Simple decision rules underlie collaborative hunting in yellow saddle goatfish

Abstract

Collaborative hunting, the coordination of animal behaviour in space and time to capture prey, is reported in several vertebrate species. However, previous studies are observational, hampering our ability to identify individual decision rules that result in collaboration. We experimentally investigated collaborative hunting in yellow saddle goatfish (Parupeneus cyclostomus) by exposing pairs to a mock prey that fled to an artificial shelter with multiple entrances. The first fish to initiate the chase (the ‘initiator’) was always closest to the prey and pursued it directly in its path. Conversely, the behaviour of the second goatfish (the ‘follower’) depended on its spatial position relative to the initiator. When the follower was less than one body length behind the initiator, it also accelerated directly towards the prey in over 95% of cases. However, if the two goatfish were separated by a distance of one body length or more, the follower chose a less direct route to reach the prey in 87% of cases. In this scenario, the follower often reached the prey's more distant refuge first, which might increase its hunting success or block the prey's escape path under natural conditions. Our findings suggest that coordinated hunting behaviour can result from simple, self-serving decisions.

1. Introduction

Collaborative hunting, the coordination of prey capture in space and time, has been documented in a handful of vertebrate species [1–3], and is often perceived as an advanced hunting strategy requiring large brains that allow high levels of coordination. However, recent studies have shown that interspecific collaborative hunting between small-brained species such as fishes (e.g. grouper and moray eel) and fish and invertebrates (grouper and octopus) is not only possible but can also be highly sophisticated, involving some form of planning, referential gestures and partner choice [4–6]. In collaborative hunts that involve partners of different species (e.g. moray eels and groupers [4]), coordination might be easier to achieve because each partner performs a specialized role corresponding to its specific ecology. Conversely, intraspecific collaborative hunting is thought to require more advanced cognitive abilities because functionally similar individuals must learn to perform different tasks [7]. For example, studies have shown that large-brained mammals such as lions and dolphins are capable of specializing in roles such as chasing prey or blocking the prey's escape path to facilitate capture by the group [2,3]. Interestingly, cases of intraspecific collaborative hunting have also been recently documented in smaller-brained vertebrates such as birds [8], reptiles [9] and fish [7]. Therefore, contrary to previous beliefs [1,10], current evidence suggests that collaborative hunting does not necessarily rely on cognitive processes that require large and complex brains. Rather, it seems possible that sophisticated hunting strategies can also emerge in relatively smaller-brained species in response to ecological needs. While it is extremely challenging to study underlying cognitive processes, it is feasible to identify the decision rules that lead to collaborative hunting and determine whether they can be simple.

We know little about the decision rules underlying intraspecific collaborative hunting because all studies to date have relied exclusively on field observations. Observational data are essential for documenting the natural occurrences of such behaviour but cannot pinpoint the specific decision rules that result in coordinated action. This limitation arises because, in the field, each hunt becomes idiosyncratic due to large variation in parameters such as habitat structure, predator and prey group size, predator motivation, or the identity, location and behaviour of prey. Such variation makes it difficult to identify the individual strategies that lead to collaborative hunts. For example, do participants display behaviours that reduce their own probability of capturing prey while increasing the probability of capture by other group members? Such helping, in which individual actions increase group benefits at a personal cost, requires an explanation centred around kin selection [11,12], reciprocity [13] based on a prisoner's dilemma or a volunteer's dilemma-type pay-off matrix [14,15], or pseudoreciprocity based on interdependence between hunting members [16,17]. In the absence of such behaviour, collaboration is said to result from simple self-serving decisions that lead to by-product benefits [18].

Understanding which ‘game’ partners are playing can help us identify how challenging it will be to achieve stable collaborative hunting. For example, individual investments in an iterated prisoner's dilemma-type game warrant more complex decision rules than self-serving decisions, which only require that one partner's behaviour is flexibly adjusted to the behavioural decisions of others. Game theoretic scenarios can be further complicated when a successful hunt leads to a carcass that is sharable, at least in principle, as is typically the case in hunts by lions and chimpanzees [1,2]. By contrast, when a captured prey is swallowed whole, as in the case of hunting crocodiles and goatfish [7,9], the ‘game’ is simpler, facilitating the identification of decision rules employed by the participants.

