To understand the origins of the concept of behavioral flexibility, we have to recall a conflict over the explanation of animal behavior that now seems quaint. In the 1930s–1950s there was a sharp distinction between two approaches. On the one hand, comparative psychologists (mainly Anglo-Saxon) were seeking to explain behavior in terms of the experiences of reward and punishment that an individual animal had had. On the other hand, ethologists (mainly European) were seeking to explain behavior in terms of the species to which an animal belonged, and the evolutionary and developmental history characteristic of that species. Of course, things were much more nuanced than this crude summary implies. But these polar positions illuminate the first significant use of the term “behavioral flexibility” that we have been able to find in the literature of what we would now call comparative cognition. Poirier (1969) introduces the term to cover the variations in behavior he observed between and within troops of Niligri langurs (Semnopithecus johnii) in the wild. In particular, troops varied in their diet, and although these variations were partly explained by variations in availability (in turn due to variations in the intrusion of human activities into the monkeys’ home ranges), Poirier found that a full explanation required an appeal to learned traditions of food acceptability. Within troops, younger animals and females were more ready to accept new foods. Thus behavior – even species-typical behavior – could not be accounted for by the kind of fixed-action pattern that the early ethologists described (e.g., Lorenz, 1932/1970; Tinbergen & Perdeck, 1950), whose very name implied inflexibility.
Fifty years after the publication of Poirier’s paper, all of this seems completely obvious, and we certainly no longer think in terms of a dichotomy between “innate” and “learned” behaviors. Poirier, however, found it necessary to devote most of the Discussion of his results to the point that “much of primate behavior results from a sizable learning component” and that “animals do not live by innate behavior alone” (p. 130). In this earliest, and still current, sense, therefore, “behavioral flexibility” simply encompasses all learning to adapt to the particular conditions of an individual’s environment – in effect, the whole of animal cognition.
Behavior is not always flexible
To say that behavior is flexible in this sense can seem like a glimpse of the obvious. To make sense of it, we need to remember the extent to which the behavior of individual animals is often inflexible – or is thought to be. Animals do not behave at random, and if there is no reason to change their behavior, we should not expect that they will do so. Some examples of relative inflexibility include:
Individual foraging site fidelity
Many animals live, roost, or nest in colonies to which they return either after each day’s foraging or (especially when incubating mates or young are being fed) repeatedly during foraging. It is common for such colonial central place foragers to be faithful to a particular foraging site, even though they would be equally capable of reaching, and using, the sites used by other colony members; and even though these areas are not defended as a territory. E.A. Morgan, Hassall, Redfern, Bevan, and Hamer (2019) list a number of examples of such individual foraging site fidelity, and explore its extent – and limitations – in the European shag (Phalacrocorax aristotelis), a colonially nesting diving bird. Across the 70 shags they studied, the overlap between the utilization distributions of foraging sites on repeated trips varied considerably, from 0.06 to 0.75 on a scale from 0 to 1; but it was far higher than would be expected by chance. Furthermore, such inflexibility may be advantageous: female shags with a higher repeatability score tended to hatch their eggs earlier and were in better body conditions than those with lower scores. Obviously, since these preferred foraging sites consisted of a patch of open sea, there was no question of territorial defense.
Consistency of foraging method
Even when attacking a particular kind of prey in a given location, animals may have more than one method of attack available. In many coastal sites in northern Europe, Eurasian oystercatchers (Haematopus ostralegus) prey on the common mussel, Mytilus edulis, especially where large beds of mussels are found, as in the drowned estuaries of south-west England and Brittany. Three distinct techniques that oystercatchers use for opening mussels have been documented. One involves stabbing at the opening between the valves of the shell. The other two both involve hammering through the shell. However, in one case the bird hammers through the dorsal surface of the shell in situ, while in the other it removes the mussel to an “anvil,” a nearby location where it will be held in place while the oystercatcher strikes it, and hammers through the ventral surface of the shell (Goss-Custard, Durell, & Ens, 1982; Norton-Griffiths, 1967). Goss-Custard et al. observed 28 individually marked oystercatchers; roughly equal numbers used each technique, and all but one of the birds used a single technique exclusively. Another example of consistent variation in foraging method has been described in humpback whales: although humpbacks generally use bubbles to herd fish prey, sometimes in complex ways, the details (for example, whether or not bubble clouds are used) vary between individuals and regions (Wiley et al., 2011).
