The multi-dimensional nature of information drives prioritization of private over social information in ants

Abstract

When personally gathered and socially acquired information conflict, animals often prioritize private information. We propose that this is because private information often contains details that social information lacks. We test this idea in an ant model. Ants using a food source learn its location and quality rapidly (private information), whereas pheromone trails (social information) provide good directional information, but lack reliable information about food quality. If this lack is indeed responsible for the choice of memory over pheromone trails, adding information that better food is available should cause foragers to switch their priority to social information. We show it does: while ants follow memory rather than pheromones when they conflict, adding unambiguous information about a better potential food source (a 2 µl droplet of good food) reverses this pattern, from 60% of ants following their memory to 75% following the pheromone trail. Using fluorescence microscopy, we demonstrate that food (and thus information) flows from fed workers to outgoing foragers, explaining the frequent contacts of ants on trails. Ants trained to poor food that contact nest-mates fed with good food are more likely to follow a trail than ants which received information about poor food. We conclude that social information may often be ignored because it lacks certain crucial dimensions, suggesting that information content is crucial for understanding how and when animals prioritize social and private information.

1. Introduction

When making decisions, animals can use both private and social information [1,2]. Private information includes any personally acquired information such as internal states and, importantly, memories. Socially acquired information includes any information generated by the behaviour of other organisms [2], such as the presence of conspecifics at a resource, food calls, or the waggle-dance of a honeybee. When private and social information sources agree, they may act additively or synergistically to improve behaviour. For example, birds may make a more rapid or accurate assessment of a food patch's quality if they feed in the company of conspecifics [3,4].

An interesting situation arises when private and social information sources conflict. One option in such a situation is to weight information from different sources and produce an intermediate value [5–7]. An alternative is to rely on one type of information and ignore the other. Understanding how animals decide which information to follow is an active field of research [2,8–11]. Evidence for various information use strategies has been found in a wide range of taxa, including insects, fishes, lizards, rats and humans [12–15]. A common finding is that, all else being equal, private information is preferentially followed or acquired when it conflicts with social information. For example, guppies that fed from one feeder type, but have observed shoal-mates feeding from another, will choose the feeder from which they have personal feeding experience [16], unless doing so is costly. Nine-spined sticklebacks also preferentially follow private information when choosing a feeding site, unless their private information is out of date [17]. Similarly, wild starlings only make use of social information for patch quality assessment if doing so does not interfere with collecting personal information [3].

Information use strategies, and the response to conflicting information, have been especially well documented in social insects such as ants and bees [9,18]. Social insects represent a special case, because individuals do not compete against each other in many situations requiring communication, such as foraging and nest-site selection. This lack of competition favours the evolution of explicit social signals, such as pheromone trails or waggle-dances. Somewhat surprisingly, however, ants and bees still predominantly follow their own memories when such intentionally produced social information is conflicted with private information [19–27], with only a few exceptions [21,28,29]. It has been proposed that the reason for this behaviour is the reliability of the information sources—how often they are associated with a positive outcome, or how outdated they are (e.g. [11,17]), or the cost of acquiring new information [30–33]. In social insects, reliance on private information may allow individual workers to specialize on different food sources, increasing overall food intake [34,35]. Widely ignored, however, is information content. Pointing in a certain direction, for example, can provide high precision and accuracy (high information content) in terms of direction, but be ambiguous (low information content) in terms of distance. We hypothesized that such information ambiguities may be driving the neglect of social information.

For ants, pheromone trails provide accurate and precise directional information, but only very noisy, imprecise and ambiguous information about the quality of a food source. Pheromone trail strength varies with resource quality, time since discovery, recruitment rates, and the individual pheromone depositions behaviour of ants: a strong trail may lead to a good food source exploited by a few ants, or to a poor one exploited by many ants; a weak trail may lead to a poor food source, or to a newly discovered high-quality food source. Importantly, there is a very large variation in the amount of pheromone deposited by individual ants to food sources of the same quality [36,37]. With no strong evidence that an advertised food source is better than a known one, foragers avoid searching for it [30,38], and rather exploit the food source they know.

