The Human Condition

Site Tools

books:shape_of_thought

The Shape of Thought

How Mental Adaptations Evolve
H. Clark Barrett

failed to fetch data: unkown error

This book is a very important update to the theoretical foundations of Evolutionary Psychology. Although the author has tried to make it accessible to a general audience, this is a scholarly book, with references, footnotes and complex sentence structure. He does define technical terms from evolutionary biology and philosophy when they first appear, but by the end the accumulation of vocabulary is considerable. People who haven't read much about the abstract mechanisms of evolution may find some sections to be heavy going. Even so, there very good extended examples of particular proposed human mental adaptations, along with suggestions of how they could have evolved. Although he touches on many issues, the argument he returns to most is that the mind is “adaptations all the way up”. There is no part of the mind that has not been designed by evolution. If we do have general mental capabilities, then this is because those capabilities evolved in an environment (EEA) which was similar in relevant ways to the contemporary environmental variation which we flexibly respond to today.

Introduction

While it is true that processes such as adaptation occur over time, and therefore things like form-function fit evolve, the details of any given case depend on–will, the details of that case. So, not without a certain sense of irony, I'd like to give this idea a name: the first law of adaptationism. This law can be summarized as follows: it depends. In its long form, what this means is that the nature of any given adaptation–the form that it takes and the nature of its fit to the world–depend on how and why it got there. p8

Some critics of Evolutionary Psychology (such as Stephen Jay Gould, see Just-So Stories), have used “adaptationism” as a name for an error of thought in the field, but Barrett embraces the term.

This is the view that I will call “additivism.” You might be surprised that I feel the need to give it a name, or even to point it out, so obvious might it seem. However, the way that many people think about the role that natural selections plays in shaping cognitions suggests that the skill-adding role of selection is not obvious at all. I will contrast it with another view, which is widespread in psychology and the behavioral sciences more generally, that I will call “subtractivism.” This view is that the role of natural selection is fundamentally to constrain systems in useful ways: to narrow down the set of possibilities of what they could otherwise do. If you do a literature search, you won't find any psychologists explicitly calling themselves subtractivists. I made the term up. What I am trying to describe is a point of view that, while acknowledging that evolution is important in explaining mind design, views its role as mainly one of adding constraints to the otherwise general-purpose learning systems. For example, many neuroscientists and psychologists view brain development as resulting from the domain-general learning properties of neural networks filtered through biological constraints that point development in certain directions. p 19

This is an important support for his argument that our seemingly general mental capabilities are largely composed of many interlinked adaptations. There is a different emphasis on constraint in much discussion of the human meaning of Evolutionary Psychology, the idea that an evolved human nature places practical constraints on culture and social organization (as in The Tangled Wing and The Blank Slate). He never tells us what the thinks about that kind of constraint.

[Discussion of how bacteria evolve the ability to sense light so that they can move away from toxic levels of UV light.] The example of phototaxis illustrates a concept that will play a central role in this book: the concept of an inductive bet. There is a sense in which negatively phototactic bacteria appear to be betting that sunlight is bad, by systematically moving away from it. But there is no conscious thought behind the bet, or indeed any mind at all in the sense that we typically use the term. Instead, the bet is embodied in the design of the organism's adaptations, in the form of physical features that cause it to move away from light. In fact, this is a general feature–one of the few, as the first law of adaptationism suggests–of adaptations: They embody statistical “guesses” about what their environment is like, and by extension, what it will be like ten minutes from now, tomorrow, and the next day p 23

The idea that adaptations incorporate inductive bets is key to understanding Barrett's theory. The term ties into both the theory of Bayesian inference in statistics and Inductive reasoning in philosophy. Evolution implicitly assumes the future will resemble the past. While this is not always true, it's the only thing evolution has to go on, and much of the time we clever humans can't do any better (see Prediction is Intractable).