Here, we examined the individual decisions of yellow saddle goatfish (Parupeneus cyclostomus) during group hunts in controlled laboratory experiments. Our objective was to identify decision rules that are likely to underlie collaborative hunting in the wild. In nature, P. cyclostomus hunts for invertebrates and prey fishes either alone or in small groups, typically consisting of two to six individuals (figure 1). In group hunts, members regularly adopt one of two roles: either attacking and chasing prey, or blocking their escape path [7]. The attacking goatfish directly pursues a fleeing prey in its path while other group members deviate around obstacles such as coral heads, thus preventing the prey from escaping. We hypothesized that factors such as an individual's proximity and ability to detect a prey determine which function each goatfish adopts because a previous observational study on P. cyclostomus found no evidence of role specialization in this species [7]. We simulated a hunting scenario by exposing pairs of goatfish to a mobile (mock) prey that sought refuge under a shelter with multiple entrance points. We then tested whether goatfish displayed collaboration by adopting different spatial routes in pursuit of the prey and what specific conditions resulted in collaborative hunts. The absence of a food reward in our experiments meant that we studied spontaneous decisions by animals that were not affected by variation in prey capture success.

Figure 1.

Yellow saddle goatfish (Parupeneus cyclostomus) often hunt collaboratively for invertebrates and prey fishes on coral reefs. Group members exhibit coordinated movements during a hunt, where one individual chases the prey and others block its escape path. When a prey seeks shelter, goatfish use their barbels to try and extract it from underneath the refuge. Photograph by M.S. (Online version in colour.)

2. Material and methods

(a). Study species and experimental set-up

Experiments were conducted during three field seasons at the Dahab Marine Research Center (DMRC) on the coast of the Red Sea, Egypt: in the summer of 2010, and in the spring and autumn of 2011. During each field season, we captured eight P. cyclostomus using a 30 m² barrier net (stretched mesh size 20 mm) and transported them in 20 l buckets to an air-conditioned room at the research station (27.0 ± 2.0°C, actual variation). Artificial light was provided following a L : D cycle that matched the local sunrise and sunset with 10 min of twilight at the beginning and end of the light phase. Fish were held in groups of four individuals in two separate round tanks (138 cm diameter, 39 cm water height) supplied with seawater at approximately 180 l h^–1 via a flow-through system with water pumped directly from the Red Sea. We grouped fish based on body size, housing the four smallest and the four largest together. The mean (±s.d.) total length (TL) and body weight of the 24 fish collected across the three field seasons were 190 ± 26.7 mm and 60 ± 38.91 g. Fish were individually recognizable based on differences in body size, coloration patterns and markings [7].

All experiments were conducted in the fish's housing tanks. An opaque partition was placed on one side of the tank to isolate non-experimental fish from experimental ones (figure 2). Holes in the partition allowed water exchanges with the rest of the tank and air stones provided mixing and aeration. The bottom of the tank was covered in a 1 cm layer of sand. Coral rocks were used to create two shelters at the periphery of the tank such that they did not obstruct the central area (figure 2).

Figure 2.

(a) Side view of the experimental arena. The mock prey was attached to fishing thread and could be pulled across the bottom of the tank to a shelter with multiple entry points. A partition separated the experimental fish from the other two fish housed in the tank. (b) Top view of the experimental arena. The letters A and B indicate the starting points from which the prey was pulled. The prey could enter the shelter via four entry points: a1 or a2 if the prey was in start location A; b1 or b2 if the prey was in start location B. Route C represents a situation in which the follower takes an identical route to the initiator, following the prey in its path. Route D represents a situation in which the follower opts to circle the shelter, deviating from the initiator's path; here, the follower would reach the prey before the initiator because the prey enters the refuge via entry point a2. (c) Top photograph of the experimental arena. The two goatfish are visible in the top right corner and the mock prey is in starting position A. (Online version in colour.)

(b). Feeding and training

Fish were fed small pieces of fish ad libitum, once a day in the late afternoon. Uneaten food was removed from the tank on the same day. Prior to the feeding session, we trained fish to pursue moving prey, beginning on day three after capture. All four fish were present in the tank during this training period. First, a food item was attached to a fishing thread and dangled in front of individuals to mimic a live prey. All fish fed in this way within 3 days from the start of training. Fish were then trained to chase a food item pulled at high velocity across the bottom of the tank. A small lead weight was attached to the end of the thread, next to the prey, so it would remain on the substrate. The start and end positions of the prey were random across trials and the prey was pulled out of the water before the goatfish could reach it. Training was completed in one tank when all four goatfish rapidly accelerated in pursuit of the prey, which is typical of this species's hunting behaviour in the wild [7]. Completing both of these training phases took between 7 and 10 days.

