Simplicity is as intuitive as it is popular.1 Simple explanations are easier to understand, simple formulae are easier to use, and when a single equation describes radically different systems, the world appears to be united through simplification. Comparative cognition is no stranger to erring on the side of simplicity, as can be seen in the long-standing debate about chimpanzee mindreading abilities.2 One of the most common heuristics in experimental comparative cognition advises that, all else being equal, one ought to prefer the simplest explanation of animal behavior: one that interprets observed behavior in terms of the simplest cognitive mechanism or ability. Unfortunately, the theoretical basis for this “simplicity heuristic” has not been adequately established.3 In this chapter, I examine how the simplicity heuristic adversely affects a relatively new tool in experimental comparative cognition: cognitive models. It does so, I argue, by directing intellectual resources into the development and refinement of putatively simple cognitive models at the expense of putatively more complex ones, which in turn directs experimenters to develop tests to rule out these simple models. The result is a state of affairs wherein putatively simple models appear more successful than less simple ones not in virtue of their epistemic superiority, but, rather, because a disproportionate amount of resources have been devoted to their development and evaluation. This has, in turn, adversely affected the design and direction of behavioral experimentation aimed at describing cognitive processes in animals, shutting down alternative research programs. I conclude that moving toward a more quantitative science of animal minds is likely to improve the explanatory and predictive power of animal cognition research, but only if these models do not fall prey to existing biases such as the simplicity heuristic.
The simplicity heuristic is intuitively appealing, widely accepted, and commonly employed in the design and interpretation of experimental results (Shettleworth 2012). However, it is currently unjustified, and may be unjustifiable, on either conceptual or empirical grounds (Fitzpatrick 2009; Meketa 2014; Starzak 2016; Sober 1998; Sober 2005), for several reasons.
First, the concept of simplicity is too ambiguous to offer any substantive guidance in either constructing or interpreting experiments, as there is no such thing as simplicity simpliciter. Not only are there many ways to “simplify” scientific entities, but there are also many dramatically different scientific entities that may be simplified. We might, for instance, simplify through homogenization by reducing the number of entity types; by reduction through reducing the total number of entities; by idealization through eliminating some features but retaining critical others; and so on. Similarly, the subject of simplification may be either the form of our explanations (i.e., the number of variables in equations, the length of the description, how easy it is to understand) or its content (i.e., the number of token or types of entities, processes, or properties posited). In addition, the simplicity of explanations may not correspond to the simplicity of the entities it postulates. For instance, an explanation using a large number of variables may be needed to explain the mechanism of a single entity (such as a human liver), while an explanation using a single formula may explain the behavior of a complex system (such as predator-prey interactions as described by the Lotka-Volterra equations).
Second, this ambiguity is compounded by the fact that the cognitive ontology to be simplified is itself underspecified: do we prefer to simplify mechanisms, processes, or structures? Answering this question is critical, as what it means for a mechanism to be simple may not be what it means for a process to be simple, or what it means for a structure to be simple, and so on.
Third, problems remain even when simplicity is indexed to particular cognitive structures, processes, or mechanisms. One problem is that some mechanisms do not lend themselves to classification along simple-complex lines. For instance, comparative cognition researchers typically believe that representational mechanisms are more complex than associative mechanisms, and yet there is no obvious intuitive sense in which this is true. The need to craft an unintuitive explanation for why representational mechanisms are more complex than associative mechanisms suggests that something other than simplicity is likely driving the intuition.4 A more general problem is that even if unintuitive answers are accepted, it is incumbent on the proponent of the simplicity heuristic to explain why we should prefer these simpler ontologies. That is, why should the fact that something is simpler make it a better default hypothesis? Why should the burden of proof be on the more complex hypothesis when explaining animal behavior?
