Humans possess cognitive and behavioural skills that remain off-limits even to the smartest of other animals, even where the latter have been subject to lengthy and highly artificial training regimes—in the extreme, full-blown human enculturation, as has been done with some non-human great apes (for example, Savage-Rumbaugh [1986]). What is the best way to conceptualize this difference between humans and other animals? What ultimately explains this difference? How did this difference become established over evolutionary time? In The Human Evolutionary Transition, Enquist et al. take up these fundamental questions of human and animal evolution, pursuing a track that is at once bold, innovative, comprehensive, and rigorous. The book has sixteen chapters and roughly divides into two parts: the first eight cover the notion of behavioural and mental sequences, and how animal behaviour and cognition can be usefully modelled in such terms. The second part then tackles human sequence learning and how it differs from that of animals. In line with many other recent approaches to explaining human cognitive and social evolution, Enquist et al. give a place of prominence to cumulative cultural learning, but their focus on sequences provides for some unique twists on this common line of thought. In this review, we summarize what we take to be the book’s main themes before lodging some of our reservations.
The first theme is that the essence of adaptive behavioural complexity lies in finding productive sequences. To survive and reproduce, animals must recognize and capitalize on the adaptive opportunities in their environments. In general, this requires that they implement behavioural sequences of various kinds. To take an extremely simple example: an animal must approach and then eat the food that it sees. But perhaps this food must first be processed in some way before it can be eaten. Then a more complex sequence is needed: the animal must approach the food, then (for example) peck it, then (for example) lick it. Enquist et al. point out that learning adaptive sequences quickly poses a combinatorial explosion, so that the demand on learning grows exponentially with sequence length, all other things being equal. More precisely, to discover a particular adaptive sequence of length l, given a repertoire of m actions, each of which the animal is equally likely to try out, the animal may have to make as many as ml attempts. So, assuming a repertoire of just ten acts, the above animal may have to make as many as 103 = 1000 attempts to find even this simple approach–peck–lick sequence! This combinatorial dilemma strongly constrains the number and complexity of sequences animals can discover on realistic time and energy budgets.
The second theme concerns the underappreciated power of associative learning. The animal world is by no means devoid of behavioural sequences. Indeed, several animals produce quite long behaviour sequences, for example, ones involving the manufacture and use of tools (for example, chimpanzee honey extraction; Fay and Carroll [1994]). How do animals overcome the above combinatorial dilemma? Many—perhaps most—animal psychologists today would have us think it’s in virtue of these animals possessing low-key versions of human-like abilities for learning, such as reasoning, planning, episodic memory, and/or cultural learning. Enquist et al. take a very different line. Boldly, they claim that associationism very likely suffices to explain all of the sequences learned by animals. But: this ain’t your good ol’ fashioned associationism, crucially. It’s associative learning, typically supplemented by domain-specific genetic information biasing learning in various ways (for example, by predisposing the animal to try certain behaviours and/or value certain stimuli), often coupled with a friendly learning environment.
Enquist et al.’s treatment of the learning environment is especially illuminating. In particular, they focus on what they helpfully call favourable entry and exit patterns associated with sequences. A sequence has a favourable entry pattern when it affords learning from a penultimate (or at least intermediate) step backwards to the start of the sequence. Imagine a chimpanzee finding a termite mound with an opening (for example, during termite spawning); this animal need only ‘eat the termites’ in order to come to positively value—via operant conditioning—the stimulus ‘opening in termite mound’. The animal’s task then becomes to learn how to open termite mounds (for example, by targeting damp areas on the mound; Brewer [1978]). A behaviour resulting in such a stimulus will be reinforced due to the value the animal has already assigned to ‘opening in termite mound’. The exit conditions of a sequence may also be favourable in the sense that a mistake or failure doesn’t lead one away from the reward state; an attempt to insert a small stick into an opening of a termite mound does not seal the mound shut again. Favourable learning patterns can exist for purely physical reasons, but often they have social origins. Others may act to bring about these conditions, as in cases of teaching (for example, provisioning young meerkats with stingless scorpions to tackle; Thornton and McAuliffe [2006]), or such conditions may arise simply as a by-product of others’ utilitarian behaviour (for example, the termite fishing behaviour of others in the service of feeding themselves). Enquist et al. make a forceful case that animals can get a lot out of such an associative learning setup, and even present a detailed but accessible computational model of such learning—the A-learning model.
A third theme revolves around the idea that thought and thinking is unique to humans. Enquist et al.’s genetically souped-up and ecologically embedded associationism allows them to credibly advance a rather incredible hypothesis—namely, animals do not really think; only humans do. Of course, they have something quite specific in mind by ‘thinking’ here. They define thinking as, ‘generalized gathering of causal information, recombination of causal information’, and/or ‘time-consuming decision making’ (p. 122). Interestingly, they argue that the thinking gap between humans and animals is ultimately rooted in costs.