Consistency of foraging parameters
Within a particular foraging method, the parameters of the behavior may show consistent individual differences. For example, air-breathing divers face a trade-off between time available for foraging underwater and time available for replenishing oxygen stocks on the surface; in general, we would expect the balance to be struck at an underwater duration somewhat less than the maximum physiologically possible (Kramer, 1988). Potier, Carpentier, Grémillet, Leroy, and Lescroël (2015) examined the repeatability of the parameters of diving behavior in the great cormorant (Phalacrocorax carbo), and found considerable variation in repeatability between individuals and between parameters; in general, however, repeatability was substantially greater than would be expected by chance. For some parameters (e.g., mean time underwater per dive) high repeatability – inflexibility – was associated with reduced overall foraging efficiency, a likely proxy for fitness. But for others (e.g., mean time spent underwater per diving bout) the relation was positive.
None of these examples is absolute. We have already seen that there was individual variation in repeatability of behavior in both the diving examples; and in the case of oystercatcher foraging methods, subsequent work by Goss-Custard and colleagues showed that individuals tend to shift from stabbing to hammering as they grow older (Goss-Custard & Sutherland, 1984), and that at each age, the predominant method is the one associated with the higher fitness (Durell, Goss- Custard, Caldow, Malcolm, & Osborn, 2001). But they do show that flexibility is neither ubiquitous nor always advantageous. Indeed, Davis, Schapiro, Lambeth, Wood, and Whiten (2019) have argued that, because of overall limitations on cognitive capacity, some degree of behavioral inflexibility (or, as they term it, conservatism) is an inevitable result as individual behaviors become more complex. In a series of experiments on captive chimpanzees, they showed that there was little conservatism (i.e., high flexibility) when chimpanzees were required to inhibit a simple well-established response (using one of two available handles) to extract food from a puzzle box. But the apes showed greater conservatism when the apes had to inhibit a more complex, also well-established response, involving two successive doors and removing an obstacle in order to access food in a different puzzle box.
Flexibility as variation in behavior in response to environmental change
It is, therefore, not unreasonable for ethologists to start from an assumption of behavioral inflexibility: that an animal’s behavior can be predicted from what kind of animal it is. Nevertheless, behavioral ecology has revealed many cases where the behavior of a given species varies depending on the environmental conditions, either between individuals living in different environments, or within the lifetime of an individual. Indeed, many of the optimizing models that have been the theoretical bedrock of behavioral ecology predict that such variation should occur. The following brief lists give a few examples of situations where it has been demonstrated. They are selected because their authors specifically referred to behavioral variation as a function of environmental change as “behavioral flexibility,” and they thus show the enduring use of this terminology over a two-decade period.
In some cases, “behavioral flexibility” is used to describe cases where behavior varies between populations living in different environments, or between groups within the population (e.g., between sexes or age classes). For example, Price, Tonn, and Paszkowski (1991) demonstrated variations in the prey choice, habitat use, and activity pattern between males and females and between adults and juveniles in fathead minnows (Pimephales promelas); Klett-Mingo, Pavoni, and Gil (2016) found that vigilance in great tits (Parus major) living near a major airport varies as a function of aircraft noise; and Ben Cohen and Dor (2018) found that the exploratory behavior and neophobia of house sparrows (Passer domesticus) varies along a climate gradient in the same way as morphological characteristics such as size and the darkness of the plumage.
In other cases, “behavioral flexibility” is used to describe behavior change that we can be confident is occurring within individuals, either because individuals have been tracked, or because the changes affect whole populations as a function of time of day or year. For example, Thompson and Baldassarre (1991) showed that the activity patterns of several species of migrant ducks in Yucatan change with foraging site and temperature; Knight, Vanjaarsveld, and Mills (1992) observed the unusual phenomenon of allo-suckling in spotted hyenas (Crocuta crocuta) after a prolonged drought; Palagi, Antonacci, and Cordoni (2007) showed how a play signal could switch the response to an ambiguous behavior from aggression to play in young lowland gorillas (Gorilla gorilla gorilla); and Christensen-Dalsgaard, May, and Lorentsen (2018) found that foraging site use in black-legged kittiwakes (Rissa tridactyla) varies as a function of distance from nest site, in interaction with weather conditions.