(c). Experimental procedure

Experiments were performed on pairs of goatfish. In a given tank, one pair remained in the experimental arena and the other was isolated behind the partition described above (figure 2). After placing the partition, we waited 30 min before starting a trial. Each pair participated in 16 trials during field season 1, and 20 trials during seasons 2 and 3. Consecutive trials were separated by 3–4 min.

To initiate a trial, the experimenter used a hook on a stick to position the mock prey (weighted plastic diamond 1.5 cm long by 1.0 cm wide) at one of two locations in the tank (labelled A and B in figure 2b). Each location was marked with two small coral rocks (figure 2b). We used a mock prey rather than a real food item to avoid fish striking when the experimenter positioned the prey in the tank and to avoid fish receiving a food reward. Rapid movement of the mock prey was sufficient to initiate pursuit by the goatfish (electronic supplementary material, video S1). Once the prey was in a starting position, the experimenter waited between 10 and 30 s before rapidly pulling the prey across the bottom of the arena and under a shelter at the centre of the tank (figure 2). As in the training phase, the mock prey was attached to fishing line and pulled by releasing a weight down a ramp, yielding constant acceleration across trials and fish pairs (figure 2). The weight was released when both goatfish were approximately within one body length of a shelter and visible to the camera. The speed at which the prey travelled was such that it could not be caught by either goatfish except on rare occasions.

The shelter to which the prey retreated consisted of an acrylic disc covered in coral rock. The disc was separated into four sections (figure 2b) and was sufficiently elevated from the substrate to allow the prey entering, but not the goatfish. When the prey was pulled from position A in the tank, it could enter the shelter directly (entrance a1 in figure 2b) or indirectly, by circling the shelter before retreating (entrance a2 in figure 2b). The same options (entrances b1 and b2) were possible when the prey was pulled from starting position B (figure 2b). Different entry points were used to mimic natural conditions on the reef, where goatfish cannot predict whether and where prey will seek shelter. We counterbalanced the prey's starting position and route to the shelter for trials performed on each goatfish pair (i.e. all options were presented to each pair, in different orders).

(d). Video analysis and data extraction

We recorded all trials at 25 Hz with a video camera positioned above the tank (Sony HDR-CX350VE, Sony Electronics, Tokyo, Japan). Videos were analysed frame by frame with the software Picture Motion Browser (Sony Electronics) to determine the exact position and time when each goatfish responded to the stimulus. We measured the straight distance between each fish's snout and the centre of the prey when fish made their first head movement in pursuit of the prey. Distances were measured in two dimensions because P. cyclostomus is a bottom dweller and the mock prey was weighted to ensure movement along the substrate. The first goatfish to respond was termed the ‘initiator’ and the second was the ‘follower’. We categorized trials into three scenarios in which the proximity of the two fish to the prey differed by: (i) one-third of a body length or less (i.e. both fish were considered equally distant from the prey; the mean body length of the pair was used), (ii) more than one-third but less than one body length (i.e. one fish was slightly closer to the prey) and (iii) more than one body length (i.e. one fish was considerably closer to the prey). In each of these scenarios, we recorded whether fish reacted simultaneously or not (i.e. in the same or different frames). Sequential frames were separated by 0.04 s. In the latter case, we noted whether the initiator and the follower directly chased the prey in its path or circled around the acrylic shelter (paths C and D, respectively, in figure 2b).

To specifically evaluate collaboration between the two hunters, we identified instances when the follower chose a different route from that of the initiator, which occurred in two situations. First, the projected future position of the prey and the position of the follower at the time it initiated its response were such that circling around the shelter was in fact the shortest route to reach the prey (e.g. path D in figure 2b). This case does not provide strong evidence of coordinated action by both partners because it could be a simple by-product of the follower opting for a route that minimizes the time it needs to reach the prey. Second, circling around the shelter caused the follower to take a longer route to reach the prey. This situation is indicative of ‘true’ collaboration because the follower bears the cost of choosing a route that lowers its chances of reaching the prey first.

(e). Statistical analysis

We used a general linear mixed-effects model (LMM; ‘lme’ function in the R package ‘nlme’) to test whether goatfish exhibited signs of fatigue or habituation due to repeated stimulations [19]. We included the response latency of the first fish to accelerate towards the prey as the response variable (i.e. the time it took the first fish to respond to movement by the prey); trial number was specified as a fixed (discrete) predictor and goatfish pair as a random factor to account for repeated measurements. We verified model assumptions using diagnostic plots for fixed and random effects (i.e. plot, qqPlot, qqnorm functions in R). We calculated the repeatability (R) of response latency using the ‘rpt’ function in the R package ‘rptR’ [20].

