How Human Are Insects, and Does It Matter?

綜合論述

台灣昆蟲 31: 85-99 (2011) Formosan Entomol. 31: 85-99 (2011)

How Human Are Insects, and Does it Matter?

Thomas F. Döring1, and Lars Chittka2*

1 The Organic Research Centre, Elm Farm, Hamstead Marshall, RG 200HR, UK 2 Queen Mary University London, School of Biological and Chemical Sciences, Biological and Experimental Psychology Group, Mile End Road, London E1 4NS, UK.

ABSTRACT

Over the past two decades, a wealth of information on cognitive function in insects has been generated and this work has been tremendously useful in pinpointing just how much can be done with how little neural circuitry in miniature brains. However, other recent claims on the occurance of ‘teaching’, ‘culture’, ‘consciousness’, or ‘personality’ in the insects seem to apply a relatively restrictive top-down approach, where researchers set out to discover human-like behaviours in animals. This approach is prone to terminological ambiguities, because terms taken from the domain of human experience often invoke more complex connotations than the restricted criteria used to test the phenomena in the animal world would permit. It also bears the risk of circularities as a consequence of selecting and shaping definitions and test criteria. Finally, the hypothesis-driven approach may be vulnerable to research biases, such as to predominantly report positive results or to mainly investigate “clever” animals. We suggest that the current approach needs to be supplemented by a complementary modus operandi. In a more exploratory, bottom-up ‘ethomics’ approach, psychological constructs would only emerge after multivariate data analysis of a broader range of animal behaviours. It is also necessary to consider the process of how we arrive at definitions as part of the research methods.

Key words: ambiguity, arbitrariness, behaviour, cognition, methods, terminology

Introduction might be considered uniquely human domains: agriculture, slavery, territorial wars, castes, Some of the attraction for scientists to division of labour, consensus building, a the insects, especially the social ones, is symbolic language, and teeming cities undoubtedly that they have evolved a with fantastic architecture (Lindauer, number of feats that, to a non-biologist, 1955; Frisch, 1967; Hölldobler and Wilson,

*Corresponding email: [email protected] How Human Are Insects, and Does it Matter? 85 1990). However, there is perhaps little fundamental conceptual problems in animal scholarly information to be gleaned from cognition science could be seen in the such similarities: insects and humans are predominant approach of reducing complex too distantly related for such comparisons behavioural or psychological constructs to reveal anything but evolutionary into small sets of experimental test criteria. convergence. The terms that label these We will explore the consequences of this features are merely metaphors, and they top-down approach and compare it to an should not mislead us to construe that alternative, bottom-up approach, thereby insects ‘think‘ like humans. However, in touching on questions of terminological recent work on insect cognition, the ambiguities, the arbitrariness of definitions, boundaries between human and insect and problems of undetected circular reasoning. performances have often been blurred, sometimes in ways that are not just The top-down approach in innocent metaphors. There is no question animal cognition that some insects perform impressive cognitive feats; recent discoveries include Current work on animal cognition, attention-like processes (Spaethe et al., where it is dealing with psychological 2006; Van Swinderen and Andretic, 2011), constructs, typically follows a type of categorization of visual stimuli (Zhang et hypothesis-driven approach that is ‘top- al., 2004), rule learning in delayed matching down’ in more than one way. First, concepts to sample tasks (Avarguès-Weber et al., and terms are mainly anchored in human 2011), context learning (Collett et al., psychology, and then superimposed on 1997), sequence learning (Chittka et al., animals. Second, the hypotheses are often 1995), various types of social learning not strictly deductive, as good hypothesis- (Leadbeater and Chittka, 2007), interval testing might be; instead the hypothesis is timing (Boisvert and Sherry, 2006), often that animals exhibit behaviour that associative recall (Reinhard et al., 2004) is human-like, and researchers are content and sensitivity to number (Chittka and with confirming this hypothesis, and indeed Geiger, 1995). However, other recent claims with adjusting both procedures and definitions on the occurance of ‘teaching’, ‘culture’, to see it confirmed. ‘consciousness’, or ‘personality’ in the insects The top-down approach in animal seem to apply a relatively restrictive top- cognition relies on taking interesting down approach, where researchers set out psychological constructs that are rooted in to discover human-like behaviours in animals. the human domain, and translating these This approach is prone to terminological into experimental paradigms for animals. ambiguities, because terms taken from the The constructs have acquired their names domain of human experience often invoke in the human world, not in the zoological more complex connotations than the restricted sciences, and an almost inevitable consequence criteria used to test the phenomena in the of this is terminological confusion, i.e. the animal world would permit. chosen terms are ambiguous. Certainly, Bordering on psychology, philosophy, this hypothesis-driven way of conducting physiology and ecology, the science of animal research on animal cognition is very cognition attracts interest from many sides. powerful and has fuelled massive progress Since all bring in their terminological in the understanding of animal behaviour traditions, cross-discipline communication and animal cognition over the last few can be difficult. However, this is not the decades (Bekoff et al., 2002). Nothing can only, and possibly not the most serious beat the excitement accompanying the source of conceptual confusion. In this successful completion of a rigorously paper we argue that one of the most designed experiment that proves the