Evolved inductive bets are not guaranteed to work out, but rather, as bets that have survived the process of natural selection. In fact, the stochastic nature of the world means that even in the environment where the mechanism evolved–what evolutionary psychologists call its “environment of evolutionary adaptedness,” or EEA–it will be wrong, is guaranteed to be wrong some of the time. For example, some bacteria that are doing just what their opsins and flagella evolved to do will nevertheless die because of it. How often, when, and why are matters of the first law. p 25

On the view of domain specificity I'm advocating, then, all adaptations that have evolved to process information, from perceptual mechanisms to reasoning mechanisms to learning mechanisms, have a domain. This is true in at least two senses. Anthropologist Dan Sperber introduced the important distinction between and adaptation's proper domain–the set of conditions in which it was designed to operate, or the inputs it was designed to process–and its actual domain, the set of conditions in which it can, in fact, operate, whether designed to do so or not. […] By analogy, the proper domain of a hammer is nails, but it can be used to hammer things other than nails, and even to do things other than hammering (e.g., serving as a paperweight is within a hammer's actual domain). p 27

When an adaptation works in a new environment, we can say this is because the actual domain is larger than the proper domain.

Body development is highly dynamic, contingent, and context-sensitive. Genes don't merely “specify” outcomes in a one-to-one fashion; the outcomes they produce depend on where and when they are expressed and on prior developmental events. Nevertheless body development is best described as a series of construction events that are designed to hit particular targets. Not point outcomes, of course, but regions of possibility space where the design features emerge through interaction–as they do everywhere in development–but at the same time emerge this way because of a history of selection to do so. What I will suggest is that this is likely to be a good description of brain development as well. However, it will require us to take into account that the kinds of targets brain development is designed to hit are not limbs and digits, but information structures. Whatever you might think about the “initial state” of the cortex, it's clear that the developed cortex of the brain is highly domain-specific, rather than uniform. p 30

We could speak of just about any active process of causal guidance in terms of constraint if we wanted to; for example, we could say that a pilot flying from Los Angeles to New York “constrains” his plane to land at JFK airport via his actions. It would be a bit odd to use the work “constraint” in this way, of course […] Both these examples illustrate the glass-half-full/glass-half-empty nature of the language of constraint and causation. Causation involves both possibilities foreclosed, or what doesn't happen, and possibilities generated, or what does. In this sens, the causal properties of all material stuff in the world are therefore both constrainers and enablers, and in any given case, we can thing and talk about both aspects of causation. In other words, we can talk about what the causal properties of stuff enable, and what the rule out. […] It is this distinction that I have in mind when I talk about “additivisim” and “substractivisim”. p 31

Some have pointed to the inherently subtractive nature of natural selection–the fact that it acts only as a sieve, pruning the variation that is there–as evidence that it does not “specify” phenotypic outcomes and is not itself a generator of design. It's true that natural selection at any given time is only a pruner and not generative. But for the reasons outlined above, natural selection, acting again and again on variation produced over time, is generative. And the developmental systems that it produces are generative as well: Because of the ratcheting process of cumulative selection, developmental systems become targeted to hit more and more specific developmental outcomes. Adult organisms are not merely “constrained” zygotes. Instead, the processes that build adults from zygotes are active, targeting certain areas of design space that would be hard to reach by a guided random walk. […] Natural selection acting over time can figure out what regions of possibility space–the space of all possible developed phenotypes–that it's best to land in, and engineer developmental mechanisms designed to hit those regions ofg the space, which evolutionary psychologists John Tooby and Leda Cosmides call adaptive targets. p 35

Let's end this chapter with one more example: imitation. This is a cognitive ability that appears to be quite general purpose, and in a sense you can say that it is. But evidence suggests that humans are in many ways better imitators that other species, and it's likely that we possess either new cognitive mechanisms or modified verions of older ones that have added new abilities of imitation that other species don't possess. In particular, imitation illustrates the point that cognition is not simply a matter of passively perceiving information in the world. Perception is important, to be sure, but in order to successfully imitation, ou have to make inferences about what it is that you are perceiving (and perception itself is, of course, inferential). Faced with all the same information, different species do different things: Some don't imitate at all, and others vary in both how successful they are at it and what they appear to be attempting to do. p 41

