Environmental Complexity and the Evolution of Cognition – Peter Godfrey-Smith

Page 1

1

Department of Philosophy
Stanford University
In R. Sternberg and J. Kaufman (eds.) The Evolution of Intelligence.
Mahwah: Lawrence Erlbaum, 2002, pp. 233-249

1. Starting Simple

2. The Environmental Complexity Thesis

3. On Complexity

4. The ECT as a Component in Many Evolutionary Scenarios

1. Starting Simple

One problem faced in discussions of the evolution of intelligence is the need to get a precise fix on what is to be explained. Terms like “intelligence,” “cognition” and “mind” do not have simple and agreed-upon meanings, and the differences between conceptions of intelligence have consequences for evolutionary explanation. I hope the papers in this volume will enable us to make progress on this problem. The present contribution is mostly

focused on these basic and foundational issues, although the last section of the paper will

look at some specific models and programs of empirical work.

Some people have a very demanding picture of what is required for intelligence,

thinking that it always involves such sophisticated skills as planning, language-use, and

perhaps even some sort of consciousness. To these people, intelligence is to be contrasted

with instinct. Perhaps in this rich sense of the term, intelligence is even to be contrasted with

the simpler types of learning, such as learning through reinforcement (operant

conditioning). From this first point of view, the problem of explaining the evolution of

________________________________________

Page 2

2

intelligence is explaining why instinct and other simple behavioral capacities were not

enough; why evolutionary processes took a few organisms so far beyond these basic

behavioral skills.

Another approach uses terms like “intelligence” and “cognition” in much less

demanding senses. On this second approach, intelligence is not restricted to a few

exceptional cases in the biological world — humans and perhaps some primates. Rather,

intelligence exists to some degree in a huge range of living systems. Humans have a lot

more of it than cockroaches do, but cockroaches do have some of it.

According to this second view, all the mechanisms that enable organisms to

coordinate their behavior with conditions in the world involve some degree of intelligence.

So there is no opposition between intelligence and what is often referred to as “instinct.” An

instinctive behavior can involve perception, and a good deal of processing and feedback to

ensure the right match between behavior and circumstances. According to this second view,

that is a low-level variety of intelligence.

Maybe it is difficult to motivate the second, less demanding approach with the term

“intelligence” which has such definite common sense usages. Perhaps it is better to present

this view with the aid of a less everyday term such as “cognition.” So in this paper I will

mostly discuss the “evolution of cognition” rather than the evolution of intelligence. The

term “mind” is another broad one, but it has its own capacity to mislead, as for many people

it is closely linked with consciousness and a sense of self.

Given this, we can describe the second approach by saying that a great range of

living things have some cognitive capacities. In many cases these capacities are extremely

limited. The capacities we habitually refer to as “intelligent” in humans, such as the

capacities for planning and conscious reflection, comprise one type of cognition. But when

a fish negotiates its way around a reef, or a rat finds its way back to a food source, the

internal processes responsible for these behaviors are varieties of cognition as well.

Does it matter which of these two general approaches to the evolution of cognition

we take? Certainly it does not much matter what we decide to refer to with the term

“intelligent” or “cognitive.” Either way, the problem remains of explaining how nervous

systems evolved at all, and the problem remains of explaining how humans became so much

smarter, in many respects, than other organisms. But I think it often does make a difference

how we view and describe the continuities between human and non-human psychological

capacities. A great deal depends on how much significance we place on the distinction

between animals that use language and those that don’t, for example. Views about non-

human cognition often have ethical consequences, and consequences for a range of issues in

the philosophy of mind.

________________________________________

Page 3

3

In any case, my own approach is very much along the lines of the second option

described above. I approach the problem with a broad and very undemanding concept of

“cognition.” My aim is to set out with a broad concept of cognition and ask: can we

formulate a generalization about why these sorts of capacities will tend to evolve? Because I

use “cognition” to refer to such a broad class of capacities, cognition is not a single

evolutionary “discovery,” restricted to a single lineage of organisms. Cognition of various

kinds has been discovered and rediscovered by evolution many times, just as eyes and

wings have been discovered independently several times. And just as we find with eyes,

cognitive machinery is very diverse. There are lots of ways to process information and

control behavior; a central nervous system is one way, but not the only way. An important

feature of this view, which will be discussed in more detail below, is that cognition “shades

off” into other kinds of biological capacities and processes. In some cases it is hard to

distinguish cognition from other control systems in the body, and hard to distinguish

behavior from such things as growth, development and the regulation of metabolism.

