Bumblebees retrieve only the ordinal ranking of foraging options when comparing memories obtained in distinct settings

  1. Cwyn Solvi
  2. Yonghe Zhou
  3. Yunxiao Feng
  4. Yuyi Lu
  5. Mark Roper
  6. Li Sun
  7. Rebecca J Reid
  8. Lars Chittka
  9. Andrew B Barron
  10. Fei Peng

  • Curated by eLife

    eLife logo

    Evaluation Summary:

    The authors investigate what type and degree of information (either absolute, relative, or a weighted combination of both) is used by bumblebees when retrieving the value of an item. There is recent evidence in humans and birds that suggests that these organisms use a combination of absolute memories and remembering of subjective ranking in these tasks. The authors conclude that bumblebees indeed use remembered ranking, but that they seem not to be able to retain (or at least utilise) absolute property information for very long. The absence of relevant work in invertebrates would make this study a potentially valuable addition to the scientific literature.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #3 agreed to share their name with the authors.)

Reviewed by

Read the full article

Listed in

Log in to save this article

Abstract

Are animals’ preferences determined by absolute memories for options (e.g. reward sizes) or by their remembered ranking (better/worse)? The only studies examining this question suggest humans and starlings utilise memories for both absolute and relative information. We show that bumblebees’ learned preferences are based only on memories of ordinal comparisons. A series of experiments showed that after learning to discriminate pairs of different flowers by sucrose concentration, bumblebees preferred flowers (in novel pairings) with (1) higher ranking over equal absolute reward, (2) higher ranking over higher absolute reward, and (3) identical qualitative ranking but different quantitative ranking equally. Bumblebees used absolute information in order to rank different flowers. However, additional experiments revealed that, even when ranking information was absent (i.e. bees learned one flower at a time), memories for absolute information were lost or could no longer be retrieved after at most 1 hr. Our results illuminate a divergent mechanism for bees (compared to starlings and humans) of learned preferences that may have arisen from different adaptations to their natural environment.

Article activity feed

  1. Version published to 10.7554/elife.78525 on eLife
  2. eLife

    Author Response

    Reviewer #1 (Public Review):

    The article by Solvi and colleagues aims to investigate what type and degree of information (either absolute, relative, or a weighted combination of both) is used by bumblebees when retrieving the value of an item. The authors reported recent evidence in humans and birds that suggest they seem to use a combination of absolute memories and remembering of subjective ranking, and an absence of relevant studies for other species, including invertebrates. Thus, the authors conducted four different experiments to study what type of information is guiding the decision of bumblebees when facing different qualitative and quantitative comparisons.

    In the first two experiments, the authors reported the use of relative ranking of stimuli instead of a memory of their absolute value. According to the authors, these results are confirmed by experiment three, where bees were presented with two equally-ranked choices which, in fact, were not treated as different by bees. In the last experiment, bumblebees showed a preference for the highest rank item.

    Despite the presentation of well-designed experiments, the conclusions that bumblebees are using only memories of ordinal comparisons, thus showing a different strategy with respect to humans and birds, seems to not be fully supported by the results. The behaviour on the first two experiments, for instance, could be explained by a recency effect, where the higher item of the last comparison is better retrieved (the work of Giurfa on transitive inferences in bees was not mentioned, though is relevant here). Furthermore, in the last experiment, bumblebees could not have used an ordinal ranking; their choice for the higher-ranking item could be based on its higher absolute quantitative value in terms of sucrose solution.

    We’re sorry for not being clearer in our descriptions in our original submission. In each of the first three experiments, the order of sessions in which the different pairs of sucrose concentrations had been used were counterbalanced. For example, in experiment 1, half of the bees experienced 45 vs 30 first and 30 vs 20 second, and half of the bees experienced 30 vs 20 first and 45 vs 30 second. Further, our GLM results show that the order of training did not affect bees’ preferences. Therefore, a recency effect cannot explain the results of experiments 1 or 2 (or 3). We now highlight this on lines 101 - 105 in the Results and in each of the Experimental descriptions in the Methods, and explain that the GLMs showed no effect of these factors on lines 424 - 426 of Methods.