Consider, for example, social learning of tool use in capuchin monkeys. Capuchins are quite clever foragers who use tools to process foods designed to thwart access, like hard nuts and seeds with painful spines, and who are known to transmit foraging skills socially in the wold. A spectacular example of this is nut-cracking by the species Cebus apella in Brazil, where the monkeys use specially selected rocks to crack nuts placed on anvil stones while standing bipedally to hurl the rocks at the nuts. Inexperienced individuals seem to realize who is an expert and preferentially watch them, but it takes them at least two years of individual trial-and-error learning to become proficient at the skill. Rather than grabbing stores and attempting to hurl them at nits when the see others doing this, novice monkeys begin by playing around with the nuts and banging them on things. It takes years–not hours, minutes or days–to “get” the relationship among hurled stone, nut and anvil. pp 43-44

This ending discussion of imitation is one of the very fine examples of how seemingly general intelligence relies on many specific adaptations.

Hill Climbing

This chapter is about the adaptive landscape, which is a view of Evolution as Algorithm. This is not so much new here, but it is important background.

We need to think about adaptive landscapes a little differently than real landscapes. We tend to think f landscapes as passive: the shapes of mountains aren't (mostly) altered by out climbing them. Although it's easy to think of organisms and environments as entirely distinct, cases like the evolution of sex, cooperation, and culture show that phenotype space and environment space aren't mutually independent spaces. Our phenotypes are both part of the environment to which adaptation occurs, and constructors of it. The term “niche construction” is sometimes used to describe this reverse form of evolutionary causation.

[…]It's impossible to understand evolutionary change without understanding changes in development. Because evolution is the developmental process looped over and over again, evolutionary change is developmental change. In this sense, evolutionary space and developmental space are not separate spaces. p 54

This rather philosophical passage is in a section discussing the importance of an information processing view of mind, with a nod to Information theory (see also The User Illusion). Information processing and evolutionary theories of mind seem wrong to many people because they don't seem to capture the personal subjective experience of having a mind. Yes, this is a weird nonintuitive way to think about ourselves, but it seems to be scientifically valid. See Intentional Design and Intentional Opacity.

This is related to Boyd and Richerson's “big mistake” critique of the 1990 evolutionary psychology synthesis, see Not by Genes Alone. We agree that the story of mismatch has been overdone (see just_so_story), and suggest that this may be partly because of the mythic appeal of the golden age followed by fall from grace (see Human Origins and Original Sin and Modern Times).

This chapter is about a theory of how information processing in the brain adds meaning to what we perceive. He uses computer-like terminology such as “tags” and “object files”, which may or may not appeal, but the basic idea that mind constructs meaning is not controversial. He goes on to give a good example of how it seems that the minds of animals (including humans) are prepared to perceive the special category of food, which makes a lot of sense from an evolutionary perspective.

Contrary to the commonsense view that separates “cognition” and emotion“, I am going to suggest that the conceptual/thinking parts and the emotional/motivational parts of mental activity, while they may involve distinct psychological mechanisms and processes, cannot be decoupled. They fit together like a lock and key, for a reason: Conceptual distinctions exist only because of their importance for, ultimately, action, the things we can do with them; and emotional and motivational systems exist only because the world is parsed in ways that allow them to make us decide to do the right thing. p 87

Amen brother!
Without emotion, people are incapable of functioning. See also Unconscious and Descartes' Error.

Hunger, of course, also involves learning–not that we learn hunger itself or how to be hungry, but rather we learn what to be hungry for. Here again, there is a mysterious-seeming but fascinating process of associative learning via experience, because we often get cravings for specific items. This is likely to be, at least in part, because our physiological nutrient evaluation systems have experienced particular foods and evaluated their contents in the past, and then updated the representational parts of our brains–perhaps even down to the perceptual level–so that when we are lacking salt, for example, those potato chips look especially good, or that Gatorade is especially thirst-quenching. Although the details are, as far as I know, not still entirely known, it seems likely that natural selections has engineered a communication channel that goes both ways between these various systems. The search for nutrient specific cravings has turned up some broad categories of cravings, such as those for carbohydrates, salt and sugar. Certain kins of foods, like meats and sweets, show up high in studies of cravings across cultures–moderated, of course, by culture-specific food traditions–and some cultures even have words or concepts for specific kinds of hunger, like “meat hunger”. p 93

A good example of an experience that's familiar and demands an evolutionary explanation, but isn't part of the standard EP example set. This is also an example of the author's ambition to bring all behavior within the scope of EP, not just uniquely human behaviors.

