Animals in captive or laboratory settings may outperform wild animals of the same species in both frequency and diversity of tool use, a phenomenon here termed ‘captivity bias’. Although speculative at this stage, a logical conclusion from this concept is that animals whose tool-use behaviour is observed solely under natural conditions may be judged cognitively or physically inferior than if they had also been tested or observed under controlled captive conditions. In turn, this situation creates a potential problem for studies of the behaviour of extinct members of the human family tree—the hominins—as hominin cognitive abilities are often judged on material evidence of tool-use behaviour left in the archaeological record. In this review, potential factors contributing to captivity bias in primates (including increased contact between individuals engaged in tool use, guidance or shaping of tool-use behaviour by other tool-users and increased free time and energy) are identified and assessed for their possible effects on the behaviour of the Late Pleistocene hominin Homo floresiensis. The captivity bias concept provides one way to uncouple hominin tool use from cognition, by considering hominins as subject to the same adaptive influences as other tool-using animals.
This review introduces the concept of ‘captivity bias’ as a means to discuss how differences in the diversity and frequency of animal tool use in captivity and in the wild may help us understand the behaviour and cognition of extinct hominins. This concept is supported by observations that animal tool use in captive or laboratory settings may outperform wild animals of the same species. These observations remain suggestive rather than definitive at this time, with the best available data coming from those animals that have most often been compared with tool-using hominins—the apes, macaques and capuchin monkeys. The principle underlying the captivity bias concept is that in cases where a species' tool-using ability has been argued to correlate with the cognitive ability of that species, it is important to consider the environmental conditions under which the relevant observations have been made. If tool-using ability varies by context, while underlying cognition does not, then the causal link between the two may not always be reliable. At the same time, if we can identify those aspects of the environment that promote more frequent or more diverse tool use, then we gain an additional analytical device for investigating differences in tool use between individuals and species. This study explores this line of reasoning, examining the factors that encourage or discourage technological expression over both evolutionary and ontogenetic timescales in known tool-using species. Versions of the captivity bias effect have been previously described as the ‘tool paradox’  or ‘captivity effect’ , but the potential effect of this bias has not previously been considered in reconstructions of the cognitive and behavioural abilities of hominins (those ancestral species closer to humans than Pan).
Based on a review of the existing animal tool-use evidence [3,4], this study contends that the ‘captivity bias’ effect is not a result of the general experience of being held captive, but derives instead from a series of atypical environmental and social circumstances generated within the captive environment. Specifically, these circumstances may include (i) increased contact between individuals engaged in tool use; (ii) guidance or shaping of tool-use behaviour by other tool-users; and (iii) increased free time and energy as a result of provisioning and lack of predation. If these (and other related) conditions are met outside of captivity, in free living populations, then I propose that it would be valuable to determine whether they result in a similar bias towards enhanced tool-use frequency or complexity.
(a) Cognition and tool use
A concise recent definition of cognition suggests that it involves ‘thinking and knowing’ . Here, the term is used in a broad sense to refer to information processing within the brain, including planning, memory, learning, decision-making and perception. Of special relevance here are those processes that have an observable behavioural outcome. Investigations of cognition through animal tool use have focused particularly on what is termed physical cognition , assessing how animals understand and act on causal relationships between physical objects. In addition to cause and effect, animal tool manufacture and the sequential use of tools have been seen as a guide to planning abilities, where the immediate aim of accessing a reward is postponed to first create or obtain the right tool for the job . It should be noted, however, that complex cognitive processes are not a prerequisite for animal tool use .
Many reconstructions of hominin cognitive abilities include a key characteristic that make them amenable to examination in terms of the captivity bias concept, namely that they are built on observations of tool use, usually stone tools as these survive most readily in the archaeological record [9–13]. For many living animal species, any effect of captivity bias would result in underestimation of cognitive capacities, if the sample were drawn purely from wild individuals, and we may therefore expect that: (i) if the conditions that promote captivity bias in non-human animals were not present in hominin environments, then we may systematically underestimate the cognitive capacities of tool-using hominins; and (ii) if some or all of the conditions that promote captivity bias are present for some or all extinct hominins, then their influence needs to be documented alongside other explanations for the patterning of hominin tool use. The existence of such a bias would not automatically determine that it would be expressed under all relevant circumstances for all tool-using animals, including hominins. However, it would offer an additional potential explanation for those circumstances where cognitive estimates of a species' ability show a mismatch with its observed technology. By treating hominin technology as a subset of more general animal tool use, we can increase our confidence that our conclusions are not being subconsciously guided by teleological or anthropomorphic reasoning.