We used Wilcoxon signed-rank tests (hereafter Wilcoxon) to account for non-normality in the data and repeated measurements on pairs of goatfish. First, we used two tests to examine whether the same fish always assumed the initiator or follower role in a given pair and whether this depended on body size. Second, we used three tests to determine how the distance separating the follower from the initiator (i.e. the three different scenarios) influenced the follower's decision to pursue the prey via a direct or indirect route. Third, we selected only instances of scenario 3 in which a decision by the follower to circle the shelter and reach the prey would result in a longer route than direct pursuit. We tested whether the follower preferentially chose the longer route over the shorter, more direct route. Statistical analyses were done in R v. 3.3.3 [21].

3. Results

In nine of the 12 pairs tested, both goatfish consistently chased the mock prey upon release, yielding a total of 143 successful trials. In 62 trials (43.4%), both fish began accelerating towards the prey at the same time (i.e. within 0.04 s; table 1). In all trials where one fish responded before the other (n = 81), the initiator always pursued the prey using the most direct route. There was no evidence of fatigue or habituation to repeated stimulations (LMM, main effect of trial number: F_1,133 = 0.521, p = 0.471). Response latency was repeatable at R = 0.142 (95% CI: 0.035–0.441). With the exception of pairs 3 and 6, both fish assumed the role of initiator and follower in approximately equal proportions (Wilcoxon V = 12, n = 7, p = 0.799; table 1). Body size (weight) did not influence which fish became the initiator and which the follower (Wilcoxon V = 14.5, n = 9, p = 0.373; table 1).

Table 1.

Body total length (TL, mm) and weight (g) of individuals in the nine goatfish pairs tested. Also indicated is the number of times each individual assumed the initiator role (i.e. was the first fish in the pair to respond to the stimulus) and the number of times individuals responded simultaneously (i.e. within 0.04 s).

pair	fish	TL (mm)	weight (g)	initiator	simultaneous
1	1	258	180	3	5
1	2	254	170	7
2	1	190	65	3	2
2	2	173	50	2
3	1	192	70	6	5
3	2	182	55	1
4	1	175	47	7	7
4	2	170	38	4
5	1	210	70	8	6
5	2	184	52	3
6	1	177	56	10	9
6	2	171	34	0
7	1	194	62	2	11
7	2	187	54	6
8	1	238	92	4	6
8	2	219	88	7
9	1	205	70	5	11
9	2	190	58	3

The initiator always reached the mock prey's closest entrance point first (a1 or b1 in figure 2b). By contrast, followers that circled the shelter tended to arrive at the prey's furthest entrance point (a2 or b2 in figure 2b) before the initiator (i.e. in 18 of 24 events; Wilcoxon V = 5, n = 9, p = 0.075).

Fourteen trials corresponded to scenario 1, whereby the distance separating each fish from the prey differed by one-third of a body length or less. In 100% of these cases, both fish responded at the same time (i.e. within 0.04 s) and always pursued the prey via the shortest route. In trials corresponding to scenarios 2 and 3, and where fish responded to the stimulus at different times (n = 81), the follower's choice of route depended strongly on its position relative to the initiator. If the initiator's lead was between one-third and one body length ahead of the follower (n = 26 trials), the follower also preferentially chose the most direct route in pursuit of the prey (i.e. in 96% of cases; Wilcoxon V = 36, n = 9, p = 0.0140; figure 3; electronic supplementary material, video S1). Conversely, if the initiator was more than one body length ahead of the follower (n = 55 trials), the follower exhibited a significant preference for circling the shelter (i.e. in 89% of cases; Wilcoxon V = 0, n = 9, p = 0.0088; figure 3; electronic supplementary material, video S2). When we excluded eight of these 55 trials in which circling the shelter resulted in a shorter distance to the prey (e.g. route D in figure 2b), followers still exhibited a significant preference for the longer route that involved circling the shelter (i.e. in 87% of cases; Wilcoxon V = 0, n = 9, p = 0.0090).

Figure 3.

Box-and-whisker plots showing the proportion of times the ‘follower’ (i.e. the second goatfish to pursue the mock prey) opted to chase the prey via a longer route than the ‘initiator’ (i.e. the first goatfish to initiate pursuit) when the distance separating the two predators was either (i) between one-third and one body length (scenario 2; n_trials = 26) or (ii) greater than one body length (scenario 3; n_trials = 47). Nine goatfish pairs were tested. Whiskers extend to the highest value within 1.5 times the inter-quartile range; data points are jittered to avoid overlap.