This hypothesis leads to a testable prediction: if a worker foraging on a poor food source can be given unambiguous information that a better food source is available, she should follow the trail pheromone, as this has a high probability of leading to better food. Here, we set out to test this prediction by first arranging for a conflict between pheromone trails and route memory, and then providing additional information about the existence of better food than the memorised one.

2. Material and methods

(a). Study species and maintenance

We used eight queenless colony fragments (ca 1000 workers) of the black garden ant, Lasius niger (Linnaeus), collected from eight different colonies on the University of Regensburg campus. The ants were fed 1 M sucrose ad libitum, supplemented with Drosophila melanogaster. Colonies were deprived of food for 4 days prior to each trial. Water was provided ad libitum.

(b). Experimental series 1 procedure

Ten different experimental treatments were carried out in this first experimental series, coded A-J (table 1). All treatments were variations on a central design. The aim of the experiments was to test whether ants with a well-established memory of finding food on one arm of a Y-maze can be induced to follow a newly presented alternate arm. Various information sources were provided about the alternate arm, including a pheromone trail, a small (0.2 µl) droplet of sucrose (simulating trophallaxis or on-trail contact with a fed ant), and odours on the new path and in the droplet. Previous work showed that when pheromone information and memories conflict, L. niger workers tend to follow their memory [21,22]. We hypothesized that adding a tiny droplet of sucrose of better quality than the memorized food would induce ants to prioritize pheromone trails over their memory, as the droplet informs the ant that somewhere in the world better food than the known source is available. In addition, we hypothesized that an added odour could link the droplet (which offers no directional information) and a path odour (which does) [39]. Droplet odour or path odour alone should not affect ant choice. To test these hypotheses, a novel scent was added either to the droplet or on the newly presented path, or both.

Table 1.

Treatment codes, their descriptions, GLMM estimates for treatment means (linear scale) and their standard error, plus p-values for a test against 0 (random choice). (Positive estimates represent a tendency to follow memory.)

treatment code	treatment name	estimate	error	p-value
A	no information (pure memory control)	1.40	0.50	0.005
B	new path odour	1.08	0.43	0.013
C	unscented droplet	1.10	0.43	0.011
D	new path odour + unscented droplet	0.76	0.46	0.096
E	new path odour + scented droplet	0.64	0.41	0.12
F	pheromone + new path odour	0.26	0.39	0.51
G	pheromone + unscented droplet	−1.11	0.46	0.014
H	pheromone + new path odour + unscented droplet	−1.46	0.48	0.0024
I	pheromone + new path odour + scented droplet	−0.81	0.40	0.04
J	pheromone + new path odour + scented droplet, trained on 1.5 M sucrose	−0.04	0.39	0.92

Each trial began by allowing an ant onto the apparatus via a drawbridge. The drawbridge led to a Y-maze with one arm drawn out just out of reach of the ants, to form an L maze (figure 1). Half the ants were thus trained to the left, the other to the right. The arms of the maze were 10 cm long and 1 cm wide, narrowing to 2 mm wide at the junction, and were covered with paper overlays. The overlay on the maze's stem was unscented, while the overlay maze's arm was scented with either rosemary or lemon essential oil, systematically varied between trials. Odour impregnation was achieved by storing the paper overlays in a sealed plastic container containing one 0.1 ml drop of essential oil on a glass Petri dish for at least 24 h. An acetate sheet affixed to the end of the L-maze arm acted as a feeder. A large drop of 0.25 M sucrose solution (1.5 M in treatment J), flavoured similarly to the maze arm leading to it, was placed onto the feeder. Sucrose solutions were flavoured by adding 10 µl essential oil 100 ml⁻¹ sucrose solution. Once the ant found the feeder it was marked with a dot of acrylic paint on the abdomen, and allowed to return to the nest. While the ant was in the nest, unloading her sucrose load, the paper overlays on the stem and arm were replaced by fresh overlays, to remove any trail pheromone the ant might have laid. The ant was then allowed to make three further return visits to the same feeder. These four visits are sufficient to ensure that, even given the relatively low quality of food, most ants will return to this arm of the maze given a choice.

Figure 1.