Flexibility mediated by learning and culture in the wild
Where behavior differs between populations, the differences could be mediated by genetic, instinctual mechanisms, and indeed in some cases authors (e.g., Ben Cohen & Dor, 2018) have sought to correlate behavioral and genetic variation. Other authors, however, invoke the term “behavioral flexibility” to mark the fact that animals learn to adapt their behavior to their varying environments. Such learning inevitably builds on the characteristic behavioral repertoire of the species concerned, and several general theories have been produced to detail the interaction of evolutionary and cognitive processes in bringing about adaptive behavioral variation. Examples include the theories of Bindra (1978), West-Eberhard (2003, chapter 18), Mery and Burns (2010), and Fawcett, Hamblin, and Giraldeau (2013) for behavior in general, and Taborsky and Oliveira (2012) for social behavior in particular.
In looking for demonstrations of the role of learning in adaptive phenotypic variation under natural conditions, we face the difficulty of disentangling learning effects from those of genetic variation. Furthermore, the mere occurrence of learning is not sufficient, since learning is involved in the emergence of many behaviors that are typical of a species (Hailman, 1969). What we are looking for are cases where behavioral differences within a population (or within the lifetime of an individual) can be attributed to different learning experiences. Despite the difficulties, learning has been claimed as an explanation of natural behavioral flexibility across a wide taxonomic range, including in the feeding of gastropods (reviewed by Elliott & Susswein, 2002), in prey capture by ladybird beetle larvae (Anisolemnia tetrastictas: Dejean, Gibernau, Lauga, & Orivel, 2003) and jumping spiders, especially Portia spp. (e.g., Jackson & Pollard, 1996, and much subsequent work from this research group), in nest-site selection by Indian house crows (Corvus splendens: Yosef, Zduniak, Poliakov, & Fingerman, 2019), in the detection of water for drinking by barbastelle bats (Barbastella barbastella; Russo, Cistrone, & Jones, 2012), in the use of sponge tools by bottlenose dolphins (Krützen et al., 2005), and indeed in the variation of foraging behavior of Niligri langurs, as in the earliest use of the term “behavioral flexibility” that we have found (Poirier, 1969).
Problem-solving capacity as behavioral flexibility
For many authors, the capacity for learning alone does not encompass what they mean by behavioral flexibility. A popular alternative, which has been used across a range of vertebrate taxa, is the capacity for learning the reversal of a task that has been well trained. Among those who have used reversal learning as an index of behavioral flexibility are Pintor, McGhee, Roche, and Bell (2014) studying Northern pike (Esox lucius); Petrazzini, Bisazza, Agrillo, and Lucon-Xiccato (2017) studying sex difference in the cognition of guppies (Poecilia reticulata); Szabo, Noble, Byrne, Tait, and Whiting (2018) studying tree skinks (Egernia striolata); Boogert, Monceau, and Lefebvre (2010) studying Zenaida doves (Zenaida aurita); Logan (2016b) studying great-tailed grackles (Quiscalus mexicanus); Gilbert-Norton, Shahan, and Shivik (2009) studying the effect of Skinnerian schedules of reinforcement on coyotes (Canis latrans); and Manrique and Call (2015) studying how great ape learning changes with age. This suggestion chimes with the popular view that a key feature of executive control in cognition is the capacity to inhibit a prepotent behavior, which is obviously implicated in reversal learning. A few authors have gone beyond that, and have considered that a better measure of behavioral flexibility would be the capacity for serial reversal learning, that is the capacity to “learn the rules” of reversal learning so that successive reversals are made more quickly. Examples include Liu, Day, Summers, and Burmeister (2016) in a study of the green and black poison dart frog Dendrobates auratus (they refer to serial reversal learning as requiring “advanced” behavioral flexibility); Bond, Kamil, and Balda (2007) in a comparative study involving three species of corvid; and Chow, Leaver, Wang and Lea (2015) studying Eastern gray squirrels.