86 台灣昆蟲第三十一卷第二期 existence of a cognitive ability in an predisposition to think, feel, and act in a animal species for the first time. Pointing particular way. To still others, personality out problems in this approach, therefore, consists of a person’s unique mixture of may carry the risk being regarded as a emotional, intellectual, and character traits spoilsport activity (Shettleworth, 2010). (honesty, courage, etc.). To more behaviorally However, we believe that it is worth oriented psychologists, personality is not taking this risk for the potential reward of something internal, but rather an externally new insights into the world of animal (and observable pattern of organized behavior specifically insect) cognition. typical of a person. Perhaps an acceptable compromise is to define human personality Terminology: the problem of ambiguity as a composite of cognitive abilities, interests, and content erosion attitudes, temperaments, and other individual Some terminological inaccuracies are differences in thoughts, feelings, and behavior. innocent. When the words ‘queen’, or This definition emphasizes the fact that ‘honeybee dance’, or indeed ‘dance language’ personality is a unique combination of are used in studies of social insects it is cognitive and affective characteristics clear that the double meaning of the words describable in terms of a typical, fairly will not endanger the conceptual clarity of consistent pattern of individual behavior. entomological studies. The semantic distance From the last definition, it follows that of the word ‘queen’ between the worlds of methods of assessing personality should insect and human societies is too large to include a broad spectrum of cognitive and lead to any misunderstandings. Thus, the affective variables.” (Aiken, 2000) metaphorical use of these anthropo- In humans, personality is now often morphisms is safe. Just using the word quantified in terms of the ‘big five’ ‘queen’ in the title of a bee study does not dimensions openness, conscientiousness, offer any increased attention from readers. extraversion, agreeableness, and neuroticism In contrast, consider the question (Goldberg, 1993), where each of these is whether the individuals of certain animal explored using a multitude of questionnaires species can be said to have ‘personality’. or tests. Contrast this complex picture of This term, without inverted commas, is personality, emphasizing the plurality of now commonly used among vertebrate approaches, with the approach to measure behavioural ecologists (Bell and Sih, 2007; the correlation between just two behavioural Frost et al., 2007), and this has recently variables in ‘personality’ studies in animals expanded into the insect literature (Wray (Bell and Sih, 2007), or indeed just a basic et al., 2011), with the curious implication level of consistency in a single behavioural that insects have person-like qualities. trait (see e.g., references in Muller et al., Although the term may be defined concisely 2010). To add to the confusion, some in an animal study, its grandeur, which authors use the terms “personality” and has been acquired in psychology, may spill “behavioural syndrome” interchangeably, over into the world of insects – at the whereas others make a distinction (Dall et considerable risk that meanings are al., 2004; Sih et al., 2004; Quinn and confounded. How is ‘personality’ defined in Cresswell, 2005). a psychology textbook? “The term personality Terms like ‘personality’ have a very has many different meanings. To some, it broad meaning in colloquial language, and refers to a mysterious charisma possessed many people might have a different by Hollywood actors and other popular, understanding of the term. The technical influential people, but not by everyone. To meanings of the same words, developed others, personality is the same as over long periods of time in various temperament, a natural, genetically based disciplines like philosophy, psychology and

How Human Are Insects, and Does it Matter? 87 sociology, may just be as broad and diverse important scientific issues. There is no as in everyday language, but they are question that the study of interindividual documented and are certainly (if only behavioural variation is important from an subtly) different from the colloquial evolutionary point of view (e.g., Thomson meanings (e.g., Aiken, 2000, p. 262). and Chittka, 2001; Raine et al., 2006; Raine When using a term that has both and Chittka, 2008) as well as for technical and colloquial meanings, it is understanding the organisation of work in quite inevitable to invoke the broad social insect colonies (Gordon, 1996; meanings associated with the non-technical Chittka and Muller, 2009) – but this is term, at least in cursory or non-specialist quite independent of whether the term readers, and especially when it is placed in ‘personality’ applies to any of such a prominent place like the title or abstract. variation. Insect researchers have been Although this certainly increases the aware of consistent behavioural differences potential impact of the study by capturing between individuals for decades (see the attention of a wider audience, many Thomson and Chittka, 2001, and references readers might feel slightly disappointed as therein; for an early example of between- soon as they arrive at the rather narrow individual differences in insects’ ‘intellect’, and restricted technical definition. In this see Christy, 1884). There is no scholarly case, it is obvious that concepts are information to be gained by giving a well narrowed down for the operational purposes established phenomenon a new label. This of the study, while the impression of a very also applies to recent claims of teaching complex concept is maintained. and culture in ants (Franks and Richardson, We may indeed (by definition) call a 2006; Reznikova and Panteleeva, 2008) or behaviour that has been observed in an consciousness in bees (Griffin and Speck, animal species ‘personality’, but this is 2004). There are many examples where only a short form of saying that a defined definitions of cognitively loaded terminology set of criteria has been met. This does not have been broadened in the process of necessarily mean that the psychological removing them from their (human-related) processes in humans and animals that are origins and applying them to animals described with the same term are across taxa (Leadbeater et al., 2006; Penn equivalent or even comparable. However, et al., 2008). Such inflationary usage of by calling the observed behaviours in terminology is also unhelpful for the goals animals by the same terms used to describe of comparative cognition i.e., to decipher human behaviour we risk invoking very the evolutionary paths to traits related to colourful but – in some cases – relatively learning and memory. Giving two phenomena unsubstantiated associations in the the same label does not turn them into the interpretation of animal behaviour. same biological trait (Chittka and Jensen, Therefore, as the attention that animal 2011). cognition studies are able to attract appears The remedies against terminological to depend at least partly on the unintentional ambiguity are obvious. Possible misunder- or deliberate use of terminological ambiguity, standings caused by double or multiple perhaps the most acute danger to conceptual meanings in colloquial and technical language clarity in animal cognition may lie in the must be anticipated and definitions pressure to transform scientific results into clarified by appropriate discriminants. sensational stories. As an example for the Anthropomorphisms should be avoided need to glamorise science to attract when it is not clear that they are purely continued funding, see Pielke and Glantz metaphorical. Where multiple technical (1995). Indeed, in some case such definitions are available it should be glamorization may detract from more justified and discussed why a particular