If a given behavior is indeed more likely to be underwritten by a simple mechanism, then we would have a strong reason to place the burden of proof onto the complex explanation. Support for this probabilistic claim has typically been couched in term of the “evolvability” of complex cognition. Arguments of this form aim to identify selection pressures or constraints that would (ceteris paribus) favor the evolution of simpler cognitive mechanisms over more complex ones. For instance, one argument holds that, ceteris paribus, the high metabolic costs of complex cognition (a product of the metabolically expensive brain tissue needed to sustain it) should favor the selection of simpler and less costly cognitive mechanisms. Such evolvability arguments are numerous and, if true, would potentially shift the burden of proof as directed by the simplicity heuristic. However, to the extent that justifications of the probabilistic claim have been attempted, few have been persuasive (Sober 2009; Meketa 2014; Mikhalevich 2015). Consider the metabolic costliness argument above. First, the argument fails to take into account the trade-offs between increased metabolic cost and the benefits that accrue to clever organisms: a bigger or denser brain needed to sustain increased cognitive function may be a reasonable price to pay for the ability to, e.g., outwit prey or predators, or to maintain close social ties needed for fitness-enhancing cooperation, and so on. Second, it wrongly presupposes that all or most increases in cognitive complexity require increases in metabolic output. While brains are indeed pound-for-pound more expensive than other organs, the addition of complex cognitive functions may not require the addition of expensive new neural tissue – e.g., it may require a no-cost repurposing of existing tissue. Moreover, the precise relationship between increases in neural tissue and changes to cognitive output remains unknown, while mounting evidence from comparative neuroscience strongly suggests a far more complex relationship between cognition and neuroanatomy than the metabolic costliness argument presupposes (Chitkka and Niven 2009; Němec 2016; Olkowicz et al. 2016). Third, the argument ignores developmental constraints that may leave more complex cognitive mechanisms as the only viable options. These objections are not exhaustive, but they illustrate that when it comes to interpreting cognitive evolution, all else is indeed far from equal. With the metabolic costliness argument, as with many similar evolvability arguments, once the “ceteris paribus” clause is unpacked, it rapidly becomes clear that the claim that evolution favors cognitive simplicity holds for only a narrow range of cases. Much more may be said about the notion that simplest mechanisms are the most likely ones to underwrite a given behavior, but in the interest of space I will proceed on the assumption that the evidence is at present equivocal.
If simplicity is such a vexed concept, one might understandably wonder how it can have any effect at all on scientific practice, much less one so significant as a shift in the burden of proof. One possibility is that the intuitive appeal of simple explanations and their promise of lightening the cognitive load encourage the practice of supposing appealing explanations to be simple, even when it is not simplicity that makes them appealing. In fact, as Elliot Sober (1998) convincingly argues, this very appeal is often due to the some (real or imagined) background epistemic virtue on which “simplicity” is parasitic (Sober 1998; Fitzpatrick 2009; Fitzpatrick 2015). In the case of comparative cognition, this apparent virtue is the careful avoidance of “over-attributing” sophisticated and, it is often thought, “human-like” cognition to nonhuman animals. This caution may be traced back to a misinterpretation of C. Lloyd Morgan’s famous “canon”5 – a single phrase that comparative cognition researchers have adopted as central to the project of comparative cognition. Yet Morgan himself notes that simplicity often favors “anthropomorphism,” since it is simpler to assume that we are just like other animals than to imagine how we might be different (Thomas 2006). Thus, even on Morgan’s view, far from protecting against anthropomorphism, simplicity encourages it.