    With regards to the experiment 6 (last experiment in our original submission), there is no reason bees could not have used ordinal ranking. However, they also could have used absolute memories. We apologise for not making the rationale for and interpretation of experiment 6 clearer. The rationale was to determine how our results spoke to a situation that was more ecologically relevant for a bumblebee. In response to Reviewer #3’s concerns, we have now added new data from an experiment which also helps better explain both our rationale and our interpretation of experiment 6. We discuss this in more detail now in the revised manuscript. In short, bumblebees must use absolute properties, otherwise they would not be able to discriminate or rank any two sequentially visited flowers. However, our results suggest that they only retain (or only utilise memory of) absolute information for a short period of time (a few minutes). Despite this, experiment 6 suggests that in normal foraging situations, bees’ preferences for the highest rewarding flowers will not be affected. This is because, in the wild, bumblebees could commonly experience short time intervals (a few minutes) between flowers, which would allow them to compare each flower’s absolute information and encode ranking information. We discuss the new data on lines 204 - 233 and add clarification for experiment 6 on lines 249 - 261.

    The different behaviours and strategies used by bees here could be better explained by differences in the experimental task proposed, rather than supporting a general statement about the evolution of different strategies in comparison to other species.

    We hope that our explanations and clarifications to your above comments and to the other referees’ comments remedy this concern.

    Reviewer #2 (Public Review):

    This manuscript analyzes if bumblebees choose feeding options based on their absolute or relative remembered subjective value. The experiments relate to previous work done in starlings where comparable questions were raised (1). The design used in the four experiments presented is elegant and provides support for the conclusion that bees guide their choices by remembered ranking of feeders instead of focusing on their absolute rewards. Bees preferred the options that were ranked higher within each experimental context experienced, irrespective of the absolute reward they provided. As a consequence, they even preferred a sucrose solution of low concentration (15%) to one that was more profitable (30%), simply because the former was experienced together with a poorer alternative (10%) while the latter was experienced together with a more attractive alternative (45%) (Exp. 2). All four experiments provide results that are consistent with the hypothesis that contextual ranking is essential to determine the bees' choices.

    Thank you for the kind and supportive words.

    Three main points require consideration to render this manuscript even more attractive than what it is already.

    1. The experiments involved in all cases four different colors and different sucrose concentrations (range: 5 - 45 % w/w). An essential requisite of these experiment is that bees should be able to discriminate the options provided, both in terms of color and in terms of reward quality. Asking about ranking or absolute value makes no sense if bees cannot distinguish, say, 15% from 10%, or yellow from orange, and so on. The authors are obviously aware of this point as they mention it explicitly (lines 267-269). Yet, although they mentioned that they verified this point, the only experimental proof available is provided in Fig. Suppl. 4, where a single comparison (from the many possible) was tested; the discrimination test provided involved blue and yellow, which were associated in a balanced way with the two highest sucrose concentrations used, 45% and 30%. In terms of color information, the choice involved the colors that were easy to distinguish (see their loci in the color hexagon). Yet, what about the other colors? Could they be equally well discriminated? Probably not, because some occupied very close loci in the hexagon. Admittedly, the tests B vs. C involved similar colors (yellow vs. orange) and bees showed significant preferences supporting the presence of color discrimination. Yet, no information is available for yellow and green and other color combinations assayed. Even more important would be to show that bees rank the different sucrose solutions differently, which is not clear in all cases. Concentrations were chosen following theoretical considerations based on Weber's law (2), but do bees really respond differently to them? Providing an experimental assessment of this question would be important.

    The perceptual distances between colours used in our experiments ranged from 0.141-0.333 hexagon units, which are intermediate to large colour distances that can be trained to high asymptotic levels of discrimination within 50-60 visits [Dyer and Chittka 2004, doi: 10.1007/s00359-003-0475-2]. Our training procedure involved 50 drinking experiences for each colour (100 drinking experiences in total), which is enough for clear discrimination by bees. Further, we adopted a counterbalanced paradigm, for which the results of each group are displayed in the presented figures. The visualisation of the individual data points indicate there was no significant difference between colour combinations, but more importantly, the GLM statistical analyses show that colour combinations had no effect on bees’ preferences (reported on lines 424 – 426). Finally, we have added data and description of new experiments using orange and yellow flowers and which show that bumblebees are able to quickly and easily discriminate the orange and yellow colours (the shortest hexagon loci distances for the colours used) even when not learned side-by-side (new Figure 2). Importantly, we used sucrose concentration pairs with much greater differences compared to what bees are capable of discriminating behaviourally (bees can tell differences in sucrose concentration as low as 1.5% [Whitney et al., 2008, doi:10.1007/s00114-008-0393-9]) and neurophysiologically (Miriyala et al., 2018, doi: 10.1016/j.cub.2018.03.070). This is now discussed on lines 348 - 353.