88 台灣昆蟲第三十一卷第二期 definition was chosen. Problems with Circularity, Experimenter Biases and ‘Clever Hans’ Effects Problems with using humans as a Any decision whether or not to include reference point criteria for psychological constructs that is Early researchers working on learning based on expected outcomes in animals and memory, e.g., Pavlov and Skinner denies us the possibility of a meaningful used various animal models to discover test of the construct in these animals. A general principles, and (perhaps sometimes revealing example is the discussion on too liberally) applied them to other animals, definitions of animal intelligence in a including humans. However, more recent textbook on animal cognition (Pearce, work on animal cognition almost universally 2008). In this source (p. 13-14), the speed uses human psychology as a point of of learning is discarded as a suitable reference. This might be in part because measure for intelligence, because bees many of the processes investigated (e.g., perform better than rabbits or indeed consciousness, episodic memories, planning, human children on a colour learning task. theory of mind etc.) are, first and foremost, Following this, any meaningful decision on accessible by introspection. In applying whether bees or rabbits are more intelligent these concepts to animals, we typically ask has become logically impossible. There how they might manifest themselves in may be good reasons to be uncomfortable observable behaviour, establish a set of with equating learning speed with criteria that might be underpinned by intelligence, but the observation that similar processes as those accessible to us humans do not top the chart should not be by introspection, and see whether they are one of them. Such circular reasoning is met by our study animal. As a consequence probably often hidden, because how authors of the necessarily human anchor in this arrive at their chosen definition (set of form of the top down approach, it has an criteria) for psychological constructs is not inbuilt tendency towards anthropomorphism. a feature described in the methods of The problem with this form of the animal cognition studies. hypothesis-driven approach is that one Recent controversies surrounding the can never discover anything genuinely possibility of misconduct in various studies novel, even though journal editors and the related to animal behaviour and cognition press love demonstrations that animals (Abbott, 2004; Ledford, 2010) quite possibly are clever and ‘human-like’. The best miss the point: incidences of academics outcome one could hope for is that the deliberately fabricating data are probably observed behaviour meets some criteria quite rare. However, experimenter biases, that have been derived from research on where observers inadvertently read their human cognition. preconceived notions into what they see, A more meaningful way of the top- are an integral part of human psychology down approach is to explore hypotheses (Rosenthal and Lawson, 1964). This might grounded in the animals’ natural environment. be an uncomfortable notion for many, and For example, von Frisch’s exploration of it might appear attractive to find scapegoats colour vision in bees, was based on the and accuse them of all-out fraud, rather observation that flowers are colourful; than addressing subconscious biases that therefore, he argued, bees should be able quite possibly affect the majority of us. to perceive the colour signals that are Researchers need to immunise themselves clearly addressed to them (Frisch, 1914). against such biases, ideally by fully automated recordings of animal behaviour. Curiously however, methods such as those used in our laboratory, using electronic

How Human Are Insects, and Does it Matter? 89 flowers with infrared light barriers double blind studies. (Chittka, 1998), 3D video recordings of bee The conclusion is that a one-dimensional flight trajectories (Ings and Chittka, 2008) search for human-ness in the animal kingdom, or radio frequency identification (RFID) when paired with unchecked experimenter (Stelzer and Chittka, 2010) often meet biases and more or less explicit encourage- with resistance in certain quarters of the ment from journal editors and the press, animal behaviour community who perceive will generate somewhat predictable outcomes, such methods as unnecessarily technological, but perhaps without making significant or too far removed from natural settings. contributions to the advancement of scientific They aren’t – automated procedures should knowledge. be standard in all recordings of animal behaviour. At the very least, observers Bottom up-approaches to animal entrusted with the test should be un-informed behaviour and cognition about what the experiment coordinator expects, as in double blind studies now We are so frequently told that scientific standard in e.g. drug testing. Where these discovery must be based on hypothesis standards are not observed, researchers, testing that it’s easy to forget that many editors and readers should be aware of the major scientific breakthroughs were entirely preliminary nature of the work. data driven (see e.g., Langley et al., 1987 An added complication of the experi- for examples). These authors caution that menter knowing what he or she expects the “very strength of the methods available from his animal subject is the Clever Hans in theory-driven discovery sometimes produce effect, named after a horse with a seemingly the danger that theories will be created prodigious gift for numeracy in the early that rationalize rather than explain the 1900s. However, a careful investigation empirical data”. As we have seen, there revealed that Hans, instead of being able are other dangers, i.e., to subconsciously to count, paid attention to subtle cues adjust terminology, procedures or interpret- from his trainer to respond in the correct tations so that they support a particular manner (Pearce, 2008, p. 243). Such effects idea. This risk is very low if large quantities are probably especially common with of data are collected first, and rules or domestic or experimental animals that patterns extracted by automated procedures. interact with humans on a daily biases, New technologies and computational ‘high- but might also occur in insect studies. A throughput’ approaches to behaviour analysis, possible example is a study by Gould as well as new neurobiological tools make (1984), where bees apparently learnt to this increasingly feasible in studies of anticipate the next location of a feeder animal behaviour. that was being moved in regular steps. A conceivable interpretation is that bees Observations of behaviour learned to use the appearance of the Many luminaries in insect behaviour experimeter himself as a landmark (for example Jean-Henri Fabre and Karl indicating the food, and if that experimenter von Frisch) have started with careful inspected a previously unrewarded location, observations of their animals in their or indeed moved the feeder to a new natural environments. Fabre made discovery location, the bees would simply follow him after discovery about the remarkable and search in the vicinity. Such complications diversity of ‘instinctual’ behaviour in his may be pervasive in studies where study organisms, precisely because he did informed experimenters interact directly not observe them with preconceived with animals, again emphasising the need notions of what he might find. (That said, for automated data recording and/or he was also a careful experimenter – his