Perhaps the most interesting and well-studied example of social learning about food comes from Norway rats, studied by psychologist Jeff Galef and colleagues. It starts in the womb: Fetal rats are sensitive to food particles in their mother's bloodstream that cross the placental barrier. Rat babies after birth are more likely to eat what the mother ate while she was pregnant. This is “social” in a sense, but it also involves, as does all social learning, individual experience: The reason the food particles are in the mother's bloodstream is a social one (they reflect her food choices and not those of the fetus), but the experience of them is individual, that of the fetus alone. Imagine the representational mapping process that must take place in order for this to occur. Something in the fetal rat's blood system detects food particles, which are then mapped to smell, which is mapped to behavioral decisions after the baby is born. This is a specialized learning system if ever there was one. The advantage of this system, of course, is that the baby benefits from the mother's lifetime of learning about what is good to eat. In fact, the process continues after birth when the baby developed preference for food detected in its mother's milk–presumably by the same, or a similar, mapping mechanism. In addition to being an example of social learning, this is an example of what is sometimes called prepared learning: learning caused by a system specialized to make inductive bets about the structure of a particular learning domain. […] as I've pointed out, all mechanisms have to be specialized for something; every adaptation has a domain. Thus, there is a sense in which all learning is prepared learning, because any learning mechanism, even the most general-purpose, must make at least some assumptions about the structure of its inputs and how that structure predicts the future. p 96

It's somewhat daunting to our ambitions to understand the human mind when we see this evidence of how complex, specific, and general is the adaptation of a rat's mind. We really have basically no idea how the rat does this, at the detailed level of cellular mechanisms.

Food substances can be subdivided in any way and still retain the affordances that matter for them as food: half a steak is still a steak. The same is not true for many of the affordances of cows that matter for our interactions with them, such as their ability to move and respond to stimuli. Cut a cow in half and those properties, which psychologists call properties of intentional agency, disappear. Not surprisingly, then, children agree that half a steak is still a steak, but half a cow is not a cow–a linguistic signature of two distinct ontologies at work, an ontology of substances and an ontology of intentional agents. p 99

This is from the author's own research, and is consistent with the idea that, when it comes to animals, we are prepared to learn that the category of food overlaps with category of animate things. All very reasonable, but I laughed when I read this, because the abstract philosophical language seemed to clash with the very vital details. This is an example of a very important point. It is plausible that humans have mental adaptations related to meat-eating, because it seems likely this was an important food source for ancient humans. We could go so far as to say that meat eating is natural. But this does not mean that we must eat meat when other replacement foods are available, and many people choose not to eat meat, for moral or practical reasons. Evolution has added a capability, but we aren't required to use it. It may also be that some of the adaptations that were originally “for” meat-eating have been repurposed to regulate consumption of non-animal foods. This was not the proper domain in the EEA, but is part of the actual domain.

Social Ontology

Humans are uniquely social animals, and our minds contain many adaptations to our social world. In many cases, it is this humanly constructed environment which is most important in determining an individual's evolutionary success (fitness). This chapter is about how we construct social meanings.

This next quote is in a discussion of how, based on studies of babies, it seems that humans are prepared to parse the behavior of others into discrete actions, and then to infer intended goals from those actions. This is important for many human interactions, including imitation (discussed before) and theory of mind. In this experiment, a hand appears and removes a bear, leaving behind another object (a ball). Then the scene resets with the objects swapped. The baby is more surprised if the hand takes the ball the next time. It seems that the baby has inferred that the hand (and attached person) wants the bear.