(a) Experimental set-up: ants are first trained for four visits on a Y-maze with access to only one arm (here the top arm, in yellow in the figure but not coloured in the experiment), leading to a 0.25 M sucrose feeder flavoured with lemon or rosemary (1.5 M in treatment J). On the 5th outward journey the second arm of the Y-maze (here bottom, green) is added, and the feeder removed. Depending on the treatment, the new arm was blank, marked with a pheromone trail, and/or scented with a novel scent (lemon or rosemary). Additionally, a small (0.2 µl) droplet of 1.5 M sucrose may be placed after a narrowing of the stem before the bifurcation, which the ant can drink but will not become satiated from. The droplet may be flavoured to match the new Y-maze arm scent or may be unflavoured. (b) Example fluorescence image of an ant (abdomen) which did not encounter fluorescent dye-fed nest-mates (negative control). (c) Fluorescence image of an ant (abdomen) which spent 2 min encountering dye-fed nest-mates. (Online version in colour.)

On the ant's 5th return, the feeder from the original arm of the Y-maze was removed, and access to the second arm was given. The new arm was either marked with a pheromone trail (treatments F-J) or not (treatments A-E). This pheromone trail was produced by immersing eight worker hindgut glands in 2 ml of dichloromethane (DCM), following von Thienen et al. [40] then 5.6 µl of this mixture was applied in an even line along the paper overlay covering the arm, using a capillary tube (Servoprax GmbH, Germany). This amount was calculated to produce a pheromone trail of a realistic strength [40], and in control trials elicited a trail following accuracy of 82% (594 out of 715), which is indistinguishable from those recorded from naive ants following a reasonably strong naturally deposited trails, as reported by [41] (82–83% accuracy, see the electronic supplementary material, S1)

The new arm was also either scented (treatments B, D, E, H, I, J) with a different scent to that of the original arm, or was not (treatments A, C, F, G). Lastly, a small (ca 0.2 µl) droplet of sucrose either was (treatments C, D, E, G, H, I, J) or was not (treatments A, B, F) placed on the stem of the Y-maze, before the bifurcation but just after a narrowing of the stem. This ensured that the ant contacted the droplet as it walked towards the bifurcation. This droplet was either flavoured similarly to the scent on the newly presented runway (treatments E, I, J) or unscented (treatments C, D, G, H). The droplet was large enough for the ant to detect and drink, but not enough to satiate the ant [42], which proceeded onwards after drinking the droplet. This droplet was designed to simulate trophallaxis or contacting fed ants on the trail. Trophallaxis is used in the nest to unload food, and also provides ants with information about the food available in the environment [43]. This information is attended to very strongly [44].

We recorded the ant's initial decision, defined by the ant crossing a line 2 cm from the bifurcation, and its final decision, defined by the antennae of the ant reaching the end of the Y-maze arm. As the ant reached the end of the Y-maze, it was allowed to walk onto a piece of paper and replaced on the path leading to the Y-maze, before the location of the sugar droplet. We made 10 observations of each ant in this way. After that, the ant was permanently removed from the colony. The number of ants tested in each treatment can be found in figure 2.

Figure 2.

Proportion of ants following their memory as a function of treatment—see table 1 for definitions of each treatment. Bars are the raw proportion of choices, circles and whiskers are the model estimates and 95% confidence intervals (CIs) for the estimates. Note that these differ, as the model controls for side bias. Top letters denote whether groups are significantly different, with groups sharing a letter not being significantly different from each other. p-values for all pairwise comparisons between treatments are provided in the electronic supplementary material, S1. Significance indications **, * and n.s. describe p-values for a test against random decisions (proportion 0.5). ** = p < 0.05 but >0.005, * = p < 0.05 but >0.005. n.s. = p > 0.05. Sample sizes are provided within the bars. (Online version in colour.)