When complexity is understood as a sign of anthropomorphism, the project of simplifying begins to take shape. Thus, in comparative cognition, certain explanations and mechanisms are commonly presumed to be simpler – as a rule of thumb, these have traditionally tended to be the mechanisms not employed by humans (although researchers are increasingly questioning the sophistication of human cognitive mechanisms as well). These include “lower-order” explanations, which postulate fewer entities or processes or representational levels than putatively more complex, “higher order” abilities such as metacognition, theory of mind, planning, and more. While it is unclear that these processes and abilities are indeed simpler, the simplicity heuristic may be, and often is, applied without such justification. For instance, comparative cognition researchers commonly assume that associative mechanisms are simpler than representational ones, such as metacognition and mindreading.6 With these labels in place, applying the simplicity heuristic yields a clear rule: prefer associative accounts over representational accounts, barring compelling evidence to the contrary. This rule shifts the burden of proof on to putatively more complex explanations of animal behavior, resulting in a systematic “under-attribution” of complex cognition to nonhuman animals (Meketa 2014; Mikhalevich 2015).
Earlier critiques of the simplicity heuristic have focused on its effects on the development and interpretation of traditional experiments in comparative cognition. But the simplicity heuristic does more than shape the design of experiments and adjudicate among competing hypotheses. It also affects novel techniques in experimental comparative cognition that the earlier critiques did not discuss, such as the design and integration of cognitive models into the broader experimental approach within the study of animal minds. These models are “mathematical or logical structures … whose terms are specified precisely enough to make testable predictions about cognitive capacities such as perception, categorization, memory, learning, and decision making” (Allen 2014: 84). Since they are neither mere explanations nor freestanding experiments, they play a unique role in the experimental methodology of animal cognition, and they are affected by the simplicity heuristic in ways similar to but distinct from traditional experimental methodology.
The introduction of cognitive modeling has energized comparative psychologists and philosophers, who are encouraged by the possibility of converting vague qualitative hypotheses about the mind, typically couched in natural language or folk psychological terminology, into quantitative cognitive models concretized in a formal language. Colin Allen, for example, is optimistic that these novel tools promise “better experiments, better predictions, and better science” (Allen 2014: 93). Allen envisions a science of animal minds that fully utilizes the modeling techniques now common in cognitive science, wherein cognitive models and traditional (material) experiments challenge and refine one another in an iterative fashion.7
Allen’s characterization nicely captures the goings-on in comparative cognition today and is properly optimistic. What it does not (and does not purport to) offer, however, are the conditions of productive or fruitful iteration. For instance, it does not specify what justifies the refinement of experiments in any given case in response to model-based findings: is it the model’s usefulness in predicting behavior, in explaining the behavior of a wider array of organisms, or in describing – at some grain of resolution – the actual mechanisms underpinning the observed behavior?
This is where the simplicity heuristic gains some traction: it offers a clear decision-procedure for which models to develop and which experiments to pursue. It does so by directing scientists to develop and refine models of the simplest cognitive mechanisms (those operating with the fewest rules or perhaps the fewest causal factors) while at the same time directing experimenters to design experiments so as to rule out explanations based on these simple models. Because associative mechanisms are presumed to be simper than alternatives, modelers invest their time in refining these models over possible alternatives. As a result, associative models have proliferated. While associative models are not the only kinds of model favored by the simplicity heuristic, they are the most common since they serve as all-purpose alternatives to numerous putatively sophisticated explanations of behavior, such as metacognition, mindreading, planning, the existence of mental maps, and much more. In the interest of space, I will consider just one example in some detail: the case of metacognition in rats.8
Jonathon Crystal and Allison Foote conducted a series of experiments to test for the presence of one type of metacognitive ability among rats: the ability to monitor their own uncertainty. The “uncertainty-monitoring” paradigm tasks animals with classifying a stimulus into one of two categories, with the option to decline to answer and move on. The stimuli can be visual arrays of different densities, audio tones of different duration, and even smells of varying concentration. The test assumes that animals who preferentially decline the more difficult trials, where the stimulus is ambiguous, must do so because they are capable of monitoring their degree of certainty about the correct choice. Foote and Crystal employed this paradigm with rats, publishing the positive results first in 2007. When given the choice to decline tests, their rats consistently opted to decline the ambiguous (“harder”) tests but not the unambiguous tests. Moreover, the overall accuracy improved when rats were allowed to opt out of difficult tests. Foote and Crystal (2007) concluded that the rats’ behavior demonstrated awareness of their metacognitive uncertainty.