    1. Figures B, D, and F are of fundamental importance to draw conclusions about the strategy used by the bees in the three adjacent experiments. Yet, the kind of representation chosen by the authors does not help to follow their conclusions. Firstly, it is not clear what the data points represent. If, for instance, in Fig. 1B, 40 bees were tested (line 247), how many bees per combination were tested (only one combination is mentioned in line 249)? Moreover, given that bees were tested with B vs. C, and if I guess correctly, there are ca. 10 data points per combination, what do these 10 proportions represent? How were these values computed? I could not find this information in the Methods section. No description of the test methodology is provided for Experiments 1 to 3. Moreover, data points in Figs B, D, F are barely visible and appear clustered around 50% in several cases, thus casting doubt on the reported significance of the comparisons. This needs to be improved by means of visible and clear graphic displays. The same kind of consideration can be applied to Fig. 2B, even if the results are clearer.

    We apologise for the lack of clarity and missing information. We now note in the figure legend that each filled circle represents the proportion of choices for a particular option by an individual bumblebee (10 individuals per group). We have now added more description on the test methodology and analyses for experiments 1-3 on lines 101 – 105, 126-127, 380 - 384. We have also increased the size of the individual data points in each of the figures for better clarity.

    1. A final point relates to results obtained in a different experimental framework but which asks whether animals can rank and order in transitive terms experienced alternatives. A considerable amount of work in the field of experimental psychology has addressed the question of transitive inferences in many species (3-12), and even in bees and wasps (13, 14). In these studies, animals are trained with premise pairs presenting different reinforcement outcomes (e.g. A+ B-/B+ C-/C+ D-/D+ E-) to determine if they establish relative rankings (A ˃ B ˃ C ˃ D ˃ E), or on the contrary use associative learning of absolute reinforcement outcomes (in which case, A ˃ B = C = D ˃ E). To determine the strategy followed by animals, they are tested with a non-overlapping pair never experienced during the training (B vs. D). In the first case, animals prefer B to D while in the second case they choose equally between both options. There might be, therefore, some parallels or contact points between these experiments and the experiments reported in this manuscript. Could the authors discuss these parallels and provide a broader view of absolute vs. relative remembered subjective value?

    Thank you for this suggestion. We believe that tests for transitive inference only resemble our methods superficially. However, we do feel that because of this we should provide a brief explanation as to how and why they are fundamentally different. We have now included a paragraph in the Introduction on lines 73 - 89.

    Reviewer #3 (Public Review):

    The central conclusion of this beautiful experimental study is that bumblebees prefer flowers on the basis of their remembered ranking in their context, but are insensitive to their absolute properties. Thus, let's say that there 4 flower types, ranked as follows in nectar concentration: A>B>C>D. However, when the bee learns about these flowers, it does in either of two 'contexts', populated as follows: A & B, or C & D. Thus, the bee experiences that B is the worse option in the context in which it is found, and C is the better one in its own context. If, at a later time, the bee has to make a novel choice, this time between B and C, its memory for ranking leads it to prefer C over B, while its (putative) memory for nectar concentration should favour B over C. The authors find, in a variety of different treatments, evidence for the influence for ranking, but they do not find any evidence for sensitivity to absolute properties (i.e., concentration).

    Thank you for the complimentary sentiments.

    One difficulty that permeates the argument is the ubiquitous difficulty in proving the null hypothesis as true: lack of significant evidence for a putative effect in one or a few experiments, does not mean reliable absence of the effect.

    We appreciate this thought, and we hope that the additional experiments on absolute information usage, together with the original experiments, might collectively form a clearer argument here.

    Another difficulty is that in my view memory for absolute properties was not given a full chance: bees were always trained in situations where both dimensions (concentration and ranking) were present. In such situations, they preferentially used ranking. However, to learn ranking between flower types in sequential encounters, they must remember the absolute properties, so that in each encounter they contrast the present flower with the memory for others. Say the bee encounters a type B flower. How does it store its ranking if it doesn't remember the properties of A at all? To take this objection into account and still maintain the claim, it is necessary to say that it remembers the properties of A when in the A & B context, but it erases it from memory when in the context B & C.