2. ‘Captivity bias’ in animal tool use
(a) Animal tool use in captivity and the wild
For a comprehensive catalogue of animal tool use, including captive data, the reader is referred to recent reviews [3,4]. Here, I focus primarily on vertebrate animals, and especially primates, as these have received by far the most attention from researchers both in natural and captive conditions, and they form the majority of comparisons with the evolution of human behaviour.
Table 1 provides an overview of known tool-use modes (primarily motor actions), tool manufacture and associative tool use used in captive and wild conditions by a variety of primates, based on the rationale and data presented in reference . These categories were intended to allow a degree of standardization across different specific tool-use behaviours exhibited by specific animals, and they help to provide a firmer basis for cross-taxon comparisons by grouping together tool-use actions that share general manipulative or combinatorial features. For example, elsewhere in this themed issue, McGrew  compares chimpanzee and corvid tool use via these categories, although his approach differs slightly as he considers only habitual or customary tool use.
The available data show that captive tool use has greater diversity than in the wild among the apes, macaques and capuchin monkeys, but not in other primates. These data are likely skewed by research effort and observation effects, which is why captivity bias must remain a speculative suggestion at present rather than a firm principle, and similar data from other tool using taxa are required. Nevertheless, the species that are currently most often compared with hominins (including all the primates that use stone tools in the wild today)  have long histories of study both in the wild and captivity, with specific attention paid to their tool use [16,17–22]. Other primate species that are considered essentially non-tool-users in nature have also been shown to be capable of tool use under captive conditions, including Japanese macaques , baboons , cotton-top tamarins and vervet monkeys . Tool use is not the only behaviour that has been observed to be more elaborate in the captive primates; for example, the number of different social signals has been demonstrated to increase in captive baboons versus their wild counterparts . Note that the rearing experiences of captive animals can differ widely [26,27], and individual animals will show natural variation in their capabilities [28,29], with the result that any suggestion of captivity bias should therefore be seen as a generality rather than necessarily applying to every member of a given species. It may be also that some wild behaviours not involving tools, such as gorilla plant-food handling , exceed in complexity those seen in captivity, highlighting the fact that tool use does not exist in a cognitive or manipulative vacuum.
While primates have formed a significant focus of research into tool use, the same effect of more frequent and diverse tool use in captivity has been noted in animals as varied as elephants , keas , small tree finches , rooks , cockatoos , rodents such as degus  and naked mole rats  and even gastropods . The examples listed here are illustrative rather than exhaustive—of the 418 instances of animal tool use tabulated by Bentley-Condit & Smith , 120 or 28.7% were found in captivity alone. For most animals, however, we do not currently have sufficient data available to judge the existence or otherwise of a captivity bias in tool use, and important avenues for future research include evaluating the effect of possible observation bias on these data, and developing criteria for quantifying the extent of the captivity bias for each taxon. For the purposes of this paper, I will assume that the apparent captivity bias results from apes and other primates reflect reality, with the caveat that if future results prove otherwise the discussion presented here will require modification.
(b) Factors promoting captivity bias
Here, I focus on three broad topics that may promote or enable greater tool use under captive conditions: (i) increased contact between individuals engaged in tool use, leading to higher potential for the observation, maintenance and spread of tool-use innovations; (ii) guidance or shaping of individual or group tool-use behaviour by humans; and (iii) increased free time and energy as a result of provisioning and lack of predation. These topics are non-exclusive and non-exhaustive, and they are intended to provide an initially broad set of targets for identifying external influences on tool use.
In recent years, researchers have explored the relevance of the number of interacting individuals to the maintenance of traditions, showing that an increase in social contacts can allow for new ideas and techniques to be retained and spread more readily in a population [39–42]. The principle is not restricted to the human lineage , and wild orangutans and chimpanzees have been found to display increased complexity, success and frequency in tool use in increasingly social and higher population density settings [44–46]. Further, wild meerkats , captive cockatoos  and wild female dolphins also show that social effects can enhance tool use . These examples suggest that both increased population size and density, where these promote increased contact between individuals, can facilitate the retention and spread of technological innovations and subsequently increase diversity in a species' tool-use repertoire.