Shortly after the publication of this study, another group of researchers – Smith et al. – offered a competing associative “response-strength” model to account for the rats’ behavior. This model introduces a “low level, flat threshold for the decline response,” which activates whenever the response strengths for either SHORT or LONG are lower than the threshold (Foote and Crystal 2012: 188). The model maps response strengths as exponential curves declining from their thresholds, and crossing in the middle, where both strengths are low. Thus, the rats’ consistent preferences for the DECLINE option during the most difficult trials is explained by the activation of both LONG and SHORT dropping below the threshold for the DECLINE option. Since DECLINE was more strongly activated than either LONG or SHORT, the model predicts that the rats will activate the button that allowed them to decline the test and move on to more tests with the possibility of future rewards – mimicking the behavior of the rats in Crystal and Foote’s experiment.
In response to Smith et al.’s model, Crystal and Foote (2009) withdrew their metacognitive interpretation, reasoning that because the rat’s behavior could be explained in terms of “simpler” associative processes relying on “primary representations,” their previous metacognitive explanation was unwarranted (Crystal and Foote 2009). The introduction of putatively simpler models, they wrote, “necessitates the development of new, innovative methods for metacognition,” something that they tried to accomplish in subsequent experiments (Foote and Crystal 2012). Having thus accepted the simplicity heuristic (on which the metacognitive or “higher-order” representational hypothesis carries the burden of proof), Foote and Crystal (2012) revised their experiments in order to accommodate the simple model.
This example illustrates how the heuristic allows associative models to shape the arc of the experimental research programs on metacognition. Without the simplicity heuristic, metacognition might have been on equal epistemic footing with alternatives, and cognitive modelers such as Smith et al. (2008) would have needed to produce a model that was in some way superior to the metacognitive explanation in terms of, e.g., predictive power. Without this ability, the model would have been unable to dislodge the metacognitive explanation, though it might rightly have slightly decreased credence in the metacognitive hypothesis.
As this example illustrates, in challenging inferences from physical experiments, these models are shaping the setup of future physical experiments. Smith et al. (2008) are explicit on this point, writing in their introduction that their “models and related discussion have utility for metacognition researchers and theorists broadly, because they specify the experimental operations that will best indicate a metacognitive capacity in humans or animals by eliminating alternative behavioral accounts.” In this way, they indeed contribute to the iterative process of model-experiment “refinement,” but this interaction is unproductive, and perhaps harmful to scientific progress, as long as the models are chosen over qualitative explanations on the grounds of their putative simplicity rather than, e.g., explanatory scope or predictive power. As long as the simplicity heuristic places the burden of proof onto the more complex model, alternative cognitive models that are not (or are perceived as not being) as simple are less likely to be developed: research is time-consuming, and when a research program is unlikely to pay off in terms of publications, funding, or other career-enhancing rewards, researchers will be rational to avoid investing their time in programs that their field deems unlikely to yield fruitful results.9 Yet without this time and effort, the simpler models will remain unchallenged and will continue to be regarded as the theoretical defaults.
The metacognition example is not an outlier. Consider a few more examples of the heuristic at work in model-building, in which “simplicity” is attributed not only to associative models, but also to models that postulate few variables, rules, or causes. Each of these is expressly non-representational and regarded as simple in that respect. For example, van der Vaart et al. (2012) propose a model, based on a single rule (“cache more when stressed”), to explain the observation that scrub jays re-cache more often when their initial caching had been observed by others (Dally et al. 2006). This behavior was interpreted as evidence that these corvids possess some mindreading abilities – the ability to take the visual perspective of another, or perhaps to impute to others the intention to pilfer. Proposing a similarly “simpler” model of nonhuman primate reconciliation behavior, van der Vaart and Hemelrijk (2014) reason that because “the same patterns arise as a consequence of simple rules about fighting and grooming, and their effects on spatial proximity [as developed in their model]” it follows that the behavior is “not necessarily the product of sophisticated social reasoning” (348). The implication is clear: an alternative model poses a legitimate challenge to a dominant interpretation of behavior just in case it is putatively simpler.