(c). Experimental series 2 procedure

The main finding of experiment 1 was that ants followed social information only when both a pheromone trail and information about a better food source are provided. To explore these findings in a more biologically realistic setting, we devised a second experiment. Training and tests were identical to treatment G in series 1, i.e. ants were trained to find poor (0.25 M) unscented sucrose at the end of a Y-maze. On the 5th visit the unrewarded arm was unscented but marked with a pheromone trail. In this experiment, the alternative Y-maze arm was present and accessible, but not rewarded, during training. Critically, additional food quality information was provided not via a drop of sucrose, but via nest-mates. On all visits the outgoing focal ant was allowed to walk into a small (5 × 5 cm) arena containing five nest-mates. On the four training visits, these nest-mates had previously been fed to satiety on 0.25 M sucrose. The focal ant was left in the arena for 2 min, and made on average 9.7 (s.d. 2.3) contacts with the fed ants. A contact was defined as the front half of the focal ant contacting the front half of any of the other ants. On several occasions (13% of visits, 47 out of 357) trophallaxis occurred. After this contacting phase the ant was allowed to continue towards the Y-maze. Thus, training was identical for all ants. Then, in half of the 5th (test) visit, the five contacting ants were fed to satiety on 1.5 M sucrose. In the other half of the trials on the 5th visit, these ants were fed with 0.25 M sucrose. Thus, ants experienced one of two treatments: 0.25 M sucrose from the contact ants during the final visit, or 1.5 M sucrose from the contact ants on the final visit.

(d). Testing for information transfer during ant contacts

To confirm that sugar solution, and thus information, was being transferred during these contacts, we used fluorescence microscopy to trace sugar flow between ants, as in [45]. A focal ant was allowed to find a 0.25 M sucrose solution on one arm of a Y-maze and return to the nest, as above. On her second outward journey, she was confined in a contact arena for 2 min with five nest-mates that had been fed 1.5 M sucrose containing 0.08 g l⁻¹ Rhodamine B. The focal ant was then removed, freeze killed, and the head, thorax, and abdomen separated and crushed between two microscope slides. The ant was then examined under green laser light (555 nm) using a Zeiss Axioplan-2 fluorescence microscope for the presence of Rhodamine B. A total of 30 ants were tested. Six control ants taken directly from the nest were also examined (figure 1). All fluorescence images are provided in the electronic supplementary material, S3.

(e). Statistical analysis

We analysed the data using generalized linear mixed models (GLMMs) using the LME4 package [46], with the ant's decision $d_{i,j}$ modelled by a binomial/logistic response, treatment and training side as fixed effects, and a random intercept for colony:

$$d_{i,j}=\mathrm{binom}\hspace{0.17em}({\pi }_{i,j})$$

$$\mathrm{logit}\hspace{0.17em}({\pi }_{i,j})=\mathrm{treatmen}{\mathrm{t}}_{i,j}+\mathrm{sid}{\mathrm{e}}_{i,j}+\mathrm{colon}{\mathrm{y}}_j\hspace{0.17em}$$

$$\mathrm{colon}{\mathrm{y}}_j=N(0,{\sigma }_{\mathrm{colony}}^2).$$

The effect of training side was added because initial data exploration revealed a leftward side bias, which is not uncommon in ants or other animals. The random intercept for colony adjusts for the expectation that ant behaviour might be more similar within colonies. The side variable was centred, so that all contrasts in the model are calculated for an average side value. We also calculate post hoc comparisons between all treatment groups, with Holm corrected p-values to control family-wise error rates. The main regression analysis (table 1) tests whether in each group ants predominantly follow their memory or not, while the post hoc analysis (electronic supplementary material, S1) allows differences between groups to be tested. We display standard errors for the regression table (table 1), and confidence intervals based on a parametric bootstrap, including fixed and random effects, for the average treatment effects on the response scale (figures 2 and 3; details see the electronic supplementary material, S1).

Figure 3.

Proportion of ants following their memory on a Y-maze rather than the alternate which was marked with trail pheromone. Ants encounter nest-mates fed on either 0.25 M (training quality) or 1.5 M (higher quality) sucrose on the way to the choice point. Note that in ca 30% of trials information transfer did not take place between the fed ants and the focal ant. The unadjusted proportion is given in the darker portion of the bars. The estimated adjustments are shown in the lighter portion of the bars, and are 0.69 for the 0.25 M treatment, and 0.35 for the 1.5 M treatment. These adjusted estimates are based on the strong assumption that the 70% successful information transfer figure we get from the fluorescence microscopy experiment is accurate, and should be treated with caution. Circle and whiskers are the modelled treatment effects and 95% CIs for the model estimate, and are not related to the adjusted proportion estimates. Asterisk represents a significant difference (p < 0.05) between the unadjusted groups. (Online version in colour.)