What is particularly interesting is that this effect appears to be specific to grasping hands. When a mechanical claw is shown reaching for an object, babies do not encode it as wanting a specific thing. This is evidenced by the fact that they are not surprised if it reached for different things later. Moreover, when babies see a hand reach out and thebacko of the hand touches the bear, rather than grasping it, the babies do not encode brushing against the bear as the specific goal of the hand; they are not surprised if the back of the hand later touches the ball. This suggests that even though the babies are undoubtedly doing a spatiotemporal analysis of the scene, it's not just the “low-level” statistical properties of it, such as what is touching what and for how long, that are influencing infants' responses. Otherwise, they would respond the same way to the claw, back-of-hand, and grasping conditions. p 123

A very cool study that does support the idea that humans are specifically prepared to learn things about grasping hands.

Minds

This chapter is about theory of mind and “mindreading”. Humans do have quite distinctive capabilities here, but Barret does an intriguing exploration of how this is not entirely distinct from the capabilities of animals. This is important, because animals with simpler, less general, adaptations to theory of mind can give us some idea of how our much more sophisticated capabilities could have evolved. He uses the example of an antelope which recognizes an object in the environment (a leopard), recognizes that it is animate, from its intentional motion and appearance, then recognizes its face and its eyes. The antelope sees that the predator is looking in this direction, and becomes alarmed, preparing to flee. We know that many animals recognize gaze, and this as obvious adaptive value, but these very sorts of capabilities are part of human capabilities for “mind reading” and predicting intentions. Some animals also have a limited ability to understand what other animals of their own species have seen, and therefore presumably know, but this seems to be limited specifically to competition for food.

Development

Making development central to evolutionary psychology is one of the key contributions of this book. Organisms only exist as they do because they have developed from a single cell. What our genome encodes is potential developmental processes, including adaptive plastic responses to environmental variation. Our minds only exist as they do because they developed that way, a process that is not even vaguely complete until our early 20's. Yet our minds are also prepared in many ways to enable us to learn to function past worlds which were in many ways very much like the world we live in today. These are evolution's inductive bets. But isn't today's world completely different? Well, no. Many of these bets are extremely easily satisfied by any environment that is capable of supporting human life. For example, children play, and in the process learn about their bodies and how they are able to affect the world. Nothing can develop without an environment, but nothing can develop without genes either. Many features of humans (and other organisms) are reliably developed when the environment has typical features. For humans, typical environmental features include other people, such as parents and peers.

A crucial concept from development is plasticity. The organism shapes its development in response to the environment. Plasticity is not an alternative to evolution and genetically controlled mechanisms. If an organism shows adaptive plasticity, this can only be because past environmental variation caused a plasticity mechanism to evolve.

Crucially, the underling design [DNA mechanism] will often be relatively invariant in the sense I have outlined here: The design features that produce adaptive plasticity need to have replication fidelity in order to be selected for. But the phenotypic outcomes of this design might be wildly variant in phenotype space, depending on the mapping properties of the design. And no matter how wildly diverse the outcomes of that mapping function produces might be, it will only produce systematically adaptive outcomes if it has been selected to do so under the various condition it might encounter. p 166

“Mapping properties of the design” refers to what developmental result (body and behavior) we get in different environments, according to the development processes encoded in the genetic mechanisms.

These [developmental] systems can be expected to produce the most adaptive (fitness-good) outcomes when the environments they currently occupy match their EEA along relevant dimensions. […] The logic of the principle follows from the fact that the trial-and-error process of natural selection happened in some sort of environments that actually occurred, and no other. Therefore, although it is possible that the designs produced by this process might produce outcomes that are even fitness-better in certain environments that the population hasn't yet experienced, the only ones that are guaranteed to reliable or systematically produce fitness-good outcomes are ones that resemble those in which the design features of the adaptations were forged. Outside of that statistical envelope of conditions, all bets are off. This does not mean that we couldn't find some previously unexperienced environments where an evolved design would do better than in the conditions under which it evolved. Indeed, we could almost be guaranteed to find some fitness-better ones, if we knew enough about the design in question and how it worked. p 169

So, we expect that fitness will generally be best when the current environment resembles the EEA, which is the basic EP mismatch hypothesis. But it is entirely possible that new environments may be in some ways fitness-better than the EEA. If that happens, it can only be regarded as luck.