Simply having increased opportunities for tool use may not be sufficient in the absence of conspecifics . For social animals at least, therefore, it is plausible to suggest that a positive feedback loop of increased opportunities for repeated observation of tool-using individuals, and increased overall frequency and diversity of tool use, may be facilitated by captive conditions. In the case of tool use, the individuals with whom captive animals are coming into contact may in fact be tool-using humans, which is of particular importance for those species that are able to view humans as behavioural role models . Not all species may be able to view humans in this manner, although studies of fairness in non-human primates and object manipulation in crows demonstrate its plausibility as a mechanism [51,52]. In some primate cases, the positive effect of human interaction on animal tool use has been directly observed, for example in human-raised capuchin monkeys , and in wild vervet monkeys that were in regular contact with humans and human facilities . Experiments with captive orangutans, bonobos and other great apes  and non-tool-using tamarins and marmosets  demonstrate that human-directed training on tool-use tasks enhances performance, which might be expected for any animal amenable to operant conditioning. The human role in shaping or enabling tool use is perhaps most apparent in the studies of animals that do not use tools in the wild, through training  or provisioning  techniques, or the creation of artificial circumstances that may not occur in the wild . Non-tool object manipulation may also be influenced by humans: stone handling by Japanese macaques, for example, (i) is found solely among provisioned groups (who therefore have extra free time and energy through lowered foraging demands); and (ii) gradually disappeared from one site after provisioning was stopped [58,59]. Stone handling by rhesus macaques is also found only in captive animals .
An obvious difference between wild and captive conditions is the lack of natural predators in the latter, and the fact that captive animals are fed, watered and kept medically fit by humans . Freed from the need to search for food or maintain vigilance for predators, captive individuals have additional time and energy to devote to exploration and play, including testing the affordances of any novel materials they may encounter [62–64]. These animals also have more time available to observe the behaviour of others, adding further impetus to the positive feedback loop suggested earlier. Suggestions that terrestriality in captivity promotes tool use among typically arboreal primates  may also rely, in part, on freedom from vigilance, as the captive environment provides security for these animals on the ground. A terrestriality effect cannot explain captivity bias seen in non-arboreal species, however, including elephants, gorillas and rodents.
Models that examine the origins and development of animal tool use do not suggest that there are intrinsic differences between wild and captive animals, and in fact, it is central to most models that observations made in laboratories and zoos are directly applicable to animals living in natural conditions. However, it should be clear from the foregoing that components of models that rely, even in part, on innate factors such as intelligence and manual dexterity  are insufficient to explain the differences between wild and captive tool-use behaviour. Extrinsic factors, whether social , behavioural  or ecological [66,67], are therefore essential to a complete explanation of how and why tool use develops. Although in a speculative form, the implications and relevance of the captivity bias effect could likely complement existing studies of technological development in each hominin species (and population within each species). For present purposes, however, we will focus on Homo floresiensis, a taxon that has generated controversy over both its cognitive abilities and associated technology.
3. Homo floresiensis and captivity bias in hominin tool use
This section contains a brief initial attempt to consider how the principles underlying the captivity bias concept may apply to an extinct animal. Homo floresiensis was a late-surviving so-called dwarf hominin species, going extinct only in the past few tens of thousand years . It was chosen here, because the target of comparison for animal tool use has often been ‘early humans’ or ‘early hominins’ [69,70–73], and the extension of comparative studies to a geologically recent hominin species helps emphasize the potentially productive role of such comparisons for our understanding of human evolution. Importantly, for discussions of cognitive abilities, this species also presents an unprecedented small brain size for a Late Pleistocene hominin, although while brain size has been shown to correlate with innovative and tool-use behaviours in primates and birds , its effect on such behaviour is disputed [75,76]. Homo floresiensis is known from only one site, Liang Bua on the Indonesian island of Flores . Its known temporal range falls within the Late Pleistocene, from 95 000 to 17 000 years ago, although there is stone tool evidence from Flores that indicates hominin occupation from at least one million years ago (made by an unknown hominin) , and the Liang Bua evidence suggests that H. floresiensis occupied other parts of the island during drier phases . The brain volume of the H. floresiensis holotype, LB1, has been estimated by different methods as between 380 and 426 cm3 [80–82]. This brain size is similar to that of chimpanzees, and in the preserved fossil record only Ardipithecus ramidus and Sahelanthropus tchadensis, both over four million years old, had smaller brains [83,84]. Neither of those two putative hominins are presently associated with evidence for tool use.
Archaeological material found with H. floresiensis skeletal elements at Liang Bua includes evidence for plant-working and consumption of Stegodon and Komodo dragons, as well as possible evidence for fire [85,86]. The site's excavators considered this as a whole to be support for complex cognition . Much discussion has centred on the supposed sophistication of the associated stone tool assemblage (including retouched ‘perforators’ ), with some researchers considering the tools to be too advanced or similar to H. sapiens artefacts to have been made by H. floresiensis, and with the small brain of LB1 explicitly given as a primary reason for such a conclusion [88–92]. The Liang Bua lithic analysts have convincingly rebutted these claims, showing that the tools are the product of an independent lineage from H. sapiens, and that as an assemblage the Liang Bua tools are not any more sophisticated than Oldowan technology dating to well over one million years ago .