Because the initial decision (crossing a line 2 cm from the bifurcation) and the final decision (reaching the end of the arm) rarely differed (less than 5% of cases), we analysed only the final decision. Moreover, although we collected 10 consecutive decisions per ant, we decided to concentrate our analysis on the first decision. This avoids various statistical and conceptual issues, including the possibility of ants becoming frustrated in later trials owing to handling or repeated failures to find the food. However, we provide an extended analysis in the electronic supplementary material, S1, re-running all analyses of the main paper with all 10 decisions, testing additional for temporal changes in treatment effects.

All analyses were performed in R v.3.5.1 [47]. Model residuals and diagnostics were checked using the DHARMa package [48]. The dataset and the analysis are provided in the electronic supplementary material, S1 and S2, and available from the Dryad Digital Repository [49].

3. Results

Our first experiment demonstrates that the type of information available to a forager strongly influenced her decision of whether or not to follow her memory (table 1 and figure 2). In all treatments in which no or only one information source is available, most ants chose to follow their memory—significantly more than chance in treatments A-C. By contrast, in all treatments where both a pheromone trail and a droplet were offered, significantly more ants followed the pheromone trail (table 1). Qualitatively similar results are found when all 10 visits of the ants are analysed (see the electronic supplementary material, S1). There were only two notable differences between the analyses: firstly, memory following rates in treatment D (path odour + unscented droplet) becomes strongly significantly different from 0, equivalent to 50% following memory (estimate 1.38, equivalent to 79% following memory, p < 0.001); and secondly, the proportion of ants following their memory in treatment J (both information sources, ants trained to 1.5 M) was higher, and significantly different from chance (estimate 0.6, equivalent to 65% following memory, p = 0.02).

The presence of odours in the drop or on the path, and whether these were similar or different, had little effect of whether ants chose to follow their memories or the pheromone trail (figure 2). The critical comparison is treatment G (no scents in droplet or on the pheromone-laden path) versus treatment I (droplet and pheromone-laden path identically scented). Here, we expected the similarity in odour between the droplet and the marked path to drive increased pheromone following, but this did not occur (GLMM, z = 0.51, p > 0.05).

Similar results were found from experimental series 2, in which unambiguous information about available food quality was provided via contact with fed nest-mates. The behaviour of focal ants which contacted ants fed on 0.25 M was significantly different from those contacting 1.5 M-fed ants (GLMM, z = 2.31, p = 0.021; figure 3), with focal ants that contacted 0.25 M-fed nest-mates showing a higher tendency to follow their memory than focal ants that contacted nest-mates fed with 1.5 M. Note, however, that it is likely that not all focal ants received information from the fed nest-mates, making the reported proportion of memory following in the 1.5 M treatment an overestimate. In the fluorescent sucrose test for information transfer, only 70% (21 out of 30) of focal ants received marked food from their fed nest-mates. Thus, we should expect around 30% of our test subjects in experiment series 2 not to have received unambiguous food quality information, and thus follow their memory at a rate of around 60% (figure 1 group F ‘only pheromone'). Adjusting for this rate of information transfer failure, we estimate 68.7% of ants in the 0.25 M treatment followed their memory after information transfer, compared to 35.1% in the 1.5 M treatment (see the electronic supplementary material, S1 for our calculations). None of the control ants showed any signs of fluorescence. Additionally, we note that of the 21 ants which showed signs of florescence, 11 did not perform trophallaxis with the dye-fed nest-mates, implying that traces of dyed-sucrose smeared on the body of the fed ants can be transferred to the focal ant even without trophallaxis.

When considering all 10 visits, the results of experiment 2 are qualitatively unchanged, although a temporal pattern was noted, with pheromone following being higher in later visits, especially in the 0.25 M treatment (see the electronic supplementary material, S1).

4. Discussion

When social and private information sources conflict, animals usually put more weight on private information [16,17,19–27]. We found that while the presence of a pheromone trail (social information) draws some foragers away from following their memory, more than half nonetheless preferentially follow their private memory over a pheromone trail if they have no further information about the quality of other potential food sources (figure 2), or if this information implies the same low quality (figure 3). If outgoing foragers are informed about a better potential food source, however, either via finding a small drop of better food or via contacting nest-mates that have fed on such food, they follow the social pheromone trail information.