    Neglecting memory for concentration may be an overshadowing effect. Overshadowing is known in learning studies, and it means that, when more than one cue is paired with an outcome, the most salient between them may reduce learning about the predicting power of the other. In this case, bees may remember and use concentration when trained in contexts where there is only flower type, so that there is no chance of using ranking, and then offered choices between pairs of them. In this case, the bees would not have access to ranking, so that there would be a stronger opportunity for absolute memory to manifest itself.

    Thank you for these suggestions. We apologise for not making our arguments clearer in our first submission. Yes, absolute information must be used at some level, otherwise no ranking, or even discrimination, could take place. Our claim is that while bumblebees must detect and use absolute information to compare flowers, our experiments show they do not retain (or utilise memory of) this information for very long (i.e. for much longer than several minutes). We now make this clearer throughout the manuscript, e.g. on lines 31 - 36 and 204 - 233. We have also added new experiments which show that when bees were trained with each of two flowers alone and then tested together, absolute information can be used if visits to the different flowers are separated by only a few minutes, but not if separated by an hour. This suggests that absolute information is retained and used by bumblebees as short-, but not mid-term memories (Menzel 2001), in order to make comparisons of options and rank them. Please see lines 385 - 401 for a detailed description of these experiments, as well as Figure 2.

    In experiment 4, during training, they could move between two zones representing the 'contexts', each with 2 flower types, and they were then given choices between the 4 types, rather than just binary choices as previously. In this case, the bees did prefer the top-quality flower type (type A), which is consistent with memory for absolute concentration and with ranking, because A offered the highest concentration of the 4-type context. Why this happened is not clear, but it indicates that the context of choice may be crucial. It is known from other studies that the number of options at the time of choice can be very influential. For instance, in one study, it was shown that starlings appeared to be risk prone when offered a binary choice and risk averse when offered a trinary choice, even if the choices were all intermingled in the same sessions. In any case, this experiment raises doubts as to the claimed insensitivity to memory for nectar concentration. Another possibility is that the separation between contexts in this experiment (a partially avoidable wall) was not extreme as in the previous ones, so that the bees could now establish a ranking among the 4 types because they were all encountered intermingled to an extent.

    Again, we apologise for the lack of clarity in our original argument. We have made clear in the manuscript that bumblebees are not insensitive to absolute metrics, and indeed require them to distinguish between sequentially visited flowers. We have also added new data and descriptions of experiments which we believe help set the stage for and help interpret the results of experiment 6 (previously experiment 4). The newly added experiments (experiments 4 and 5) show that when bees learn flowers in isolation, and therefore have no ranking information available, they still do not retain or utilise memory of absolute information in a new context, unless the temporal separation between flower experiences is short (a few minutes). The results of experiment 6 (previously experiment 4) essentially help show that in a more ecologically realistic scenario (what bees normally experiment in the wild), the time between flower visits are short enough so that absolute information can be compared and used to rank flowers. We now explain this better on lines 204 – 233 and 249 - 260.

    There is one potential mechanism that may also be discussed. It is known from other species, that state at the time of learning influences subjective value of alternatives. To explain this effect I will exemplify the problem with a non-eusocial consumer. Say that food sources B and C are of equal caloric value. Say, further, that B is encountered when the subject is less food deprived than when it encounters C. Then the hedonic (conditioning) power of B will be lower, because it causes a smaller improvement in fitness (this was Daniel Bernoulli's argument regarding the concept of utility). In animal studies this effect is called State-Dependent Valuation Learning (SDVL). Since in the present experiments the context A & B was richer than the context C & D, the bees would have been in a consequently more favourable state (maybe carrying bigger sugar loads), so that each encounter with B would cause a smaller improvement than each encounter with C. This effect is totally different from remembering the ranking of flower types. The two alternative explanations for preference of C over B (ranking and SDVL) can, fortunately, be confronted because it is possible to change the state of the bees by a common 3rd source that could be used to equate or manipulate the average richness of the contexts.

    Thanks for this suggestion. Although we believe SDVL cannot account for bumblebees’ behaviours as well as ordinal ranking, we agree that it would be valuable to discuss SDVL with reference to some of the literature on the subject. We have now added description of SDVL to Table 1 on lines 156 – 162 and added a paragraph in the Discussion on lines 283 – 290.