More fully, whereas associative cognition simply involves activating the behaviour with the highest estimated value, given the stimulus the animal is currently experiencing, bona fide thinking involves the construction of a mental sequence of intermediate memory states and computational procedures generated using a mental model of some kind. Mental models are understood as organized mental representations of some domain, which, among other things, support the drawing of novel inferences. (A paradigm case of a mental model would be a cognitive map in the classical ethological sense.) To build a rich mental model, the animal must pick up and store in a computationally accessible format information for which it perceives no immediate value. Moreover—lest one’s model fall into disrepair—the animal needs to update its model as new information comes in. Mental models solve the twin combinatorial problems implied by behavioural and mental flexibility, that is, of finding productive behavioural and mental sequences, on realistic time and energy budgets. However, mental models are costly to learn, costly to maintain, and costly to use. In addition, inaccurate models can produce costly mistakes. Given such costs, and given the surprising power of associative learning to find productive sequences, Enquist et al. claim that, from an evolutionary point of view, their flavour of associative learning often yields a greater net benefit than thinking. This, they propose, explains the (supposed) situation with animal cognition.
The fourth and final key idea is that virtually all distinctively human mental skills are culturally evolved. If building, maintaining, and using mental models to fuel thinking frequently carries prohibitively high costs, then how did humans end up such powerful and diverse thinkers? Enquist et al.’s answer lies in the special capacities humans show for cumulative culture. They acknowledge, of course, that some biological differences must exist between us and our great ape relatives—it cannot be cultural evolution all the way down, so to speak, else we would not expect pronounced differences between humans and human-enculturated great apes (of which there are many). However, Enquist et al. hypothesize that these biological differences are in fact quite modest. More specifically, they posit the genetic evolution of small, domain-general cognitive adaptations that enabled our ancestors to better ‘represent sequential information and to learn new ways to process information’ (p. 142). These changes are envisaged as having occurred against a backdrop of even slower development than is characteristic of other apes, combined with increased (allo)parental investment, allowing for extended learning periods in an enriched social-learning world. The idea is that in such an environment the posited genetic adaptations for enhanced sequence representation and mental flexibility would have earned their keep from the start. The result of all this (supposedly) was that humans discovered more and better behavioural but also mental sequences that rewarded copying, sequences that were then culturally transmitted and ratcheted-up over time. This included, crucially, the very mental skills needed to profitably build and use mental models. Thus, in line with an increasing number of other theorists in this area (especially Cecilia Heyes; see, for example, Heyes [2018]), Enquist et al. propose that cultural evolution is largely behind both the contents and cognitive mechanisms of distinctively human cognition.
In our view, there is much to like about this package. In particular, the sequence-centred approach—which has not been developed anywhere near as thoroughly by anyone else to our knowledge—strikes us as highly useful. Sequences may well provide a sorely needed, intuitive, and mathematically tractable common denominator for comparing animal and human types of cognition and culture, especially in terms of costs. The costs of cognitive mechanisms and the mental sequences based on them have been largely neglected by theorists—to the detriment of the field, we think. This is understandable, given that there’s still so much we don’t understand about how computation is implemented in the brain. It is a highly attractive feature of Enquist et al.’s approach that it allows us to at least partially side-step this issue. In short, then: everyone interested in the cognitive and behavioural evolution of humans and animals can benefit from reading this book! That said, plenty of questions remain. We finish by highlighting a few, dividing them into proximal and ultimate issues.
Proximal first. The majority of the book deals with the proximate causes of human and animal behaviour. Much of this material is persuasive, though some cases of animal learning do not receive the full attention they deserve. A prime example is animal song learning, which is mentioned only in passing. For example, there is evidence that some untrained, unenculturated members of certain bird species spontaneously copy acoustic sequences that they would not have produced otherwise (Gardner et al. [2005]). If so, then these cases involve copying-dependent know-how; the know-how—the linear order, in this case—must be copied (Tennie et al. [2020]; Tennie and Call [2023]). We mention this here because such cases bear on Enquist et al.’s claim that the human capacity for ‘genuine imitation’ is culturally evolved—a ‘gadget’ in Heyes’s sense—as opposed to being (at least partly) biologically motivated (as is presumably the case in the relevant bird species, given that the tested birds were untrained and unenculturated).
