Why study cognition in the wild (and how to test it)?

An animal’s behavior is affected by its cognitive abilities, which are, in turn, a 21 consequence of the environment in which an animal has evolved and developed. Although behavioral ecologists have been studying animals in their natural environment 23 for several decades, over much the same period animal cognition has been studied 24 almost exclusively in the laboratory. Traditionally, the study of animal cognition has 25 been based on well-established paradigms used to investigate well-defined cognitive 26 processes. This allows identification of what animals can do, but may not, however, 27 always reflect what animals actually do in the wild. As both ecologists and some 28 psychologists increasingly try to explain behaviors observable only in wild animals, we 29 review the different motivations and methodologies used to study cognition in the wild 30 and identify some of the challenges that accompany the combination of a naturalistic 31 approach together with typical psychological testing paradigms. We think that studying 32 animal cognition in the wild is likely to be most productive when the questions 33 addressed correspond to the species’ ecology and when laboratory cognitive tests are 34 appropriately adapted for use in the field. Furthermore, recent methodological and 35 technological advances will likely allow significant expansion of the species and 36 questions that can be addressed in the wild. 41


5
This question is as broad as the question of why one should study animal 90 cognition at all. The benefits and challenges of working in the field, however, lend 91 themselves to asking certain questions rather more readily than others. In this section, 92 we describe some of the various reasons why scientists choose to work in the field, both 93 in terms of the aims of their research programmes, but also in terms of the practical 94 benefits of working outside of the laboratory. The Ecological Approach 97 Research programmes within the Ecological Approach involve the testing of 98 hypotheses that concern how natural selection might have shaped animal cognition. McGregor & Healy, 1999). 112 Although that work was located in the laboratory, ecologically-based questions 113 have also been addressed in the field. For example, a long-running study of 114 6 hummingbird cognition in the wild has tested a range of a priori predictions about the 115 information to which hummingbirds "should" pay attention to in order to forage 116 effectively (Healy & Hurly 2013). Using field experiments that create a simplified 117 version of their natural environment, it is possible to investigate whether hummingbirds 118 can pay attention to various types of information present in the environment, as well as 119 the kinds of information they preferentially use during foraging. One of the challenges 120 of the cognitive ecology research programme, however, is to objectively identify a 121 priori predictions about types of information to which animals "should" pay attention.  Similarly, captive and wild nectivorous bats trained to feed from 133 echoacoustically distinctive flowers also preferred to use spatial cues rather than the 134 flowers' unique acoustic shape when returning to feed at a rewarded flower (Thiele & 135 Winter, 2005). As many flowers may look similar but each sits in a unique location, a 136 possible post-hoc explanation for the preference of spatial rather than feature cues (e.g.  Recently, however, interest has begun to include the direct investigation of the 152 fitness consequences of cognition, inspired by the success of the work on the evolution 153 of learning in Drosophila, in which flies respond to artificial selection on their 154 associative learning abilities (e.g., Mery & Kawecki, 2003. Unlike the cognitive 155 ecology focus on the ability of animals to learn particular ecologically relevant 156 information, this more recent interest has tended to be directed towards "general" 157 cognitive ability, typically assessed using one or more "problem-solving" tasks.

158
One commonly-used example is the "lid-flipping" task often presented to birds 159 as a novel or innovative foraging task (e.g., Boogert, Giraldeau, & Lefebvre, 2008;    Perhaps the first of these, and one that motivates many keen to investigate the 205 evolution of cognition, is that by working with animals in the wild, one can potentially 206 access a much wider range of study species than just those suited to the laboratory.

207
Additionally, this might mean gaining access to investigating the mechanisms that 208 underlie "natural" behaviors, which are not easily produced or tested in the laboratory.