    All the reasons mentioned above should make it clear that this reviewer finds the study of very great interest and much merit, but considers that the conclusion for exclusive impact of ranking on preference should be tempered, or at least defended more strongly against these doubts.

    Thanks again for the valuable and positive critique.

  3. eLife

    Evaluation Summary:

    The authors investigate what type and degree of information (either absolute, relative, or a weighted combination of both) is used by bumblebees when retrieving the value of an item. There is recent evidence in humans and birds that suggests that these organisms use a combination of absolute memories and remembering of subjective ranking in these tasks. The authors conclude that bumblebees indeed use remembered ranking, but that they seem not to be able to retain (or at least utilise) absolute property information for very long. The absence of relevant work in invertebrates would make this study a potentially valuable addition to the scientific literature.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #3 agreed to share their name with the authors.)

  4. eLife

    Reviewer #1 (Public Review):

    The article by Solvi and colleagues aims to investigate what type and degree of information (either absolute, relative, or a weighted combination of both) is used by bumblebees when retrieving the value of an item. The authors reported recent evidence in humans and birds that suggest they seem to use a combination of absolute memories and remembering of subjective ranking, and an absence of relevant studies for other species, including invertebrates.

    Thus, the authors conducted four different experiments to study what type of information is guiding the decision of bumblebees when facing different qualitative and quantitative comparisons.

    In the first two experiments, the authors reported the use of relative ranking of stimuli instead of a memory of their absolute value. According to the authors, these results are confirmed by experiment three, where bees were presented with two equally-ranked choices which, in fact, were not treated as different by bees. In the last experiment, bumblebees showed a preference for the highest rank item.

    Despite the presentation of well-designed experiments, the conclusions that bumblebees are using only memories of ordinal comparisons, thus showing a different strategy with respect to humans and birds, seems to not be fully supported by the results. The behaviour on the first two experiments, for instance, could be explained by a recency effect, where the higher item of the last comparison is better retrieved (the work of Giurfa on transitive inferences in bees was not mentioned, though is relevant here). Furthermore, in the last experiment, bumblebees could not have used an ordinal ranking; their choice for the higher-ranking item could be based on its higher absolute quantitative value in terms of sucrose solution.

    The different behaviours and strategies used by bees here could be better explained by differences in the experimental task proposed, rather than supporting a general statement about the evolution of different strategies in comparison to other species.

  5. eLife

    Reviewer #2 (Public Review):

    This manuscript analyzes if bumblebees choose feeding options based on their absolute or relative remembered subjective value. The experiments relate to previous work done in starlings where comparable questions were raised (1). The design used in the four experiments presented is elegant and provides support for the conclusion that bees guide their choices by remembered ranking of feeders instead of focusing on their absolute rewards. Bees preferred the options that were ranked higher within each experimental context experienced, irrespective of the absolute reward they provided. As a consequence, they even preferred a sucrose solution of low concentration (15%) to one that was more profitable (30%), simply because the former was experienced together with a poorer alternative (10%) while the latter was experienced together with a more attractive alternative (45%) (Exp. 2). All four experiments provide results that are consistent with the hypothesis that contextual ranking is essential to determine the bees' choices.

    Three main points require consideration to render this manuscript even more attractive than what it is already.

    1. The experiments involved in all cases four different colors and different sucrose concentrations (range: 5 - 45 % w/w). An essential requisite of these experiment is that bees should be able to discriminate the options provided, both in terms of color and in terms of reward quality. Asking about ranking or absolute value makes no sense if bees cannot distinguish, say, 15% from 10%, or yellow from orange, and so on. The authors are obviously aware of this point as they mention it explicitly (lines 267-269). Yet, although they mentioned that they verified this point, the only experimental proof available is provided in Fig. Suppl. 4, where a single comparison (from the many possible) was tested; the discrimination test provided involved blue and yellow, which were associated in a balanced way with the two highest sucrose concentrations used, 45% and 30%. In terms of color information, the choice involved the colors that were easy to distinguish (see their loci in the color hexagon). Yet, what about the other colors? Could they be equally well discriminated? Probably not, because some occupied very close loci in the hexagon. Admittedly, the tests B vs. C involved similar colors (yellow vs. orange) and bees showed significant preferences supporting the presence of color discrimination. Yet, no information is available for yellow and green and other color combinations assayed. Even more important would be to show that bees rank the different sucrose solutions differently, which is not clear in all cases. Concentrations were chosen following theoretical considerations based on Weber's law (2), but do bees really respond differently to them? Providing an experimental assessment of this question would be important.