(a). Information type is a key driver of information use decisions

Information use strategies have previously been mainly divided into when and who strategies – when should information be used, and who should one listen to [9,50]. Here, we argue that what is an important question to ask as well—what type of information is conveyed? The information source's dimensions—the type of ‘data' provided—is extremely relevant to deciding how to respond to an information source. Pheromone trails and honeybee waggle-dances can, in fact, encode information about food quality: individual foragers only recruit to sufficiently high-quality food sources [51], and deposit more pheromone to food sources they perceive as of high value [36,37]. However, this information is extremely imprecise both owing to individual variation in responses to fixed food quality [36,37,52] and, for pheromone trails, owing to being ‘overwritten' by other ants or evaporating. Private information, and contacts with nest-mates, does provide more unambiguous and precise quality information. We propose that our observations can be explained by this difference in information quality, causing otherwise valuable social information to be neglected if it is uncertain. The neglect of ambiguous options, and the preference for certainty, are robust economic behaviours in humans, and have also been demonstrated in non-human primates [53–55]. In humans, these effects are termed ‘ambiguity aversion' and ‘the certainty effect', respectively.

Trail following in ants is often modelled in simple non-cognitive terms, with the proportion of ants following a trail acting as a function of trail strength [40,56,57]. Recently, trail following has been described using psychophysical functions, with the presence of conflicting or congruent memories decreasing or increasing pheromone following rates, respectively [7]. However, our results cannot be well explained by such pre-cognitive psychophysical models. Specifically, we show that the addition of directionless information (quality information) influences pheromone following. The decision to attend to pheromone trails or not appears to be an explicit decision, and psychophysical models of trail following seem to be insufficient to fully describe trail following behaviour.

(b). Information type asymmetry explains information use patterns described in other studies

Understanding information use strategies in terms of information asymmetry can explain previously puzzling findings. Leadbeater & Florent [58] unexpectedly found that bumblebees do not rate social information above personal experience, even when their personal experience becomes outdated. Bumblebees which had experienced a novel odour in the nest (social information) were just as unlikely to sample novel flowers scented with this odour as bees which did not experience the novel odour in the nest. Both groups continued choosing known-odour flowers, even when those flowers became unproductive. However, once the bees with the social information did, eventually, sample the novel flowers, they were much more likely to switch to the novel flowers than the group without social information. This is directly analogous to the situation we report: neither private sensory information providing reliable information in the dimension of quality (sampling the novel flowers), nor social information (smelling the novel odour in the nest) is sufficient to make bees change their choices. However, when both are provided together (here, after the first sampling of the novel flower), bees are willing to switch.

Similarly, understanding information use in terms of information quality can explain situations in which social information does override personal information. In such cases, social information does provide unambiguous quality information. For example, Dunlap et al. [11] and Smolla et al. [59] found that social information alone can drive decision making during flower choice in bumblebees, and even outweighed private information when the two competed [11]. This result could be understood through the psychological concept of blocking [18]. However, differences in information content between the two sources can also explain these results: in these studies, the social information is based on making associations between a social cue (the presence of a conspecific) and the direct personal experience of the forager with the flower quality [18]. There is thus no quality ambiguity. Indeed, if ambiguity exists, it is lower in the social information: the forager not only has her own experience of finding the flower acceptable, but also the presence of a conspecific suggests that she too finds the flower acceptable, resulting in more information about quality for the socially advertised flower. Likewise, Josens et al. [60] describe how trophallactic interactions with a nest-mate providing scented, untainted food can cause foraging ants to accept similarly scented food tainted with poison, which they would otherwise reject. Here again, the social information component is based on direct sensory experience of the food (trophallaxis), and the reasons the ants overweigh this information over their own later sensory information is, as above, owing to the concurrent presence of social information and personal sensory information.