Nevertheless, the fact remains that no hominin with a brain size of ca 400 cm3 has previously been associated with knapped stone tools of any kind, and there remains no doubt that H. floresiensis anatomy is distinct from that of known tool-making hominins, including in the shoulder, arm and wrist configurations [94–96]. While it remains possible that H. floresiensis dwarfed from an isolated Southeast Asian H. erectus population , cladistics analysis suggests an earlier, non-erectus origin for the species . In the light of the data, the questions to be asked are not whether the Liang Bua Pleistocene tools are unusually sophisticated (they are not), but rather (i) why H. floresiensis was a stone tool maker at all, given how physically distinct it is from other tool-making hominins, and (ii) how its lineage sustained stone tool-use traditions on Flores for what appears to be at least a million years .
The first factor to consider here is evidence for population interaction. We do not yet know the population size or density of H. floresiensis, although it has been suggested that population sizes may have increased during wetter periods on the island . A testable hypothesis based on the captivity bias effect is therefore that such periods will show increased tool diversity and frequency on Flores, irrespective of whether foraging activities changed during the same periods. Homo floresiensis remains have been found in a single site; however, the species must have been more widespread on Flores (which is a reasonably large island at 13 500 km2) to maintain continuity over at least tens and potentially hundreds of thousands of years, including periods with very little evidence of occupation at Liang Bua. This population also survived the Toba volcanic super-eruption around 74 000 years ago on nearby Sumatra , and in fact, the highest concentrations of stone tools at Liang Bua are found immediately following that event . Studies of endemic primates on islands in Southeast Asia and elsewhere have shown that these populations typically have high population densities relative to mainland groups owing to release from interspecific competition [99,100], and there is no reason to assume H. floresiensis would have differed from this pattern. It is a reasonable hypothesis therefore that hominin encounter and interaction rates would have been higher on Flores than over a similarly sized area of mainland. This hypothesis includes encounters with artefacts left by other H. floresiensis individuals , possibly further increased through the circumscribed nature of island living. The likely cooperation necessary for hunting and transporting large or dangerous prey such as Stegodon and Komodo dragons  may have acted as an additional spur to bring individuals together.
Second, is there evidence for contact with other tool-using hominins, especially H. sapiens? Surrounding currents make Flores a difficult island to reach , and the lithic sequence shows little change over time, so on current data, the direct influence of other hominins on H. floresiensis tool use and manufacture is improbable. Nevertheless, the site's lithic team raised the possibility that continuities in stone technology before and after the disappearance of H. floresiensis at Liang Bua (near the Pleistocene/Holocene boundary) resulted from incoming H. sapiens groups incorporating techniques from H. floresiensis . If this suggestion is supported by further data, then the direction of information transmission from the resident species to the new arrivals would be as expected under the captivity bias effect, where it is resident population size and interconnectedness rather than cognitive ability that determines whether technological attributes are maintained.
Third, we can consider evidence for predation reduction and increased free time. The dwarfing of H. floresiensis (its stature has been calculated at 1–1.1 m)  has been attributed to the ‘island rule’, in which large-bodied animals reduce in size and some smaller animals grow larger. Despite debate over its generality, this rule has been found to hold for primates , with the driving factors including lower levels of predation and diminished resources. One effect of diminished resources may have also been to increase range sizes, however, leading again to greater rates of contact between different groups on the island. Predation reduction was likely an important influence in H. floresiensis dwarfing (although predators were certainly present on the island ), and features in the lower limbs that favour ‘low gear’ locomotion and stability over speed or agility [104,105] may be indirect evidence for a relaxation of the need for rapid escape. The similarly dwarfed prey species Stegodon florensis insularis is found throughout the Pleistocene Liang Bua sequence, only disappearing when H. floresiensis does, giving one indication perhaps that over-hunting owing to resource pressures did not occur on the island. Further, Late Pleistocene Stegodon do not decrease in size through the Liang Bua sequence , and despite overall low species richness on Flores, other endemic fauna, including the Komodo dragon, a smaller varanid, and three species of giant rat also survive throughout the Liang Bua Pleistocene record [85,104].
