“The core question for animal navigation,” according to Jeffery et al. “concerns what the reference frame might be and how the coordinate system encodes distances and directions within the frame” (sect. 2, para. 2). To answer this question, they hypothesize that a bicoded cognitive map underlies the navigational capacities of animals. Their presumption that navigation requires a cognitive map leads them to conceive of navigation as an abstract computational problem that requires a representational schema powerful enough to encode geometrical features of the familiar environment of the navigating animal. We argue that this presumption leads them to ignore important ethological evidence against cognitive maps.
The work of ethologists is especially relevant to the problem of navigating in a three-dimensional world. Ethologists have produced concrete methods for testing the cognitive map hypothesis during navigational tasks. Despite their passing references to ethology, Jeffery et al. fail to note that most ethologists believe that the experimental evidence runs against the hypothesis that cognitive maps underlie animal navigation (e.g., Bennett Reference Bennett1996; Dyer Reference Dyer1991; Wray et al. Reference Wray, Klein, Mattila and Seeley2008; Wystrach et al. Reference Wystrach, Schwarz, Schultheiss, Beugnon and Cheng2011). In one such experimental paradigm, the “displacement experiment,” an animal is moved prior to navigating to a familiar goal in a familiar range (e.g., see Wehner & Menzel Reference Wehner and Menzel1990). If the animal were to take a novel path from the displacement point to the goal, that would count as evidence for map-like navigation. However, typically the animal assumes a heading based on some other kind of information, providing evidence that the animal is using navigational systems (e.g., landmark learning, path integration, systematic search behavior, or a combination thereof) whose representational structure does not preserve the geometrical properties assumed by the cognitive map hypothesis. These studies, unlike those to which Jeffery et al. refer, propose alternate explanations that account for the navigational abilities of animals without presupposing the kinds of complex computation implied by a cognitive map. Simpler navigational strategies (e.g., visual snapshot matching) have been proposed to explain navigation to a goal (Judd & Collett Reference Judd and Collett1998), with landmarks serving to provide context-dependent procedural information rather than geometric positional information (Wehner et al. Reference Wehner, Boyer, Loertscher, Sommer and Menzl2006). This work shows that an animal may not need to abstract higher-order spatial characteristics of its environment in order to successfully navigate in it. However, without first determining what kind of information an animal is actually using while navigating, it is premature to be investigating the nature of the metric properties of the presumed cognitive map.
We suggest that reconceiving the problem of navigation in a three-dimensional world in ethological terms would shift the focus from describing the computational complexity of the geometric properties of the navigable environment to one wherein the information used by the animal to determine its location relative to some goal is investigated under ecologically valid conditions. By failing to consider the alternative explanations of how animals navigate in real environments, Jeffery et al.'s bicoded map hypothesis puts the hypothetical cart before the evidential horse.
Having ignored evidence against cognitive maps in animals, Jeffery et al. appeal to human behavioral evidence suggesting that the representations utilized in navigation preserve the planar properties of the navigation environment. For example, they summarize several studies indicating that humans are relatively poor at estimating the height of landmarks in a familiar environment (sect. 3.1, para. 7) and other studies suggesting that people overestimate the steepness of hills (sect. 3.2, para. 3). They mention yet another study suggesting that people do poorly when attempting to estimate the direction of objects in a multilayer virtual environment (sect. 3.3, para. 4). What is not clear is how these kinds of judgments are related to the information a navigating animal may use in an online navigational task. Jeffery et al. fail to acknowledge the possibility that these kinds of distance judgments (and their failures) may not be evidence for a bicoded cognitive map precisely because the information contained in such a representation may not be that used by a navigator in a real navigational task. Differential performance on vertical versus horizontal distance of objects, though perhaps having import for spatial perception in general, may simply be irrelevant to navigation.
We applaud Jeffery et al. for acknowledging that, in general, the problem of navigation is an evolutionary one. However, we find the possibility they endorse to be problematic. They argue that a fully three-dimensional map has never evolved, on the grounds that “[it] is possible that the cost-benefit ratio of the required extra processing power is not sufficiently great” (sect. 5.2, para. 4), and they suggest that a lower dimensional map-like representation supplemented with unspecified encoding heuristics could achieve successful navigational behavior at lower cost. Parallel reasoning suggests that an even lower dimensional representation (not even a map at all) augmented with appropriate heuristics would have evolved under similar cost constraints. That is to say, given their assumption that evolution works to minimize the cost of a trait (i.e., the cost of a more complex computational system), it becomes plausible that selection would act to produce ever less costly representational systems for navigation, as long as those systems are capable of approximating the same result. This would be consistent with the computationally simpler navigational systems found by ethologists to be sufficient for animals to navigate. In our view, Jeffery et al.'s hypothesis as to the evolutionary origins of the bicoded cognitive map suffers from a deficiency common in attempts to connect the presence of a trait with a story about selection history – that is, that mere consistency between the presence of some trait and some hypothetical selection pressure is the sole measure for plausibility. They have provided a list of how-possibly stories (two of which they reject, one they accept), when what is needed is a how-actually explanation that is properly grounded in the ethological evidence. Finally, we agree that a fully volumetric map may not have evolved, but perhaps because no cognitive map at all is needed for an animal to navigate in a three-dimensional world.