4. Discussion

Our results support the hypothesis that collaborative hunting in yellow saddle goatish is based on simple decision rules. The response of the individual that first detected the moving prey (the initiator) was always to initiate a direct pursuit. Similarly, goatfish that were second to react in our experiments (i.e. the ‘followers’) directly pursued the prey in over 95% of trials when they were in close proximity to the initiator (distance ≤ 1 BL). We observed a sharp drop in direct pursuits by the follower when it lagged behind the initiator by a distance greater than 1 BL. Here, the follower opted for a longer, less direct path to the prey in 87% of cases, often reaching the prey's more distant refuge first. Such a decision could increase the follower's chance of capturing the prey or blocking the prey's escape path under natural conditions. Therefore, a simple distance-based rule explains nearly all of the observed variation in the occurrence or absence of collaborative group hunting in our system.

(a). Functional hypotheses for collaboration

Our experiment was not designed to quantify hunting success because we used a mock prey that could not be captured. Nevertheless, our study allows inferring the potential benefits of individual decisions. First, we note that direct pursuit of the prey by the initiator is a decision that probably maximizes its net benefits. Choosing an alternative route that is not in the prey's direct path would only be beneficial if the prey has to circle a shelter because it cannot immediately access an entry point (figure 2b). However, because predators do not know the location of all hiding places in the reef and/or a prey's potential preference for a particular shelter [22], direct pursuit should, on average, be the most effective (and hence preferred) option for the initiator to achieve capture.

From the follower's perspective, direct pursuit of the prey might also be the best strategic option when it is at a short distance from the initiator and has limited time for decision-making. The follower's decision to follow the initiator when both fish are in close proximity suggests that its probability of capturing the prey under these circumstances is non-negligible in nature. Alternatively, once the follower has initiated pursuit, it might lack the ability to effectively adjust its direction during the chase even if an alternative path becomes more advantageous (e.g. if the distance between the two predators widens during the chase). Our data show that, when both fish followed the prey in its path, the initiator always arrived first at the closest shelter entrance under which the prey could seek refuge. Hence, at least under these experimental conditions, followers were unable to catch up to the initiator but still never strategically adjusted their trajectory during the chase.

When the distance between hunters widens, the further the follower lags behind the initiator, the less likely it becomes that a direct pursuit will yield hunting success as the prey would be swallowed by the initiator or escape into a refuge. Conversely, given the uncertainties about the distribution of locally available refuges and unknown prey preferences for shelters, opting for a non-direct route to the prey has the potential to increase the follower's hunting success relative to a near-zero baseline. In our experiment, variable prey, behaviour and the presence of multiple possible entry points under the artificial shelter were intended to mimic the uncertainties that predators face in nature. Opting for a path that differed from that of the prey resulted in an advantageous position for the follower (i.e. closer to the prey's point of entrance to the shelter) in instances when the prey was pulled to the furthest (i.e. second) entrance around the shelter (a2 or b2 in figure 2b). Therefore, it is conceivable that, under natural conditions, deviating from the prey's escape path occasionally increases the follower's chance of intercepting and capturing the prey. In other words, a self-serving decision by the follower would explain the occurrence of collaborative hunting.

Accurately determining the breakpoint in the functional form describing when the follower's choice of strategic option changes (i.e. opting for an indirect versus a direct path to the prey) would have required data with higher temporal resolution than we were able to collect. Owing to the rapidity of the goatfish's displacements, the mean distance covered by an individual once in movement was approximately 6 cm between frames in our videos (i.e. in 0.04 s). As a result, the time resolution of our video recordings only allowed us to categorize trials into three different scenarios (see Material and methods) rather than perform a more detailed analysis using continuous predictors. Nevertheless, despite this caveat, our results convincingly demonstrate that (i) the breakpoint of this functional form is situated near a distance of 1 BL between the hunters (see figure 3 showing that followers opt for a direct path to the prey in 96% of cases when this distance is less than or equal to 1 BL and an indirect path in 87% of cases when the distance is greater than 1 BL), and (ii) the relative orientation of the hunters appears unimportant because the differences in strategic choices observed are overwhelmingly explained by distance alone, irrespective of the existing variation in the fish's relative angular position. We also note that the shape of this functional form will likely change in natural or other experimental settings that differ in the number and distance to refuges.

(b). Is collaborative hunting indicative of cooperation by yellow saddle goatfish?