The presumption that such putatively complex explanations, such as “social reasoning,” comes through in the common refrain that these explanations are rendered unnecessary whenever simpler models become available. Consider another example. Cruse and Wehner (2011) propose a cognitive model of ant and bee navigation, which is intended to displace explanations adverting to “cognitive maps” – representations of an animal’s environment that organisms may use to navigate. On this model, information from peripheral systems acts on disparate parts of the organism to produce behavior that’s only apparently coordinated. It thus bypasses the need for a central processor required to collate information into a single representation – a feature that the authors regard as an extravagant hypothesis. They conclude that since their model is purportedly simpler than the cognitive map model, there is “no need” for a cognitive map. Similarly, Puga-Gonzalez et al. (2009) argue that “the use of grooming [among primates] as a ‘currency of exchange’ is dangerously anthropomorphic according to us and others” and that “often simple rules suffice to cause many of the observed patterns and herewith an integrative theory,” concluding that “fewer cognitive processes may suffice as shown for instance in a model for dominance style” (2; emphasis added).
But note that a claim that a putatively more complex account is not needed or that a simple one may “suffice” is a claim about the proper direction of the research program. It is a claim, guided by the simplicity heuristic, that advises against developing alternative explanations of putatively complex mechanisms – or of developing quantitative models that may account for the behavior as well as or better than the simple models. Instead, the heuristic suggests that experiments must be redesigned in order to rule out the inference from the simple model, and it repeats this request each time a “simple” model challenges the experimental finding.
As long as the simplicity heuristic directs the development of application of cognitive models, it will discourage the development of alternatives. This appears to be the situation in comparative cognition today, where models of simple mechanisms far outnumber models of putatively complex mechanisms. The result of this heuristic is cumulative: because the simplicity heuristic encourages researchers to channel their efforts into refining “simpler” and allegedly non-anthropomorphic models, these models gain an unearned competitive advantage over alternatives that are deemed “complex.” As a result, these alternatives are more likely to remain qualitative, which further diminishes their apparent value in the eyes of researchers, who would be rational to view a qualitative explanation with suspicion when a quantitative explanation is on offer. Thus, without understanding the history of the models, it may appear that they have outcompeted more complex alternatives on their epistemic merit. Yet, as I suggest, their staying power may be due to having benefitted from a biasing heuristic that funneled more intellectual capital into their development while simultaneously discouraging the proliferation of viable (quantitative) alternatives.
One may object that simple models are defaults because simplicity is a virtue insofar as simple models may be more predictive as well as easier to use than complex models.10 I grant that the predictive power of a model is a virtue that justifies the further exploration of that model. However, predictiveness does not necessarily track accuracy, and simplicity is not a virtue if the models are intended to explain animal behavior by describing actual cognitive processes or mechanisms. There is no reason to suppose a priori that the target mechanism described by the model is in fact simple.11 Thus, simple models ought not to be defaults just as long as these models are intended to describe the cognitive system.
Another objection holds that simple models of animal minds are well developed not because they are favored by the simplicity heuristic, but because alternatives such as metacognition are so vague as to be impossible to model quantitatively. However, if metacognitive hypotheses appear vague, this may be because fewer attempts at quantifying these models have been made. This objection may prove true of some putatively complex cognitive explanations of animal behavior, but the best means of putting this objection to the test is by attempting to formulate alternative models of, e.g., metacognition.