209
In some cases, the behavior of interest is carried out on a scale that excludes it 210 from being studied in any real way in the confines of the laboratory environment. For 211 example, determining whether avian migrants truly know the location of their wintering 212 grounds, rather than just the distance and direction to fly in order to reach them, relies 213 10 on experiments carried out on a grand scale impossible in the laboratory (Perdeck,214 1958; Thorup et al., 2006). 215 Similarly, the homing flights of pigeons are impressive because of the distances 216 involved. Pigeons released in unfamiliar territory, many kilometres from their home 217 loft, can reliably find their way home using multiple sources of information from their 218 surroundings to fix their position and chart a homeward trajectory (Wallraff, 2005).  One key feature of the laboratory species commonly used to investigate animal 237 cognition, such as pigeons, rats, and zebra finches, is their ability to thrive in captivity.
Pigeons and rats in the laboratory can also readily be trained to search for food or to 239 modify their behavior to gain reward, e.g., through pressing levers (e.g., Adams & 240 Dickinson, 1980) or pecking at lights (Brown & Jenkins, 1968  Access to "natural" environments 255 The information available to an animal in the wild is very different from the 256 information available to an animal in the laboratory. In some respects, this may seem to 257 be obvious. What may be less obvious is that the difference in information between the 258 laboratory and the wild can be qualitative as well as quantitative. 259 Typically, differences between the laboratory and the wild are discussed in 260 quantitative terms: the laboratory is barren or sparse, whereas the field has more 261 confounded variables. The implication is that there is more information available to the 262 animal in the wild, more potentially confounded cues, which make understanding how 263 12 animals use a particular source of information more challenging. Even critiques of the 264 laboratory environment rely on this logic, arguing that the lack of information makes 265 the laboratory somehow unnatural, which then limits its value for testing ecologically 266 relevant cognition (e.g., Jacobs & Menzel 2014).

267
What may be less often appreciated, however, is that the environment of the 268 laboratory can structure the kinds of information that animals acquire. Take, for   (Baenninger, 1967), preventing it from either learning a task appropriately or not at all 335 (Bowman, 2005). This can then lead to the conclusion that the animal cannot learn 336 information that it actually did learn or to the interpretation that the behavioral response 15 is a result of impaired cognition, rather than that the impairment is due to a stress 338 response.

339
For example, male rats Rattus norvegicus outperform female rats in spatial tests concern what an animal can learn and those that concern what an animal has learned.    the locations of their neighbors' territories, treating the song of a neighbor apparently 438 sung in the "wrong" territory as they would the song of a stranger (Godard, 1991). 439 Multiple playbacks can also be used to assess whether the information that an 440 animal has learned is the same for different stimuli, using a habituation-dishabituation    Ideally, animals suitable for the experimental study of cognition in the wild 465 should be reliable, observable, and amenable. Reliable animals are those that can be 466 found easily on multiple occasions and will perform the behavior of interest sufficiently 467 frequently to allow collection of adequate data. Animals that are rare or perform 468 behaviors that occur sporadically would not be reliable and may be challenging to study 469 in the wild. 470 Rufous hummingbirds have been a useful example for studying cognition in the 471 wild because they are very reliable. Throughout the breeding season, males are almost 472 always found within their individual feeding territories, which they fiercely defend from 473 rivals (Kodric- Brown & Brown, 1978). As they are highly motivated to find food and 474 typically feed every 10-15 min, it is relatively simple to collect sufficient data even 475 though their breeding season may be as short as 6 weeks.

476
Observable animals are those that can be identified and whose behaviors can be    The enthusiasm for investigating cognition in the wild is being greatly benefitted 541 by recent advances in technology, which are enabling access to many more species and 542 questions that require animals to be followed over long distances, for long periods of  likely that these devices will become increasingly useful in the future.

571
Although less often used so far, computer vision also has significant potential for 572 studying 'wild' cognition. Unlike PIT tagging and biologging, which involve attaching 573 devices to animals, computer-vision technology allows researchers to track and record 574 the behavior of animals without requiring the animal to carry any equipment.  In addition to providing economical tracking solutions, similar methods can be 586 used to reconstruct the visual information available to animals navigating in the wild. Using multiple overlapping photographs of an area, for example, three-dimensional 588 reconstruction techniques can be used to generate a three-dimensional model of natural 589 environments, which alongside the reconstructed paths of an animal, allow researchers 590 access to the "view from the cockpit" of animals travelling through their worlds (Stürzl,591 Grixa, Mair, Narendra, & Zeil, 2015). These data can be used alongside experiments 592 and computational modelling to quantify and manipulate information available to 593 animals in their natural environments in unprecedented ways. The study of cognition in the wild, especially spatial navigation, seems likely to 597 continue gathering momentum as technological advances increase our access to ever 598 more species and their behaviors in the field. We are optimistic about the implications 599 of such work.