90 台灣昆蟲第三十一卷第二期 explorations of homing in parasitoid wasps, and excitement about diversity in animal and the question of how they identify nest information processing, the promise of locations, were early masterpieces in alien perceptual worlds – how sad, in behavioural ecology.) comparison is the one-dimensional search A bottom-up ‘inductive’ approach also for human-like qualities in the animal laid the foundations for von Frisch’s work, kingdom! The take-home message is, quite and that of many of his disciples who simply, to keep your eyes open for the started an investigation by carefully unexpected. observing their study organisms in their natural settings, taking everything in, and Ethomics – automated behaviour getting to know the study organism classification thoroughly from many angles; this often Behaviour is essentially a coordinated resulted in the discovery of unexpected or series of precisely timed muscle contractions inexplicable phenomena. Only then were from simple ones such as leg extension to testable hypotheses developed and rigorous extended behavioural sequences, such as experiments set up to zero in on how the waggle dance of honeybees. However, a particular processes might be explained. complication is that the classification of One might argue that this approach is not behaviour typically follows entirely subjective entirely hypothesis-free either – that it is criteria and is therefore difficult to only possibly to discover the unexpected if compare across animals: for example ‘leg there are certain expectations. But the extension’, ‘bee waggle dance’ and ‘consensus point here is that the bottom-up approach building in bee swarms’ are not equivalent to animal behaviour is open-ended: it does behavioural classifications, and are to not start with a predefined goal that it some extent nested within one another. It seeks to reach, or an idea it wants to is therefore desirable to classify behaviour confirm. by automated procedures so they can be For example, von Frisch would certainly objectively measured and compared using never have discovered the bee ‘dance what Branson et al. (2009) have called an language’ if he had deliberately set out to ‘ethomics’ approach. find a form of symbolic communication in New technologies in motion capture the animal kingdom (following a top down- and video analyses make an entirely approach). Or would Lubbock have discovered hypotheses free approach to the analysis UV sensitivity in ants if he had looked for of behaviour feasible. The complete a colour vision system exactly like humans’? classification of behavioural repertoires by Probably not – and indeed, these would automated procedures might soon be have been rather misguided endeavours. J. within reach; see for example Branson et Gould wrote about von Frisch: “His al. (2009), Braun et al. (2010), Benjamini pioneering work inspired the discovery of et al. (2011), for promising developments. several otherwise unimaginable sensory For the future, it will be necessary to systems in animals: infrared detectors in develop large-scale automated analytical night-hunting snakes, ultrasonic sonar in tools to effectively mine the data collected dolphins and bats, infrasonic hearing in with the aim of finding prototypical and birds, and magnetic field sensitivity in a atypical behaviours. To discover prototypical variety of animals. Doubtless, other systems activity models a useful approach is to are still to be discovered. The lesson is a implement procedures for clustering melancholy one: We are blind to our own accumulated trajectory data into repre- blindness, and must try not to read our sentative patterns (Braun et al., 2010). own disabilities into the rest of the animal We will ultimately have to break down kingdom.” (Gould, 1980). Note the awe behavioural sequences into their smallest

How Human Are Insects, and Does it Matter? 91 identifiable units, and quantify them in these might consist of small-scale hardwired terms of trajectories of body parts, their motor routines strung together in flexible speed, distance, acceleration, deceleration ways (Gould and Marler, 1984). Again, we etc., thereby identifying behavioural would only pick this up by breaking down prototypes (Braun et al., 2010) that can behaviour into small components, using subsequently be strung together into methods as described above. various sequences to reconstruct meaningful natural behaviour. To encode the variability Bottom-up approaches to cognition of the behavioural repertoires, it will be from neuroscience necessary to develop a dedicated alphabet While the fully automated, bias-free and syntax of movements that can be analysis of behaviour is now becoming effectively compared between individuals increasingly feasible, this is less straight- and species. An automated ‘data mining’ forward with cognitive processes that might approach will also facilitate the discovery or might not have observable outcomes. of new behaviour patterns that have so far There is no question that hypothesis driven escaped the attention of human observers. behavioural experiments are a necessary In this bottom-up approach, we are, at ingredient of studying animal cognition the early stages, less focussed on one (although the hypotheses might more particular concept, but we are also much suitably be driven by the motivation for freer to observe the peculiarities of the scientific enquiry rather than for confirming animals’ behaviours. Insect behaviour, in that animals are clever). Nonetheless all its diversity, is unlikely to be wholly there is plenty of room for more bias-free describable by the concepts that are approaches in the study of cognitive derived from just one, very unusual approaches as well. We need to understand species, Homo sapiens. By looking at many the neural circuitry that underpins cognitive more animal species, under more test processes in more detail, not just because conditions, and by testing many, and we really still do not understand how the smaller, i.e. more basic criteria (cognitive brain works, but also to understand the or behavioural units), we probably could evolution of cognitive capacity. ‘Intelligence’ build psychological constructs that are is not a biological trait that can be mapped more independent from the human world onto an evolutionary tree in any meaningful and therefore, ultimately, more meaningful. way. We need to know how many neurons It is clear that such an approach (and with which connections) are engaged cannot stand on its own, and needs to be in any defined cognitive feat and how many coupled with experimental manipulation sequential stages of information processing which in turn needs to be hypothesis there are. Insects’ small nervous systems driven. Learnt behaviours can only be should make it feasible to explore these tested in meaningful ways by exposing questions at a very fine-grained level. animals to controlled stimuli, and these The quest for understanding the precise cannot be generated in a hypothesis free neuron-to-neuron connectivity underlying manner. Additionally, learning may cognitive processes can be addressed both generate a practically infinite number of from an empirical angle, as well as from a behavioural routines, which might be modelling perspective. Multi-electrode difficult to pick up in clustering algorithms – recordings provided profound insights into consider, for example, that the diversity of how the brain stores and organises flower handling procedures required of a memories in the rodent hippocampus in generalist pollinator to extract nectar or the 1990s (Wilson and McNaughton, 1993; pollen is presumably as great as the McNaughton et al., 1996) and the number number of floral morphologies. Nonetheless of cells recorded from simultaneously (>