“The core question for animal navigation,” according to Jeffery et al. “concerns what the reference frame might be and how the coordinate system encodes distances and directions within the frame” (sect. 2, para. 2). To answer this question, they hypothesize that a bicoded cognitive map underlies the navigational capacities of animals. Their presumption that navigation requires a cognitive map leads them to conceive of navigation as an abstract computational problem that requires a representational schema powerful enough to encode geometrical features of the familiar environment of the navigating animal. We argue that this presumption leads them to ignore important ethological evidence against cognitive maps.
The work of ethologists is especially relevant to the problem of navigating in a three-dimensional world. Ethologists have produced concrete methods for testing the cognitive map hypothesis during navigational tasks. Despite their passing references to ethology, Jeffery et al. fail to note that most ethologists believe that the experimental evidence runs against the hypothesis that cognitive maps underlie animal navigation (e.g., Bennett Reference Bennett1996; Dyer Reference Dyer1991; Wray et al. Reference Wray, Klein, Mattila and Seeley2008; Wystrach et al. Reference Wystrach, Schwarz, Schultheiss, Beugnon and Cheng2011). In one such experimental paradigm, the “displacement experiment,” an animal is moved prior to navigating to a familiar goal in a familiar range (e.g., see Wehner & Menzel Reference Wehner and Menzel1990). If the animal were to take a novel path from the displacement point to the goal, that would count as evidence for map-like navigation. However, typically the animal assumes a heading based on some other kind of information, providing evidence that the animal is using navigational systems (e.g., landmark learning, path integration, systematic search behavior, or a combination thereof) whose representational structure does not preserve the geometrical properties assumed by the cognitive map hypothesis. These studies, unlike those to which Jeffery et al. refer, propose alternate explanations that account for the navigational abilities of animals without presupposing the kinds of complex computation implied by a cognitive map. Simpler navigational strategies (e.g., visual snapshot matching) have been proposed to explain navigation to a goal (Judd & Collett Reference Judd and Collett1998), with landmarks serving to provide context-dependent procedural information rather than geometric positional information (Wehner et al. Reference Wehner, Boyer, Loertscher, Sommer and Menzl2006). This work shows that an animal may not need to abstract higher-order spatial characteristics of its environment in order to successfully navigate in it. However, without first determining what kind of information an animal is actually using while navigating, it is premature to be investigating the nature of the metric properties of the presumed cognitive map.
We suggest that reconceiving the problem of navigation in a three-dimensional world in ethological terms would shift the focus from describing the computational complexity of the geometric properties of the navigable environment to one wherein the information used by the animal to determine its location relative to some goal is investigated under ecologically valid conditions. By failing to consider the alternative explanations of how animals navigate in real environments, Jeffery et al.'s bicoded map hypothesis puts the hypothetical cart before the evidential horse.
Having ignored evidence against cognitive maps in animals, Jeffery et al. appeal to human behavioral evidence suggesting that the representations utilized in navigation preserve the planar properties of the navigation environment. For example, they summarize several studies indicating that humans are relatively poor at estimating the height of landmarks in a familiar environment (sect. 3.1, para. 7) and other studies suggesting that people overestimate the steepness of hills (sect. 3.2, para. 3). They mention yet another study suggesting that people do poorly when attempting to estimate the direction of objects in a multilayer virtual environment (sect. 3.3, para. 4). What is not clear is how these kinds of judgments are related to the information a navigating animal may use in an online navigational task. Jeffery et al. fail to acknowledge the possibility that these kinds of distance judgments (and their failures) may not be evidence for a bicoded cognitive map precisely because the information contained in such a representation may not be that used by a navigator in a real navigational task. Differential performance on vertical versus horizontal distance of objects, though perhaps having import for spatial perception in general, may simply be irrelevant to navigation.
We applaud Jeffery et al. for acknowledging that, in general, the problem of navigation is an evolutionary one. However, we find the possibility they endorse to be problematic. They argue that a fully three-dimensional map has never evolved, on the grounds that “[it] is possible that the cost-benefit ratio of the required extra processing power is not sufficiently great” (sect. 5.2, para. 4), and they suggest that a lower dimensional map-like representation supplemented with unspecified encoding heuristics could achieve successful navigational behavior at lower cost. Parallel reasoning suggests that an even lower dimensional representation (not even a map at all) augmented with appropriate heuristics would have evolved under similar cost constraints. That is to say, given their assumption that evolution works to minimize the cost of a trait (i.e., the cost of a more complex computational system), it becomes plausible that selection would act to produce ever less costly representational systems for navigation, as long as those systems are capable of approximating the same result. This would be consistent with the computationally simpler navigational systems found by ethologists to be sufficient for animals to navigate. In our view, Jeffery et al.'s hypothesis as to the evolutionary origins of the bicoded cognitive map suffers from a deficiency common in attempts to connect the presence of a trait with a story about selection history – that is, that mere consistency between the presence of some trait and some hypothetical selection pressure is the sole measure for plausibility. They have provided a list of how-possibly stories (two of which they reject, one they accept), when what is needed is a how-actually explanation that is properly grounded in the ethological evidence. Finally, we agree that a fully volumetric map may not have evolved, but perhaps because no cognitive map at all is needed for an animal to navigate in a three-dimensional world.