Another important omission is primate social cognition. Specifically, we have in mind the results of Cheney and Seyfarth’s ([2007]) many ingenious ‘playback’ experiments with baboons. These experiments strongly suggest that baboons and presumably many other primates have rich mental models of their social environments that they use to, among other things, interpret vocal sequences, including entirely novel sequences. These models combine different types of social information—for example, both kin and dominance information—into a single, cohesive mental representation that is continually updated as the social environment changes, allowing baboons to make fine-grained discriminations among perceptually similar stimuli, such as between a call sequence indicating a within-family feud (less consequential) and one indicating a between-family feud (more consequential). Indeed, Cheney and Seyfarth persuasively argue that baboons’ social representations exhibit a hierarchical or nested part–whole structure akin to that which is thought to underlie mental representations of natural language sentences in humans. (We note that discussion of how animals learn hierarchical representations—as opposed to linear sequences—is all but absent from Enquist et al.’s book; on learning hierarchies, see, for example, Byrne and Russon [1998]; Planer [2023a].) To the extent that this picture is even roughly accurate, primate social inference holds obvious relevance for questions about the origins of human cognition and communication—and hence human culture—given humans’ primate heritage (Cheney and Seyfarth [2007]; Planer [2021]).
A final proximal issue. In developing their brand of associationism, Enquist et al. make use of rather abstract categories to describe behaviour and stimuli (a point they are up-front about; see, for example, p. 29). Abstractness per se is not a problem. But the worry is that in some cases Enquist et al. may be letting in the very complex computational capacities they hope to eschew by using such high-level representations. The issue is a familiar one for associationism, expressed both by some of its most ardent supporters (for example, Dickinson [1987]) and critics (for example, Pylyshyn [1984]). The primate social domain is again instructive here. A strong predictive relation exists, no doubt, between being threatened by a dominant monkey and being attacked. This is an associative bond worth making! However, as Cheney and Seyfarth have shown, whether a particular vocalization is interpreted as a threat towards oneself in the first place can depend on factors such as the recent history of social interactions between the signaller and the receiver, and even on the recent history of social interactions between the signaller’s kin and the receiver. To the extent that categorization is informed by such background information, and to the extent that such information is retrieved or computed from an organized mental model of the social environment, then the act of categorization is far from computationally innocent.
As for ultimate causation: The book only briefly discusses such factors. The main proposal has already been mentioned, namely, that it was prolonged development in an environment rich in social-learning opportunities that allowed human cumulative culture to take off. We have no doubt that changes in human life-history and sociality were crucial drivers of human cultural evolution. However, there are questions of uniqueness—why only us, given that such conditions ring true of some other species, including some primates?—as well as of timing. Here, we’ll limit ourselves to just the latter.
The lifeway changes Enquist et al. point to are typically linked to the establishment of cooperative breeding in hominins (although they do not personally make this link). On one highly influential line of thought, increased allomaternal investment was fundamental to the increase in hominin childhood and adolescence (for example, Hrdy [2009]; van Schaik and Burkart [2010]). At the same time, such investment leads to a marked increase in social complexity (Planer [2023b]). Most obviously, hominin young interact with a larger number of caregivers; less obviously, cooperative breeding drives down the inter-birth interval in ways that lead to more (half-)siblings, spaced closer in age, in one’s residential group. More carers also means a safer learning environment. The problem is that cooperative breeding is widely thought to have established in our line around the time of Homo erectus, and possibly even before. Evidence for this comes from a variety of changes in hominin brains, bodies, and environments around this time (for overviews, see, for example, Hrdy [2009]; DeSilva [2021]). And yet, there are few or no signs of unambiguous cumulative culture take-off until around 800 kya or even later (Paige and Perreault [2024]; Tennie [forthcoming]; Planer et al. [forthcoming]). So, there is a puzzling time lag—perhaps on the order of a million years—between the establishment of the conditions that drove the evolution of cumulative culture, according to Enquist et al., and actual signatures of such culture in the archaeological record.
Our aim in this review has been twofold: first, to communicate the impressive scope and novelty of Enquist et al.’s recent book and, second, to illustrate the book’s promise to move the field forward—in part, by stimulating critical challenges to its bolder claims from researchers such as ourselves. But more generally, we hope to have made clear that The Human Evolutionary Transition is a highly valuable contribution to the field that will reward careful study.
https://i0.wp.com/www.thebsps.org/wp-content/uploads/2024/10/Thumbnail-Brown-on-Anderson-and-Mitchel.png?fit=398%2C280&ssl=1
280
398
The BJPS
http://www.thebsps.org/wp-content/uploads/2015/06/BSPS-blue.gif
The BJPS2025-03-04 07:00:382025-03-04 20:38:47Andersen and Mitchell, The Pragmatist Challenge // Reviewed by Matthew J Brown