Flat reaction norms seem to be at least one part of an intuitive notion of “innateness”: Sometimes when people say that a trait is innate, they mean that it must be identical in everyone. But I do hope you can see that innateness in this sense is not the same thing as phenotypic outcomes being “caused by genes” or “caused by evolution,” since the outcomes of plastic reaction norms are also caused by genes and evolution, just like flat ones. Some people might like to solve this problem by just defining the reaction norm as innate, whether or not it's flat, which is a defensible position. But if you do that, you have to agree not to point at some phenotypic outcomes as innate and others not. p 174

A “reaction norm” is a technical name for how an organism develops in response to the environment, given its genetic makeup. A flat reaction norm means that the same result develops, regardless of the environment. A reaction norm might be flat because the organism struggles against the environment to maintain a single desirable outcome, or it might be flat because the organism is simply indifferent to those changes. How does this relate to our intuitions about innateness? A reaction norm, flat or otherwise, seems like the sort of thing we mean by innate, but this does nothing to get us to a place where we can say that some developmental outcomes are innate and others aren't. He goes on to suggest that in complex organisms like humans, flat reaction norms are probably rare. The concept of “reliably developed” is closer to a useful conception of innate, but it explicitly includes the possibility that there are individual exceptions, and that this developmental outcome is only reliable in some range of environments. If for some reason there is a desire that humans not develop this outcome, then we can't rule out the possibility that there are environments which would have that effect.

Perhaps the simplest type of developmental plasticity involves what biologists call morphs–discrete types or phenotypic designs within a population. […] In humans and many other sexually reproducing species, the most obvious morphs of this kind are males and females: Humans born with two X chromosomes develop into the the female phenotype, and humans born with an X and a Y develop into the male phenotype. I say “develop into” males or females to emphasize that human males and females are, in terms of the genes and developmental systems they inherit, virtually identical, or at least very highly overlapping. Both sexes have almost all the DNA required to develop into either males or females, the critical difference being the possession of the DNA on the Y chromosome that will make you male, if you have it. This is oversimplifying a bit. But the basic point stands: there is a kind of a switch, determined by a tiny bit of DNA that you are born with, that turns you into one of the two possible morphs. This falsifies a common misunderstanding about the relationship between genes and phenotypes, namely, that lage differences in phenotypic design between individuals require corresponding large differences in DNA between those two individuals. […] differences between the sexes can be, and largely are the result of how the same genes are regulated, turned on and off and interacting during development, rearranging hor particular developmental resources are used. p 176

Other examples of morphs are the different forms seen in social insects, such as ant warrior, worker and queen types.

The regulatory systems that turn things in your brain on and off depending on circumstance–for example, in happy times or sad times, hungry times or full times–may be examples of plastic developmental systems that operate over relatively short timescales. There is no question that this is a nonstandard use of terminology, but it might make sense to think of ourselves as containing a variety of morphs–what might be called contextual morphs–that we can slide into or out of, like Jekyll and Hyde, as circumstances dictate. Such phenotypic states would, of course, have reaction norms, and would have to obey the rules of mind-world fit if the were to have evolved. p 179

This is somewhat similar to, but importantly different from, some ideas from the nature/nurture controversy. We can see that many human behaviors are fairly reliably developed in current environments, and some of those reliable developments, such as sex differences in behavior, are regarded by some as (morally) bad, leading to a desire to change that developmental outcome. This simple fact of reliable development in existing environments does not mean that something different couldn't happen in a completely new environment (such as a new culture), though given the wide range of current environments, it does suggest that the environmental changes would have to be pretty drastic, and might have unintended consequences, both fitness-bad and moral-bad. We argue that if you want to create new environments to improve the human condition, then you're a lot more likely to succeed if you find out “enough about the design in question and how it worked.” Karl Marx has been paraphrased as saying “The question is not how to understand the world, but how to change it.” But it's very hard to improve the world without understanding the relevant bits.

I add that the outcome might be different when genetics differ, because the reaction norm is defined by the interaction between the genetic plasticity mechanisms and the environment. So in a genetically diverse population, this could result in diverse outcomes, even in the same environment. And even in genetically identical lab animals raised in environments that have been made identical as possible, there is still variation in some outcomes. This is developmental_noise.