation for this pattern involves a difference between our task and the situation that cleaners experience in the wild. For instance, it is possible that such cues are short-lasting under natural conditions where partners may stay out of sight for prolonged periods and both males and females are often distracted by new clients demanding service. Under these conditions, any cues associated with female choices may not be noticed by males. By contrast, our studies allowed males to reunite with their females immediately after a trial with no distractions. Thus, females may attend to what males can and cannot see under natural conditions and may have demonstrated this pattern of performance—at least initially—in our experiments.
Another possibility worth considering is that female cleaners developed their ability to represent others’ perspectives in a totally different context, such as when represented what image-scoring bystander clients can and cannot see25,26. Cleaners are highly sensitive to how bystander clients respond to their treatment of current clients28,29,30. Therefore, paying attention to whether or not bystander clients can see how cleaners treat current clients might be the primary selective force on perspective-taking abilities, which are then also applied to male cleaners.
In conclusion, our results suggest that cleaner wrasse can detect what their partners can and cannot see—one important key feature of human-like ToM capacities—in the context of their cooperative cleaning mutualism. These findings support the view that ecological pressures for strategic deception can drive surprisingly complex human-like cognitive abilities even in a distantly related fish species.
Source: Ecology - nature.com