Finally, one may grant both that simple models receive more investment and that they do so for a poor reason (simplicity), but nevertheless maintain that the iterative process of model-experiment development is not sensitive to initial conditions – that eventually, the models and experiment will converge on the truth. In fact, there is some reason to believe that simple associative models have become more complex by having to accommodate more experimental information and, as a result, have come to resemble cognitive models of more sophisticated cognition. However, even if this turns out to be true for association, this need not be so in other cases. The optimism must be grounded in something like the view that no matter what the starting point, scientific research programs (or the research programs of comparative cognition specifically) are virtually guaranteed to ultimately arrive at the best (most accurate, most useful, etc.) answer. Thus far, however, no adequate justification for the assumption that experimental research will asymptotically approach the best explanation has been given. In the absence of persuasive reasons to retain the simplicity heuristic, and given its epistemically undesirable consequences, there remains little reason to continue to deploy it in the study of animal minds.
Suppose that we eliminate simplicity as a guide for shaping the arc of model-experiment development. Without simplicity acting as a gatekeeper, what is to prevent an influx of empirically adequate and predictive models that presuppose metaphysically suspect entities, such as chakras and spirits? Should scientists open up the borders to every conceivable model, including those that seek to explain animal behavior in terms of such entities? At least simplicity permits the modeler to claim that we do not “need” such entities in our explanations. Without simplicity, how do we discourage such promiscuous proliferation of models, and, relatedly, how do we adjudicate at all among empirically adequate models?
These are important concerns, and while space does not permit a full answer to the second query, I believe that we have reason for optimism with respect to the first. Reasons to reject these metaphysically suspect entities go beyond the desire for clean ontologies. These are the same reasons that generate the suspicion in the first place: namely, that most of these entities do not fit into a naturalistic view of the world on which science is premised. Admittedly, not all such unwelcome entities are metaphysically suspect – for instance, interventions by aliens or global government conspiracies – but the sheer improbability of these explanations is enough to guard cognitive models from their influence. We have excellent reasons to restrict model-building to the range of entities and causes in an epistemically responsible fashion, and these reasons stem from background theoretical assumptions about the kinds of entities that populate our world. These theories tell us that metacognition is real but chakras are probably not, and for this reason to permit models describing metacognition but not those describing chakras.
Simplicity, therefore, is not a unique bulwark against an anything-goes science of metaphysical extravagance. Rather, it is an occasionally useful but frequently detrimental guide that risks closing off potentially fruitful research programs. Instead of the simplicity heuristic, comparative cognition flourishes when all epistemically equal models are encouraged equally, rather than being ruled out without a fair trial or being saddled with an impossible burden of proof, in the courtroom of scientific ideas. Such model diversity would put a new and productive pressure on experimenters, encouraging them to devise experiments not to rule out “simpler” alternatives, but to identify novel means of experimentally adjudicating among competing models of animal mind.
1 I am grateful to participants of the Washington University in St. Louis Philosophy of Science Reading Group for helpful comments on an earlier version of this essay, and to Russell Powell for his thorough reading and thoughtful suggestions.
2 See Call and Tomasello (2006) and Povinelli and Vonk (2003) for examples of appeals to simplicity. For recent discussions of the chimpanzee mindreading controversy, see Chapters 22, 21, and 13 in this volume by Halina, Lurz, and Proust, respectively.
3 My “simplicity heuristic” roughly resembles what Fitzpatrick (Chapter 42 in this volume) calls the “Conservative Canon.”
4 To say that the reasons are not intuitive is not to say that they are necessarily mistaken.
5 For a thorough review of the multiplicity of meanings of Morgan’s Canon, see Thomas 2006. For a historical analysis, see Kimler 2000; see Burkhardt 2005 for a detailed history of the study of animal behavior which places the canon into proper context. For examples of explicit appeals to the canon, see Kennedy 1992, Wynne 2007a, and Wynne 2007b. For analyses and critiques of specific interpretations, see Fitzpatrick 2008; Meketa 2014; Allen-Hermanson 2005; and Sober 1998 and 2005. For arguments against any plausible interpretation of the canon, see Starzak 2016 and Fitzpatrick (Chapter 42 in this volume). See Dacey (2016) for a limited defense of some forms of the canon.