In recent years, however, there has been an increasing tendency to use the term “behavioral flexibility” in particular connection with animal problem solving. Although any situation requiring learning can be described as solving a problem, “problem solving” typically refers to the spontaneous solution of physical problems, most often the extraction of food from inaccessible places. The places concerned could either be naturally occurring or experimentally contrived, but experimental situations – puzzle boxes – are easier to study, though they may be deployed in natural situations as well as in laboratories. Authors using the term “behavioral flexibility” either as a synonym for problem-solving ability or in an attempt to explain it, include Webster and Lefebvre (2001) in a comparative study of several species of birds in Barbados; Isden, Panayi, Dingle, and Madden (2013) studying spotted bowerbirds (Ptilonorhynchus maculatus); Mangalam and Singh (2013) examining urban bonnet macaques’ (Macaca radiata) strategies for extracting food from anthropogenic sources; Loukola, Perry, Coscos, and Chittka (2017) studying bumblebees (Bombus terrestris); and Chow, Lurz, and Lea (2018) comparing Eurasian red with Eastern gray squirrels (Sciurus vulgaris and S. carolinensis). Griffin and Guez (2014) reviewed the relation between experimental studies of problem solving and the emergence of innovative behaviors in the wild, invoking behavioral flexibility as a linking mechanism.
Brain size, behavioral flexibility, and innovation
This tendency to identify behavioral flexibility with problem-solving ability is seen most prominently in attempts to make large-scale comparative assessments of cognitive differences between taxa, and to relate them to possible causes and consequences. The majority of these papers have come from Lefebvre and his colleagues. They have used published anecdotal reports of novel foraging methods in birds as an index of the species’ capacity for problem solving and hence of its behavioral flexibility, and have then examined how this measure correlates with residual forebrain size. Residual forebrain size is the excess of forebrain mass over what would be predicted from the correlation across a wide range of species between brain and body mass (cf. Jerison, 1985). Lefebvre, Whittle, Lascaris, and Finkelstein (1997) first demonstrated such a correlation between innovation reports and residual brain size across 17 different orders of birds, using data from North America and the British Isles, and Lefebvre, Gaxiola, Dawson, Timmermans, Rosza, and Kabai (1998) reported a similar correlation across orders and parvorders of Australasian birds. Similar correlations have been reported in other taxa, for example across species in the mammalian orders Primates (Reader & Laland, 2002) and Carnivora (Benson-Amram, Dantzer, Stricker, Swanson, & Holekamp, 2016). The basic analysis has also been refined. For example by Nicolakakis and Lefebvre (2000) using an enlarged sample of northern European birds, while Overington, Morand- Ferron, Boogert, and Lefebvre (2009) showed that the variety of innovations recorded was more important than the mere number of innovations. There have also been attempts to refine the areas of the brain responsible for the correlation: in Reader and Laland’s analysis, they focused on neocortex size, while Timmermans, Lefebvre, Boire, and Basu (2000) returned to the dataset used by Lefebvre et al. (1997), and showed that the key brain area in birds was the hyperstriatum. Because of the reliability of this correlation, residual brain size is sometimes used as a proxy for observed behavioral flexibility in comparative studies, as in the analysis carried through by Sol, Szekely, Liker, and Lefebvre (2007).