    2. Figures B, D, and F are of fundamental importance to draw conclusions about the strategy used by the bees in the three adjacent experiments. Yet, the kind of representation chosen by the authors does not help to follow their conclusions. Firstly, it is not clear what the data points represent. If, for instance, in Fig. 1B, 40 bees were tested (line 247), how many bees per combination were tested (only one combination is mentioned in line 249)? Moreover, given that bees were tested with B vs. C, and if I guess correctly, there are ca. 10 data points per combination, what do these 10 proportions represent? How were these values computed? I could not find this information in the Methods section. No description of the test methodology is provided for Experiments 1 to 3. Moreover, data points in Figs B, D, F are barely visible and appear clustered around 50% in several cases, thus casting doubt on the reported significance of the comparisons. This needs to be improved by means of visible and clear graphic displays. The same kind of consideration can be applied to Fig. 2B, even if the results are clearer.

    3. A final point relates to results obtained in a different experimental framework but which asks whether animals can rank and order in transitive terms experienced alternatives. A considerable amount of work in the field of experimental psychology has addressed the question of transitive inferences in many species (3-12), and even in bees and wasps (13, 14). In these studies, animals are trained with premise pairs presenting different reinforcement outcomes (e.g. A+ B-/B+ C-/C+ D-/D+ E-) to determine if they establish relative rankings (A ˃ B ˃ C ˃ D ˃ E), or on the contrary use associative learning of absolute reinforcement outcomes (in which case, A ˃ B = C = D ˃ E). To determine the strategy followed by animals, they are tested with a non-overlapping pair never experienced during the training (B vs. D). In the first case, animals prefer B to D while in the second case they choose equally between both options. There might be, therefore, some parallels or contact points between these experiments and the experiments reported in this manuscript. Could the authors discuss these parallels and provide a broader view of absolute vs. relative remembered subjective value?

    References
    1. L. Pompilio, A. Kacelnik, Context-dependent utility overrides absolute memory as a determinant of choice. Proc Natl Acad Sci U S A 107, 508-512 (2010).
    2. K. L. Akre, S. Johnsen, Psychophysics and the evolution of behavior. Trends Ecol Evol 29, 291-300 (2014).
    3. E. L. Maclean, D. J. Merritt, E. M. Brannon, Social complexity predicts transitive reasoning in prosimian primates. Animal Behaviour 76, 479-486 (2008).
    4. L. Grosenick, T. S. Clement, R. D. Fernald, Fish can infer social rank by observation alone. Nature 445, 429-432 (2007).
    5. Y. M. C. G. Paz, A. B. Bond, A. C. Kamil, R. P. Balda, Pinyon jays use transitive inference to predict social dominance. Nature 430, 778-781 (2004).
    6. H. Markovits, C. Dumas, Can pigeons really make transitive inferences? Journal of Experimental Psychology: Animal Behavior Processes 18, 311-312 (1992).
    7. G. Jensen, Y. Alkan, F. Munoz, V. P. Ferrera, H. S. Terrace, Transitive inference in humans (Homo sapiens) and rhesus macaques (Macaca mulatta) after massed training of the last two list items. J Comp Psychol 131, 231-245 (2017).
    8. O. F. Lazareva, E. A. Wasserman, Transitive inference in pigeons: measuring the associative values of Stimuli B and D. Behav Processes 89, 244-255 (2012).
    9. A. B. Bond, C. A. Wei, A. C. Kamil, Cognitive representation in transitive inference: a comparison of four corvid species. Behav Processes 85, 283-292 (2010).
    10. M. Vasconcelos, Transitive inference in non-human animals: an empirical and theoretical analysis. Behav Processes 78, 313-334 (2008).
    11. J. J. Bryson, J. C. Leong, Primate errors in transitive 'inference': a two-tier learning model. Anim Cogn 10, 1-15 (2007).
    12. A. B. Bond, A. C. Kamil, R. P. Balda, Social complexity and transitive inference in corvids. Animal Behaviour 65, 479-487 (2003).
    13. E. A. Tibbetts, J. Agudelo, S. Pandit, J. Riojas, Transitive inference in Polistes paper wasps. Biol. Lett. 15, 20190015 (2019).
    14. J. Benard, M. Giurfa, A test of transitive inferences in free-flying honeybees: unsuccessful performance due to memory constraints. Learn Mem 11, 328-336 (2004).