(c). Direct transmission of quality information to foragers enables strategic information use

Outgoing ants repeatedly encounter returning ants on a trail. We have shown that during these contacts, small amounts of food, and thus information, can be transferred to the outgoing ant. Information transfer can be either via trophallaxis, or by detecting food remains smeared on the fed ant while it was feeding. We also observed transfer of food to active foragers from ants inside the nest, as reported recently in another species [45]. Similarly, dancing honeybees often engage in trophallaxis with followers. Ants receiving trophallaxis may even respond as if they have themselves successfully foraged and deposit trail pheromone [61]. Information gained during such contacts, especially trophallaxis, seems to be particularly well attended to—even more so than information gained while foraging [44,60]. Such information gained from nest-mates can drive strategic information use: an outgoing forager encountering returning ants which collected poorer food may continue exploiting her own known food source, while outgoing foragers encountering more successful returning ants may begin to follow pheromone trails, thus playing a ‘copy if better' strategy. The ‘copy if better' strategy is uncommon in biology, as direct quality comparison is not usually possible and accurate social comparison may require advanced cognitive abilities [62]. If a forager has only found poor quality food, and encounters only other foragers with equally poor food, it may be more likely to begin scouting, following an ‘innovate if dissatisfied' strategy. Thus, provision of unambiguous quality information may well be a major role of worker interaction on a trail. Sharing food-associated cues, such as odour [43,60], is certainly an important role of trophallaxis, although we could not find any effect of odour in the current experiment (figure 2).

The ability to redirect workers to the highest quality underexploited food source will aid colonies in securing resources from competition [63] and most efficiently use a colony's limited processing and storage capacity. More broadly, this system produces an elegant solution to the allocation of producer-scrounger roles within a colony: some workers must produce information via scouting, which can be ‘scrounged' by other workers, who copy the producers' choices. Models of competitively interacting social animals show that, as the frequency of social learners increases, the value of social learning declines, until the success of producers and scroungers is identical [64]. However, this will not maximize total success rates, which in social insects is the salient driver of selection. By deciding whether to be an information producer or scrounger based on the relative success of their fellow foragers, social insects could adapt their producer-scrounger balance to suit the current foraging conditions.

(d). Workers in a colony may synthesize and disseminate foraging information

Trophallactic networks within a nest are complex [45]. Food returned to the nest is distributed and mixed among the ants in the nest, producing an average ‘reading’ of the food being returned by all foragers. Returning foragers interacting with these ants may then compare this reading to their currently exploited food source. If they find their own food source to be worse, they may begin following social information. In line with this reasoning, ants which receive high-quality food via trophallaxis are less willing to accept medium quality food, and vice versa for ants receiving poor quality food [37]. Honeybee foragers experiencing an influx of higher-quality nectar into the hive are more likely to switch from nectar foraging to pollen foraging [65]. It seems likely that social insect foragers are very well informed about the various food qualities available in their environment, and scout, follow recruitment signals, or exploit known food sources adaptively depending on this information. Using newly developed fluorescent imaging technologies allows the flow of food via trophallaxis in a colony to be tracked and quantified [45], opening the possibility of empirically testing these suggestions.

5. Conclusion

Here, we explored how three different information sources, one private and two public, interacted. It appears that the type of information in an information source, and its ambiguity, is a key driver in whether it will be heeded or not. Some information sources may be highly informative in one dimension, and still be neglected if they are uninformative along another dimension. We propose that as well as considering when and who strategies for information use, what is an important question to ask: what type of information is provided? Animals demonstrably use complex mechanisms to decide which information to attend to, and integrate information from a wide variety of sources in order to do so.

Data accessibility

The raw data on which this work is based is provided in the electronic supplementary material, S2. The complete analysis pathway and all data required to reproduce are available from the Dryad Digital Repository: https://doi.org/10.5061/dryad.t63gm4r [49].

Authors' contributions

T.J.C. conceived of, planned and coordinated the study, and wrote the manuscript. A.-L.H. collected the data for the ant-contact and florescence experiments. J.B. collected the data for the droplet presentation experiment, and assisted in planning and experimental design. F.H. carried out the statistical analysis. All authors edited the text and gave final approval for the manuscript.

Competing interests

We declare we have no competing interests.

Funding

This work was funded by a DFG grant to T.J.C. (CZ 237/1-1).