6 This assumption is problematic for two reasons. First, it is unclear that association is simpler than metacognition, since the ontology of metacognition and association are under-specified. For example, see Heyes 2012 and Gallistel 2000. Second, even if association is simpler than metacognition, it is far from clear that simplicity should play any role in adjudicating among hypotheses.
7 See Buckner 2011 for a discussion of such iterative model-experiment development.
8 This trend is mirrored in developmental psychology as well. For instance, associative explanations for language acquisition in human infants continue to be preferred despite their explanatory and predictive failures. Their staying power may be attributed not to an epistemic virtue, but to something extra-epistemic, such as theoretical simplicity. I thank Richard Moore for this point.
9 This is not to say that no research will be conducted without substantial rewards, much less to suggest that scientists are driven exclusively or predominantly by career-enhancing goals. It is, however, to acknowledge a sociological constraint on the development of intellectual tools and the cascading effects of these social dimensions of science.
10 See, for instance, Sober 2009.
11 One might argue that my objection rules out all models since all models are idealizations. However, there is a difference between simplicity and idealization. All models are idealized, including metacognitive models. Yet metacognitive models would be considered more “complex” by the simplicity heuristic because they are representational or because they contain more elements. Thus, it is not the fact that putatively simple models are more idealized that makes them ineligible as defaults, but, rather, that there is no a priori reason to prefer them over putatively more complex models.
For additional readings on simplicity in biology, psychology, and philosophy, see Elliott Sober’s Ockham’s Razors: A User’s Manual, Cambridge: Cambridge University Press, 2015.
Allen, C. (2014) “Models, Mechanisms, and Animal Minds,” The Southern Journal of Philosophy 52: 75–97.
Allen-Hermanson, S. (2005) “Morgan’s Canon revisited,” Philosophy of Science 72(4): 608–631.
Buckner, C. (2011) “Two Approaches to the Distinction Between Cognition and ‘Mere Association,’” International Journal of Comparative Psychology 24: 314–348.
Burkhardt, R. W. (2005) Patterns of Behavior: Konrad Lorenz, Niko Tinbergen, and the Founding of Ethology. Chicago: University of Chicago Press.
Chitkka, L., and Niven, J. (2009) “Are Bigger Brains Better?” Current Biology 19: R995–R1008.
Cruse, H., and Wehner, R. (2011) “No Need for a Cognitive Map: Decentralized Memory for Insect Navigation,” PLoS Computational Biology 7: e1002009.
Crystal, J. D., and Foote, A. L. (2009) “Metacognition in Animals,” Comparative Cognition & Behavior Reviews 4: 1–16.
Dacey, M. (2016) “The Varieties of Parsimony in Psychology,” Mind & Language 31: 414–437.
Dally, J. M., Emery, N. J., and Clayton, N. S. (2006) “Food-Caching Western Scrub-Jays Keep Track of Who Was Watching When,” Science 312: 1662–1665.
Fitzpatrick, S. (2008) “Doing away with Morgan’s Canon,” Mind & Language 23(2): 224–246.
——— (2009) “The Primate Mindreading Controversy: A Case Study in Simplicity and Methodology in Animal Psychology,” in R. Lurz (ed.) The Philosophy of Animal Minds, New York: Cambridge University Press.
——— (2015) “Nativism, Empiricism, and Ockham’s Razor,” Erkenntnis 80: 895–922.
Foote, A. L., and Crystal, J. D. (2007) “Metacognition in the Rat,” Current Biology 17: 551–555.
——— (2012) “‘Play It Again’: A New Method for Testing Metacognition in Animals,” Animal Cognition 15: 187–199.