The consequences of behavioral flexibility
There is a potential feedback relationship between behavioral flexibility and brain size: The greater the adaptive advantages of behavioral flexibility, the greater the selective pressure to increase the size of the brain. However, within the life of an individual organism, species brain size can reasonably be taken as a precursor of behavioral flexibility, rather than a consequence – although engaging in flexible behavior might cause brain growth, and has been claimed to induce desirable changes in brain activity (e.g., Belleville et al., 2011). But what might the consequences of higher behavioral flexibility be? Several surveys have suggested that, between taxa, higher behavioral flexibility contributes to fitness. Nicolakakis, Sol, and Lefebvre (2003) showed that, across the parvorders of birds worldwide, those with higher recorded innovation rates (hence, behavioral flexibility) tended to include more species, suggesting that evolution (and in particular, speciation) is speeded up in such taxonomic groups. Sol et al. (2007) extended this result by showing that, across both populations and taxonomic families of birds, higher residual brain mass is associated with lower adult mortality per year, even taking into account factors such as body mass and social structure. Within the parrots, Schuck-Paim, Alonso, and Ottoni (2008) showed that those with greater residual brain size (implying greater behavioral flexibility) tended to live in a wider range of climate types. A more ambiguous result was found by Sol, Lefebvre, and Rodriguez-Teijeiro (2005): They showed that migratory species of passerine bird tended to have lower residual brain size than related non-migrant species. Sol et al. interpret this result as implying that sedentary species have to rely on innovative feeding techniques to cope with winter conditions in the temperate zone, though it is not obvious why a change of seasons would impose more demands on foraging technique than a change of continents.
All these studies compare averages for species or higher taxonomic groups with one another. Within species, evidence that behavioral flexibility is correlated with fitness has been harder to find, regardless of how behavioral flexibility is measured. Within species, not all evidence supports the idea that flexibility enhances fitness: In their studies of spotted bowerbirds, Isden et al. (2013) used a battery of six cognitive tests including reversal learning, but could not derive any measure that would predict breeding success. Huebner, Fichtel, and Kappeler (2018) found a similar result in gray mouse lemurs (Microcebus murinus). And Madden, Langley, Whiteside, Beardsworth, and Van Horik (2018) found that speed of reversal learning in pheasant chicks (Phasianus colchicus) was negatively correlated with their survival as adults.
There is one particular context, however, in which many authors have associated greater success with higher behavioral flexibility, and that is in adaptation to anthropogenic environmental change and in particular to urbanization. The spread of urban development commonly results in a great loss of biological diversity in the wildlife of an area, but this may be accompanied by an increase in wildlife biomass; a few species do very well in the urban environment (e.g., Chace & Walsh, 2006). This may be due to dominance of resources by a few invasive species (Shochat et al., 2010). Both success in the urban environment and success in novel environments generally, and hence invasion success, have been repeatedly linked with behavioral flexibility (e.g., Griffin & Diquelou, 2015, for the case of the invasive Indian myna in Australia). There is an obvious logical link, in that to succeed in any new environment, and in particular in the urban environment, an animal has to be able to vary its behavior, but there is also substantial empirical evidence of a link, as demonstrated in the reviews by Sih (2013), Lowry, Lill, and Wong (2013) and Barrett, Stanton, and Benson-Amram (2019).
So what does behavioral flexibility mean?
We have seen that across the past half-century “behavioral flexibility” has been used to mean everything from an absence of instinctual predetermination of behavior, through to relative performance under quite specific kinds of cognitive challenge. There is no simple historical progression between these different meanings, but the trend is towards using “behavioral flexibility” more specifically where learning or problem solving are demonstrably involved, with the term “phenotypic flexibility” being used where adaptive variation has an unknown origin, and might be due in part to genetic differences between individuals and populations.
Unsurprisingly, these different definitions do not cohere, empirically speaking. There are manipulations that leave problem solving unimpaired but damage learning of the reversal of the same problem (e.g., early social deprivation in rats; M. J. Morgan, 1973). Within species individual differences in simple associative learning speed are not reliably correlated with speed of solving more complex problems (e.g., in great-tailed grackles; Logan 2016a). Where a number of different cognitive tests are given to the same animals, correlations between their performance levels tend to be low (e.g., Isden et al., 2013). Specifically, the two capacities that have been most often associated with behavioral flexibility, reversal learning and innovative problem solving, do not necessarily correlate across individuals, as for example in the study of common mynas (Sturnus tristis) in Australia by Griffin, Guez, Lermite, and Patience (2013).
Inevitably, these conceptual and empirical inconsistences have resulted in some confusion (Audet & Lefebvre, 2017). One possible response would be to advise against using the term “behavioral flexibility” at all. However, we want to argue for a different approach: to give it a more precise meaning. The remainder of this paper explores how that might be done.