Taken together, therefore, the available data would suggest an interconnected, cooperating, stable population on Flores, likely facing predation pressures lower than that on the mainland, but likely also lower resource availability (tempered by the large size of the island and continued presence of large prey taxa). Interactions with other tool-using species may have been non-existent, based on lithic continuity evidence and the body size changes seen in both the hominins and other Flores fauna , which suggest isolation from outside influences. The long-term stability in stone tool forms suggests stable social groups, but at population densities too low to sustain cumulative changes in the technology. This maintenance of tool traditions, despite a suggested decrease in brain size of a third to a half depending on whether Homo habilis or early Indonesian Homo erectus is considered the founding population , was likely facilitated by increased free time granted by decreased vigilance requirements.
Although larger-brained than H. floresiensis, East African australopiths were manufacturing stone cutting tools from at least 2.6 million years ago . Even if the Flores dwarfing process initiated from a relatively small-bodied and small-brained hominin ancestor , tool use at Liang Bua may be seen as a context-dependent outcome of existing behaviour, rather than an improbable de novo occurrence . From a captivity bias perspective, H. floresiensis' maintenance of a variety of stone tool forms over many thousands of years, likely supported by non-stone tools, may therefore be explained, in part, as an instance of relaxed predation coupled to long-term social observation opportunities in an island setting. A prediction based on this perspective is that were H. floresiensis individuals ever to leave Flores (accidentally or otherwise) and reach the Southeast Asian mainland, their technological output would suffer a dramatic decline as the protective factors on Flores were removed. Although it is not the only plausible explanation, the Flores hominins could be viewed as naturally ‘captive’, and their technology assessed not from a priori expectations about brain size, cognition and anatomy but instead as another example of animal tool-use development in an enabling environment.
4. Hominins as tool-using animals
The case study included here has touched on only certain aspects of the ways in which animal tool use may provide additional options for developing our views on the behaviour of our hominin ancestors. Factors such as an increase in social contact, and the risk of predation or conflict can be linked in complex ways and, naturally, it is not only tool use that they influence, as shown for example by recent work tying the evolution of primate social living to diurnal predation .
Further, not every instance of increased free time and observation opportunity will automatically result in wholesale changes to toolkits, and the averaged picture of hominin technological evolution is actually one of prolonged periods of stasis (especially before the Middle Pleistocene). The key is to understand why change happens when it does, and here the animal tool-use record is useful in that it demonstrates that variations in technology characteristically involve the use of individual tools in sequence, and the addition of new tool forms and materials to the repertoire (i.e. increases in diversity), rather than cumulative modification of existing forms. The few possible examples of cumulative change in animal tools, such as the brush-tipped termite fishing probes made by Goualougo chimpanzees , are important but very rare exceptions. Where changes are observed in hominin behaviours that may be expected to influence tool use, such as ranging or diet, without accompanying changes in stone tool forms, we may therefore hypothesise that either (i) additional (non-lithic) tools were being incorporated into the toolkit, or (ii) existing tools were being sequentially combined for greater energy returns or to open up new extraction niches. Further, the captivity bias effect suggests that hominin groups, even if members of the same species, will show widely varying levels of tool use as a result of both cultural and stochastic processes . Cognition is irrelevant to these processes.
This discussion of the captivity bias effect is not intended to present an overarching hypothesis for all animal or even hominin tool use. It offers, instead, a complementary approach to the potentially circular arguments that assess the complexity of hominin tool use based on inferred cognition, where the tools themselves provide an important line of evidence for cognitive ability. Combined consideration of palaeoenvironmental, zooarchaeological and site occupation data (to reconstruct available resources, predator/prey abundance and information on possible group sizes from the density of material remains , respectively) can allow assessment of the living conditions for any hominin taxon. While these will suffer the same lacunae as any archaeological data, the independent lines of evidence that they provide may at least augment reconstructions based on cranial capacities and brain organization , whereas lessening reliance on often-scarce hominin fossil evidence. Variation seen among different groups of the same wild animal species [48,111,112], and between wild and captive individuals, suggests that we should focus on the specific ecological and social circumstances faced by the specific hominin groups if we want to posit anything other than a baseline cognition underlying tool use within and between those groups. The captivity bias effect posited here, while speculative at present and requiring further validation, provides a caution against generalizing findings of artefactual diversity, complexity or frequency and their behavioural implications from one hominin group to others, even of the same species, without explicit reasons for doing so.
Funding support was provided by the European Research Council though Starting Investigator grant no. 283959 (Primarch).
I thank Dora Biro for organizing the Royal Society International Scientific seminar ‘Tool use as adaptation’, for many helpful discussions about this paper, and for assistance preparing table 1. I also thank Helen Eaton, and the anonymous reviewers for their insights.
One contribution of 15 to a Theme Issue ‘Tool use as adaptation’.
- © 2013 The Author(s) Published by the Royal Society. All rights reserved.