Gallistel, R. (2000) “The Replacement of General-Purpose Learning Models With Adaptively Specialized Learning Modules,” in M. Gazzaniga (ed.) The New Cognitive Neurosciences, Cambridge, MA: MIT Press.
Heyes, C. (2012) “Simple minds: A qualified defence of associative learning,” Philosophical Transactions of the Royal Society of London B: Biological Sciences 367(1603): 2695–2703.
Kennedy, J. S. (1992) The New Anthropomorphism, Cambridge: Cambridge University Press.
Kimler, W. C. (2000) “Reading Morgan’s Canon: Reduction and Unification in the Forging of a Science of the Mind,” American Zoologist 40: 853–861.
Meketa, I. (2014) “A Critique of the Principle of Cognitive Simplicity in Comparative Cognition,” Biology & Philosophy 29: 731–745.
Mikhalevich, I. (2015) “Experiment and Animal Minds: Why the Choice of the Null Hypothesis Matters,” Philosophy of Science 82: 1059–1069.
Olkowicz S., Kocourek, M. Lučan, R. K., Porteš, M., Fitch, W. T., Herculano-Houzel, S. and Němec, P. (2016) “Birds Have Primate-Like Numbers of Neurons in the Forebrain,” PNAS 113: 7255–7260.
Povinelli, D., and Vonk, J. (2003) “Chimpanzee Minds: Suspiciously Human?” Trends in Cognitive Sciences 7: 157–160.
Puga-Gonzalez, I., Hildenbrandt, H., and Hemelrijk, C. K. (2009) “Emergent Patterns of Social Affiliation in Primates: A Model,” PLoS Computational Biology 5(12): e1000630.
Shettleworth, S. J. (2012) “Modularity, Comparative Cognition and Human Uniqueness,” Philosophical Transactions of the Royal Society B 367: 2794–2802.
Smith, J. D., Beran, M. J. Couchman, J. J., and Coutinho, M. V. (2008) “The Comparative Study of Metacognition: Sharper Paradigms, Safer Inferences,” Psychonomic Bulletin & Review 15: 679–691.
Sober, E. (1998) “Morgan’s Canon,” in C. Allen and D. Cummins (eds.) The Evolution of Mind, Oxford: Oxford University Press. 224–242.
——— (2005) “Comparative Psychology Meets Evolutionary Biology: Morgan’s Canon and Cladistic Parsimony,” in L. Datson and G. Mitman (eds.) Thinking With Animals: New Perspectives on Anthropomorphism, New York: Columbia University Press, 85–99.
——— (2009) “Parsimony and Models of Animal Minds,” in R. W. Lurz (ed.) The Philosophy of Animal Minds, New York: Cambridge University Press, 237–257.
Thomas, R. K. (2006) “Lloyd Morgan’s Canon: A History of Misrepresentation,” History & Theory of Psychology Eprint Archive. Available at http://archive.is/kQasr and by the author at https://faculty.franklin.uga.edu/rkthomas/sites/faculty.franklin.uga.edu.rkthomas/files/MCPrintOptimal.pdf
Tomasello, M., and Call, J. (2006) “Do Chimpanzees Know What Others See – Or Only What They Are Looking At?” in S. Hurley and M. Nudds (eds.) Rational Animals? New York: Oxford University Press.
Starzak, T. (2016) “Interpretations Without Justification: A General Argument Against Morgan’s Canon,” Synthese: 1–21. doi: 10.1007/s11229-016-1013-4
van der Vaart, E., Verbrugge, R., and Hemelrijk, C. K. (2012) “Corvid Re-Caching Without ‘Theory of Mind’: A Model,” PLoS One 7: e32904.
Wynne, C. D. (2007a) “Anthropomorphism and its Discontents,” Comparative Cognition & Behavior Reviews 2: 151–154.
Wynne, C. D. (2007b) “What Are Animals? Why Anthropomorphism Is Still Not a Scientific Approach to Behavior,” Comparative Cognition & Behavior Reviews, 2: 125–135.