92 台灣昆蟲第三十一卷第二期 100 in some cases) is impressive, and the number of available neurons (and contributed substantially to the under- their connections)? Circuit functionality standing of the neural ensemble code of can be assessed by varying all parameters the mammalian hippocampus. Although in discrete steps, and keeping a complete the settings in which the data are recorded record of simulation results, including the are of course hypothesis driven, their output for each varied parameter value subsequent analysis is largely data-driven, under all combinations of parameter in that information contained about settings, as has been done in simulations individual neuronal firing patterns, and to derive optimal colour coding systems their propagation and intergration needs (e.g., Chittka, 1996) where thousands of to be extracted from the raw data by data systems were analysed. In conclusion, mining processes. The adaptation of such because the bottom-up approach reduces the techniques for insects could generate a importance of introspection for the much more comprehensive picture of “verification” of psychological constructs, it ensemble neural code of any one brain could contribute to create a more objective, area in insects than in mammals. A recent more independent, and less arbitrary neuroanatomical study, for example, methodology. succeeded in identifying and mapping 16% of the ~100,000 neurons of the Drosophila Bottom-up approaches in terminology brain (Chiang et al., 2010), illustrating At present, the plurality of definitions just how far we have progressed in of psychological constructs seems to make understanding the full circuitry of insects’ positive results relatively easy to obtain brains. Technology for multielectrode when looking for human-like behavioural recording has already been adapted for complexes in animals. However, as use in insects (Du et al., 2009; Bender et demonstrated above, this situation creates al., 2010) but has yet to be tested on both confusion and room for exaggerated problems related to cognition. Other recent conclusions. As a first measure for more developments make the imaging of entire clarity we therefore suggest to increase neural circuits, and possibly whole brains the diversity of terms (each with a precise at micron resolution, increasingly feasible meaning), rather than a multitude of (Dodt et al., 2007; Jahrling et al., 2010). meanings attached to each of a few grand It is quite conceivable that, as such words. If findings do not fit under the technologies develop, we will discover umbrella of a particular and established neural-computational solutions to cognitive cognitive construct they may still be close tasks in insects (and other animals) quite by and deserve a novel term. This is unlike those in primates (even where the certainly better than repeatedly (and more behavioural outcomes might be similar). or less forcefully) adjusting the size of the Neural network modelling can provide umbrella. useful pointers for neuroscientists to Consider, for example, Franks and concentrate their efforts on what to Richardson's (2006) claim to have found explore, and here, too, a more open-ended an instance of ‘teaching’ in ants, approach is needed. Often, modellers search explaining (correctly) that the behaviour for a single computational solution that they observed met or indeed exceeded the best explains an empirically determined criteria for teaching as defined for other phenomenon. An alternative approach animals. However, the behaviour had would be an emphasis on diversity – for already been described 30 years earlier example, for a given cognitive problem, (Hölldobler et al., 1974; Möglich et al., how many neural solutions might be 1974), and its function had been (at least generated, and how does this depend on in part) revealed. Indeed the behaviour