  6. eLife

    Reviewer #3 (Public Review):

    The central conclusion of this beautiful experimental study is that bumblebees prefer flowers on the basis of their remembered ranking in their context, but are insensitive to their absolute properties. Thus, let's say that there 4 flower types, ranked as follows in nectar concentration: A>B>C>D. However, when the bee learns about these flowers, it does in either of two 'contexts', populated as follows: A & B, or C & D. Thus, the bee experiences that B is the worse option in the context in which it is found, and C is the better one in its own context. If, at a later time, the bee has to make a novel choice, this time between B and C, its memory for ranking leads it to prefer C over B, while its (putative) memory for nectar concentration should favour B over C. The authors find, in a variety of different treatments, evidence for the influence for ranking, but they do not find any evidence for sensitivity to absolute properties (i.e., concentration).

    One difficulty that permeates the argument is the ubiquitous difficulty in proving the null hypothesis as true: lack of significant evidence for a putative effect in one or a few experiments, does not mean reliable absence of the effect.

    Another difficulty is that in my view memory for absolute properties was not given a full chance: bees were always trained in situations where both dimensions (concentration and ranking) were present. In such situations, they preferentially used ranking. However, to learn ranking between flower types in sequential encounters, they must remember the absolute properties, so that in each encounter they contrast the present flower with the memory for others. Say the bee encounters a type B flower. How does it store its ranking if it doesn't remember the properties of A at all? To take this objection into account and still maintain the claim, it is necessary to say that it remembers the properties of A when in the A & B context, but it erases it from memory when in the context B & C.

    Neglecting memory for concentration may be an overshadowing effect. Overshadowing is known in learning studies, and it means that, when more than one cue is paired with an outcome, the most salient between them may reduce learning about the predicting power of the other. In this case, bees may remember and use concentration when trained in contexts where there is only flower type, so that there is no chance of using ranking, and then offered choices between pairs of them. In this case, the bees would not have access to ranking, so that there would be a stronger opportunity for absolute memory to manifest itself.

    In experiment 4, during training, they could move between two zones representing the 'contexts', each with 2 flower types, and they were then given choices between the 4 types, rather than just binary choices as previously. In this case, the bees did prefer the top-quality flower type (type A), which is consistent with memory for absolute concentration and with ranking, because A offered the highest concentration of the 4-type context. Why this happened is not clear, but it indicates that the context of choice may be crucial. It is known from other studies that the number of options at the time of choice can be very influential. For instance, in one study, it was shown that starlings appeared to be risk prone when offered a binary choice and risk averse when offered a trinary choice, even if the choices were all intermingled in the same sessions. In any case, this experiment raises doubts as to the claimed insensitivity to memory for nectar concentration. Another possibility is that the separation between contexts in this experiment (a partially avoidable wall) was not extreme as in the previous ones, so that the bees could now establish a ranking among the 4 types because they were all encountered intermingled to an extent.

    There is one potential mechanism that may also be discussed. It is known from other species, that state at the time of learning influences subjective value of alternatives. To explain this effect I will exemplify the problem with a non-eusocial consumer. Say that food sources B and C are of equal caloric value. Say, further, that B is encountered when the subject is less food deprived than when it encounters C. Then the hedonic (conditioning) power of B will be lower, because it causes a smaller improvement in fitness (this was Daniel Bernoulli's argument regarding the concept of utility). In animal studies this effect is called State-Dependent Valuation Learning (SDVL). Since in the present experiments the context A & B was richer than the context C & D, the bees would have been in a consequently more favourable state (maybe carrying bigger sugar loads), so that each encounter with B would cause a smaller improvement than each encounter with C. This effect is totally different from remembering the ranking of flower types. The two alternative explanations for preference of C over B (ranking and SDVL) can, fortunately, be confronted because it is possible to change the state of the bees by a common 3rd source that could be used to equate or manipulate the average richness of the contexts.

    All the reasons mentioned above should make it clear that this reviewer finds the study of very great interest and much merit, but considers that the conclusion for exclusive impact of ranking on preference should be tempered, or at least defended more strongly against these doubts.

  7. Version published to 10.1101/2022.04.05.487177v1 on bioRxiv