Demonstrating that collaborative hunting by yellow saddle goatfish constitutes cooperation requires showing that a pair's hunting success is more than double that of a single individual. We could not examine each partner's capture success in our study; however, by pursuing the prey via an indirect path, the follower reduces the prey's escape options, effectively fulfilling the role of a ‘blocker’ [1]. Such collaboration is believed to increase hunting success at the group level in species such as lions and dolphins, where coordinated hunting is perceived as cooperative [2,3]. However, unlike these species that share a captured prey (or school of prey), only one yellow saddle goatfish can be successful during a hunt. Which hunter, if any, captures the prey will depend on the prey's behaviour when it is cornered. For example, the initiator might chase the prey towards a follower that opted for an indirect path, or the follower might surprise the prey and alter its movement (i.e. slow down or halt) or path (i.e. change direction), facilitating capture by the initiator. In addition to collaborating during the chase, goatfish must also coordinate their movements when trying to extract a cornered prey from underneath a shelter using their barbels [7] (figure 1). Thus, in yellow saddle goatfish, the relative success of groups versus singletons depends on the potential benefits of group hunting on both phases of a hunt. Future studies should explicitly focus on quantifying the increase in hunting success resulting from group collaboration versus solitary hunts.

Short of demonstrating cooperation, our study suggests that collaborative chases by yellow saddle goatfish are a by-product of individual decisions in an overall competitive situation where a captured prey is not shared by hunters. Here, each goatfish tries to detect a suitable prey first because the first fish to respond has the highest probability of capturing the prey. The follower then makes the most out of a bad situation, either by attempting to catch-up to the initiator and pursuing the prey in its path if it is separated from the initiator by a small distance (less than or equal to 1 BL), or by opting for an alternate route if the initiator's lead is larger (greater than 1 BL). Reciprocation is not necessary to explain the stability of this collaboration because individuals invariably behave in a self-serving way. Similarly, reciprocation is probably unnecessary to explain collaboration in other systems where field studies have described group hunting of non-shared prey [3,9]. For example, simple by-product benefits probably explain why sailfish hunt in groups, allowing them to herd schools of sardines and strike at prey in turn [23], and why spotted sea trout coordinate their attacks on prey schools, inhibiting collective antipredator responses [24].

(c). Future directions

Further insights into the decision rules underlying collaborative hunting would be gained by studying other systems which are amenable to experimentation and in which prey are shared. For example, sharing the prey may offer an immediate cooperative solution to individuals that display costly behaviour (e.g. adopting a role that reduces individual success in favour of increased group success) if it can be shown that such investments are rewarded at the end of the hunt. There is evidence for this phenomenon among chimpanzees in the Taï National Park, Ivory Coast [25], offering opportunities to study the role of inequity aversion [26,27] as a mechanism to regulate pay-offs in return for investments.

We found little evidence for role specialization among goatfish during collaborative hunts, and no obvious indication of factors underlying the few cases in which fish tended to consistently adopt the same role (two of nine pairs). Role specialization may not be expected in a scenario where the first individual to detect a prey must respond immediately to maximize its chances of capture. By contrast, role specialization may emerge in situations where a specific behaviour can improve over time through experience; for example, in dolphins, where members of a pod tend to specialize as ‘herders’ or ‘blockers’ of fish schools [3], or in lions, where several individuals chase prey towards a hiding pride member that takes the prey by surprise [2]. These questions are ripe for future investigation.

5. Conclusion

Our experimental study provides strong support for the hypothesis that simple, self-serving decision rules can result in instances of coordinated hunting by partners. Therefore, the possibility that similar forms of collaboration in other species are based on sophisticated decisions that warrant reciprocity and hence advanced cognitive processes (e.g. individual recognition, memorization of past interactions (book keeping), negative inequity aversion and a degree of self-control that allows playing immediate pay-off-reducing roles) should be demonstrated rather than assumed by default.

Ethics

This study was approved by the Dahab Marine Research Center and the Suez Canal University. Fish were released at their site of capture after the experiments.

Data accessibility

The data and script for this study are archived in the repository figshare following best practices [28] and were made available to editors and reviewers upon initial submission https://doi.org/10.6084/m9.figshare.4877783.

Authors' contributions

M.S. and R.B. designed the study. M.S. and Hanaa Sarhan collected the data. D.G.R. and M.S. analysed the data. M.S., D.G.R. and R.B. wrote the paper.

Competing interests

We declare no competing financial interests.

Funding

This study was funded by a Swiss National Science Foundation grant to R.B.