How Human Are Insects, and Does it Matter? 93 had been given a name, ‘tandem running’, to phylogenetic analysis. Words from the that emerged directly from what the human cognitive world are unlikely to be authors observed, and which appropriately sufficient for analysing all animal cognition described the activity without making and behaviour, and sticking to the human inferences about its function. The main world terms will most likely impede advance by Franks and Richardson (2006) progress, because it will just not leave is in discovering that the behaviour fulfils enough freedom to capture novel processes. the criteria for teaching as defined by Caro The recommendation would be to let terms and Hauser (1992), although it is evident emerge from patterns observed in the data that these criteria have been adjusted to themselves. In short: first the data, then allow the term teaching to be expanded to the words. animals and fall far short of how one would define teaching in the human realm Consideration for Reporting (Leadbeater et al., 2006). The scientists Hypothesis Building and Validation who originally discovered the behaviour In the task to avoid over-restriction or labelled it exactly as we advocate here: narrowness of psychological constructs they gave it a ‘bottom-up’ term, whereas (such as ‘personality’) that seem to limit (Caro and Hauser, 1992) started in a the potential achievements of the current typical top-down fashion – adapt the top-down approach, much can be learnt definition of an established behaviour in from the psychological sciences. The first humans so that it includes animals, then thing to acknowledge is that the con- find animals that match the definition. struction of the concept – the definition, as There are more risks here than just it were – is part of the methods, rather those related to sensationalism of than an issue that can be safely put away inappropriate terminology. In some cases, into the introduction as something given. permissive terms are misconstrued as The bottom-up approach in animal cognition being biological traits that one might map would need more effort invested in the onto phylogeny. Following on from the development of such methods. But also (clearly rather inclusive) definition of within the top-down approach, a critical teaching by Caro and Hauser, (1992), consideration of its methods could be Hoppitt et al., (2008) set out to investigate possible (Hare, 2001; Sarter, 2004). The the taxonomic distribution of the essential methodological tool that would phenomenon. They write: “Teaching was be useful in both domains is a validity regarded as a uniquely human faculty. test. However, recent studies suggest that teaching might be more common in “The construct validity of a psychological animals than previously thought.” – well, assessment instrument refers to the extent yes, but that is a result of broadening the to which the instrument measures a particular meaning of the term so as to include many construct, or psychological concept, such as forms of animal signalling. The authors anxiety ...or intelligence. [...] it involves a then note the “bizarre taxonomic distribution” network of investigations [...] designed to of teaching, but the problem is that a determine whether an assessment instrument colloquial term like teaching (like that purportedly measures a certain ‘intelligence’, or ‘speed’) is, quite simply, psychological variable is actually doing not a biological character that can be so.” (Aiken, 2000, p. 95) studied in evolutionary terms. Conversely, bottom-up terms, such as ‘tandem running’ Investigating the validity of a test for or ‘bee dance’ that are designed to describe a cognitive construct is not an easy task, specific behaviour patterns, are accessible especially when dealing with animals,

94 台灣昆蟲第三十一卷第二期 because one of the most targeted tools for allow too many or too strong research determining validity is barred from zoology, biases, or terminological inaccuracies. the in-detail and direct questioning of These biases rest on the current scientific examinees “about their responses to a test reward system that strongly favours in order to reveal the specific mental studies with a sensational slant that can processes involved in responding to the bring animal behaviour closer to human items...” (Aiken, 2000, p. 95). However, experience. Such a system invites usage of there are other approaches to test the grand words with narrow definitions, to validity of an instrument that claims to concentrate on the smart animals, and to reveal a cognitive construct. These include report only positive results. The proposed the (documented) judgement of experts approach for animal cognition and behaviour about the relationship between the science would solve the problems of the contents of the test instrument and the positive results bias and the smart-animal construct, as well as experimental studies bias, because it does not create failures in that link the construct test instrument to the results, as all possible results are external variables. One particularly equally positive. Clearly, an approach that interesting tool in the validity test tool box is only bottom-up would not lead us very is discriminant validation, which compares far, but a careful combination of both how the same instruments (methods) approaches could be very powerful indeed. measure different psychological constructs. As Francis Bacon points out in the Novum It can be expected that tests of psycho- Organum (1620), the bee combines the logical constructs in animal cognition best of both worlds: “Empiricists, like ants, science would greatly benefit even from merely collect things and use them. The “quick and dirty” validity tests and that Rationalists, like spiders, spin webs out of such tests might form an important tool to themselves. The middle way is that of the detect over-restriction in psychological bee, which gathers its materials from the constructs. flowers ... but then transforms and digests it by a power of its own.” Conclusion References Currently, many studies seem to ask to what degree animals (including insects) Abbott A. 2004. Prolific ecologist vows to have cognitive properties that can be fight Danish misconduct verdict. likened to human cognition. We suggest Nature 427: 381. that the question, at present, is as difficult Aiken LR. 2000. Psychological testing and to answer as it is unimportant, at least in assessment. 10 ed. Boston: Allyn and the insects. Looking for cognitive homologies Bacon. between humans and their closest relatives Avarguès-Weber A, Deisig N, Giurfa M. is useful to understand the evolution of 2011. Visual cognition in social insects. cognitive traits, although unless we know Ann Rev Entomol 56: 423-443. the underlying circuitry, and have good Bekoff M, Allen C, Burghardt GM. 2002. estimates about how quickly cognitive The cognitive animal: empirical and traits might evolve, the possibility of theoretical perspectives on animal convergence needs to be considered even cognition. Cambridge, Massachusetts: in such comparisons between close relatives. MIT Press. More broadly, we will not be able to Bell AM, Sih A. 2007. Exposure to predation answer the question of how cognitive generates personality in threespined abilities are distributed in the animal sticklebacks (Gasterosteus aculeatus). kingdom, and how they evolved, if we Ecol Lett 10: 828-834.

How Human Are Insects, and Does it Matter? 95 Bender JA, Pollack AJ, Ritzmann RE. specialization, efficiency and task 2010. Neural activity in the central allocation in social insects. Commun complex of the insect brain is linked to Integr Biol 2: 151-154. locomotor changes. Curr Biol 20: 921- Chittka L, Jensen K. 2011. Animal 926. cognition: concepts from apes to bees. Benjamini Y, Fonio E, Galili T, Havkin GZ, Curr Biol 21: R116-R119. Golani I. 2011. Quantifying the Chittka L, Kunze J, Geiger K. 1995. The buildup in extent and complexity of influences of landmarks on distance free exploration in mice. Proceedings estimation of honeybees. Anim Behav of the National Academy of Sciences. 50: 23-31. doi: 10.1073/pnas.1014837108. Christy RM. 1884. On the methodic habits Boisvert MJ, Sherry DF. 2006. Interval of insects when visiting flowers. J timing by an invertebrate, the bumble Linn Soc 17: 186-194. bee Bombus impatiens. Curr Biol 16: Collett TS, Fauria K, Dale K, Baron J. 636-1640. 1997. Places and patterns - a study of Branson K, Robie AA, Bender J, Perona P, context learning in honeybees. J Dickinson MH. 2009. High-throughput Comp Physiol A 181: 343-353. ethomics in large groups of Drosophila. Dall SRX, Houston AI, McNamara JM. Nature Methods 6: 451-U77. 2004. The behavioural ecology of Braun E, Geurten B, Egelhaaf M. 2010. personality: consistent individual Identifying prototypical components differences from an adaptive per- in behaviour using clustering algo- spective. Ecol Lett 7:734-739. rithms. PLoS One 5. doi: e936110. Dodt HU, Leischner U, Schierloh A, 1371/journal.pone.0009361. Jahrling N, Mauch CP, Deininger K, Caro TM, Hauser MD. 1992. Is there Deussing JM, Eder M, Zieglgansberger teaching in nonhuman animals. Q Rev W, Becker K. 2007. Ultramicroscopy: Biol 67: 151-174. three-dimensional visualization of Chiang AS, Lin CY, Chuang CC, Chang neuronal networks in the whole HM, Hsieh CH, Yeh CW, Shih CT, Wu mouse brain. Nature Methods 4: 331- JJ, Wang GT, Chen YC, Wu CC, Chen 336. doi: 10.1038/nmeth1036. GY, Ching YT, Lee PC, Lin CY, Lin HH, Du J, Riedel-Kruse IH, Nawroth JC, Wu CC, Hsu HW, Huang YA, Chen JY, Roukes ML, Laurent G, Masmanidis Chiang HJ, Lu CF, Ni RF, Yeh CY, SC. 2009. High-resolution three- Hwang JK. 2010. Three-dimensional dimensional extracellular recording of reconstruction of brain-wide wiring neuronal activity with microfabricated networks in Drosophila at single-cell electrode arrays. J Neurophysiol 101: resolution. Curr Biol 21: 1-11. 1671-1678. Chittka L. 1996. Optimal sets of colour Franks NR, Richardson T. 2006. Teaching receptors and opponent processes for in tandem-running ants. Nature 439: coding of natural objects in insect 153. vision. J Theor Biol 181: 179-196. Frisch Kv. 1914. Demonstration von Versuchen Chittka L. 1998. Sensorimotor learning in zum Nachweis des Farbensinnes bei bumblebees: long term retention and angeblich total farbenblinden Tieren. reversal training. J Exp Biol 201: Verhandl Deutsch Zool Gesellsch 515-524. Freiburg 34: 50-58. Chittka L, Geiger K. 1995. Can honeybees Frisch Kv. 1967. The dance language and count landmarks? Anim Behav 49: orientation of bees. Cambridge: Harvard 159-164. Univ. Press. Chittka L, Muller H. 2009. Learning, Frost AJ, Winrow-Giffen A, Ashley PJ,

96 台灣昆蟲第三十一卷第二期 Sneddon LU. 2007. Plasticity in segmentation of intact Drosophila by animal personality traits: does prior ultramicroscopy. Front Syst Neurosci experience alter the degree of 4: 1: doi: 10.3389/neuro.06.001.2010 boldness? Proceedings of the Royal Langley P, Simon HE, Bradshaw GD, Society B-Biol Sci 274: 333-339. Zytkov JM. 1987. Scientific discovery - Goldberg LR. 1993. The structure of computational explorations of the phenotypic personality traits. Am scientific processes. Cambridge, Psychol 48: 26-34. Massachusetts: MIT Press. Gordon DM. 1996. The organization of Leadbeater E, Chittka L. 2007. Social work in social insect colonies. Nature learning in insects - from miniature 380: 121-124. brains to consensus building. Curr Gould JL. 1980. Navigation by honeybees. Biol 17: R703-R713. pp 135-141. In: Horovitz N, Hutchings Leadbeater E, Raine NE, Chittka L. 2006. E (eds). Genes, Cells, and Behavior. Social learning: ants and the meaning W.H. Freeman, San Francisco. of teaching. Curr Biol 16: R323-R325. Gould JL. 1984. Natural history of honey Ledford H. 2010. Harvard probe kept bee learning. pp 149-180. In: Marler P, under wraps. Nature 466: 908-909. Terrace HS (eds), The Biology of Lindauer M. 1955. Schwarmbienen auf Learning. Springer, Berlin. Wohnungssuche. Z Vergl Physiol 37: Gould JL, Marler P. (1984) Ethology and 263-324. the natural history of learning. pp McNaughton BL, Barnes CA, Gerrard JL, 47-74. In: Marler P, Terrace HS (eds). Gothard K, Jung MW, Knierim JJ, The Biology of Learning. Springer, Kudrimoti H, Qin Y, Skaggs WE, Berlin. Suster M, Weaver KL. 1996. Griffin DR, Speck GB. 2004. New evidence Deciphering the hippocampal polyglot: of animal consciousness. Anim Cogn 7: the hippocampus as a path 5-18. integration system. J Exp Biol 199: Hare B. 2001. Can competitive paradigms 173-185. increase the validity of experiments Möglich M, Maschwitz U, Hölldobler B. on primate social cognition? Anim 1974. Tandem calling-new kind of Cogn 4: 269-280. signal in ant communication. Science Hölldobler B, Möglich M, Maschwitz U. 186: 1046-1047. 1974. Communication by tandem Muller H, Grossmann H, Chittka L. 2010. running in ant Camponotus sericeus. J 'Personality' in bumblebees: individual Comp Physiol 90: 105-127. consistency in responses to novel Hölldobler B, Wilson EO. 1990. The ants. colours? Anim Behav 80: 1065-1074. Berlin: Springer. Pearce JM. 2008. Animal learning and Hoppitt WJE, Brown GR, Kendal R, cognition. 3rd ed. Hove & New York: Rendell L, Thornton A, Webster MM, Psychology Press. Laland KN. 2008. Lessons from Penn DC, Holyoak KJ, Povinelli DJ. 2008. animal teaching. Trends Ecol Evol 23: Darwin's mistake: explaining the 486-493. discontinuity between human and Ings TC, Chittka L. 2008. Speed-accuracy nonhuman minds. Behav Brain Sci 31: tradeoffs and false alarms in bee 109-178. responses to cryptic predators. Curr Pielke RA, Glantz MH. 1995. Serving science Biol 18: 1520-1524. and society: lessons from large-scale Jahrling N, Becker K, Schonbauer C, atmospheric science programs. B Am Schnorrer F, Dodt HU. 2010. Three- Meteorol Soc 76: 2445-2458. dimensional reconstruction and Quinn JL, Cresswell W. 2005. Personality,

How Human Are Insects, and Does it Matter? 97 anti-predation behaviour and behavioural Spaethe J, Tautz J, Chittka L. 2006. Do plasticity in the chaffinch Fringilla honeybees detect colour targets using coelebs. Behaviour 142: 1377-1402. serial or parallel visual search? J Exp Raine NE, Chittka L. 2008. The correlation Biol 209: 987-993. of learning speed and natural foraging Stelzer RJ, Chittka L. 2010. Bumblebee success in bumble-bees. Proceedings foraging rhythms under the midnight of the Royal Society of London Series sun measured with radiofrequency B-Biol Sci 275: 803-808. identification. Bmc Biology 8. doi: 93 Raine NE, Ings TC, Dornhaus A, Saleh N, 10.1186/1741-7007-8-93. Chittka L. 2006. Adaptation, genetic Thomson JD, Chittka L. 2001. Pollinator drift, pleiotropy, and history in the individuality: when does it matter? pp evolution of bee foraging behavior. 191-213. In: Chittka L, Thomson JD Adv Stu Behav 36: 305-354. (eds). Cognitive Ecology of Pollination. Reinhard J., Srinivasan M.V., Zhang S. University Press, Cambridge. 2004. Scent-triggered navigation in Van Swinderen B, Andretic R. 2011. honeybees. Nature 427: 411. Dopamine in Drosophila: setting Reznikova Z., Panteleeva S. 2008. An arousal thresholds in a miniature ant's eye view of culture: propagation brain. Proc R Soc B. 278: 906-913. of new traditions through triggering Wilson MA, McNaughton BL. 1993. dormant behavioural patterns. Acta Dynamics of the hippocampal ensemble Ethologica. 2: 73-80. code for space. Science 261: 1055- Rosenthal R, Lawson R. 1964. A 1057. longitudinal study of the effects of Wray MK, Mattila HR, Seeley TD. 2011. experimenter bias on the operant Collective personalities in honeybee learning of laboratory rats. J Psychia colonies are linked to colony fitness. Res 2: 61-72. Anim Behav 81: 559-568. doi: 10.1016/ Sarter M. 2004. Animal cognition: defining j.anbehav.2010.11.027. the issues. Neurosci Biobehav Rev Zhang S, Srinivasan MV, Zhu H, Wong J. 28:645-650. 2004. Grouping of visual objects by Shettleworth SJ. 2010. Clever animals honeybees. J Exp Biol 207: 3289-3298. and killjoy explanations in comparative psychology. Trends Cogn Sci 14. Received: March 31, 2011 Sih A, Bell A, Johnson JC. 2004. Accepted: April 3, 2011 Behavioral syndromes: an ecological and evolutionary overview. Trends Ecol Evol 19: 372-378.

98 台灣昆蟲第三十一卷第二期一定要把昆蟲想成人類那樣嗎？

Thomas F. Döring1, and Lars Chittka2*

1 英國有機研究中心 Elm 農場 2 倫敦瑪麗皇后學院生物暨化學科學系生物與實驗心理學組

摘要

過去二十年來已有大量的資料在探討昆蟲的認知功能，即使昆蟲腦中的神經迴路是如此之小，也已有相當令人滿意的成果。然而近年的一些研究聲稱昆蟲存在「教導」、「文化」、「意識」或「性格」，科學家們開始著手探索動物的擬人行為，似乎涉及了相對限制性由上而下的方法。此方法容易導致術語的歧義性，由於這些術語出自人們的經驗，往往產生了比在動物世界中有限制條件試驗下可被允許的現象更複雜的含義。此結果也將承擔因主觀選擇或重新塑造了定義與測試標準所造成的後果之風險。最後，假說驅動的方法容易導致研究偏差被忽略，例如報告主要著重於正面的結果或以「聰明」的動物為主要研究對象。我們建議現行的方法需要加入互補性作法加以補充，以自動化程序進行行為的分類，透過由下而上的方法做更進一步的探索，在更廣範的動物行為多元數據分析後，也將揭露動物的心理構念，當然在研究方法中我們也必須注意我們是如何得到定義的。

關鍵詞：歧義、獨斷、行為、認知、方法、術語。

*論文聯繫人 Corresponding email: [email protected] How Human Are Insects, and Does it Matter? 99