Cognition is diverse, but it might be possible to find a common type of evolutionary

story that applies in all or most of these diverse cases. With such a general framework in

hand, we can then ask more specific questions about why certain types of cognition evolved.

When is learning favored over less flexible strategies of dealing with the world? And if one

is to learn, when is learning through individual trail-and-error better than learning by

imitating a parent? What brings about the transition to a planning intelligence? When is

what primatologists refer to as “theory of mind” (taking other individual organisms to have

a mind) favored? And what on earth lay behind the explosion of mental capacities found in

the evolution of humans?

It might turn out that all these explanations are so diverse that it is pointless to try to

link them under a general principle. I do not deny that possibility. But my own approach

here is to outline and cautiously defend one possible generalization about the adaptive value

of cognition. The generalization is intended to be a fairly obvious one, something that has

been expressed in partial or imperfect ways dozens of times before, dating back to the 19th

century. My aim here is more to make a vague existing idea into a precise one, than to

present an novel idea. I do think it could be of considerable help to discussions about the

evolution of cognition if this underlying idea, and its possible rivals, were made explicit and

precise in people’s minds.

2. The Environmental Complexity Thesis

Here is my proposal for a general “first principle” about the evolution of cognition:

________________________________________

Page 4

4

Environmental Complexity Thesis (ECT):

The function of cognition is to enable the agent to deal with environmental complexity.

Each of the key terms in the ECT requires a good deal of clarification. The term “function”

is understood here in a strong sense. To ascribe a function in this sense is to offer an

evolutionary hypothesis. The function of a trait or structure is the effect or capacity it has

which has been responsible for its success under a regime of natural selection. When we

say that the function (in this strong sense) of the thorns on a plant is to deter herbivores

from eating the plant, we are not just saying that the thorns help the plant by deterring

herbivores. We are saying that thorns were selected for in evolutionary processes because

they tended to have the effect of deterring herbivores. The ECT makes a similar claim about

cognition.

The ECT is a broader and more abstract claim than the one about the thorns, but

very abstract functional claims can certainly be made. Eyes have evolved many times, and

they make use of various different types of mechanisms. But a general claim can be made

about their evolutionary function: the function of eyes is to respond in discriminative ways

to light, and hence to enable the organism to make use of information about the world that is

carried in light (Gibson 1966, Dretske 1981). We can also formulate an even more general

thesis about the function of perceptual mechanisms: they all respond, with some degree of

sensitivity or discrimination, to some physical or chemical variables that impinge causally on

the organism, in such a way as to enable the organism to make use of information about the

world that is carried by these variables.

Some might say that when we formulate a broad generalization like this, we are

saying something so obvious or empty as to make it not worth the effort of saying. I

disagree. First, it is important to be able to embed specific functional claims made about

biological structures within a more general picture of what the organism as a whole is doing.

Second, it is actually quite hard to get these generalizations right, and lots of puzzling

questions can get raised along the way. For example, in the claims about perception I made

above, I used the concept of information. Although lots of people, including scientists, talk

about information in a casual and unreflective way, it is a subtle and difficult concept. The

philosopher Fred Dretske (1981) developed a detailed theory of where information is found

in the physical world, and of how everyday talk about information is related to the technical

discussions of “information theory” in engineering. Information, for Dretske, is found

where there is contingency and correlation. Any variable in the world which has a range of

possible states is a source of information. When the state of a source of information is

________________________________________

Page 5

5

correlated with the state of another variable, as a consequence of physical laws, the second

variable carries information about the source. For Dretske, information is a resource that

organisms use to make their way through the world; cognitive systems are information

consuming, or information exploiting, systems.

Without careful, explicit discussions like Dretske’s, it would be unclear whether or

not it is really justifiable to use the concept of information when making generalizations

about the function of eyes and other perceptual systems. Whenever I use the term

“information” in this paper, I have Dretske’s sense in mind. Below I will discuss the concept

of “environmental complexity” in some detail; this is another concept which it can be easy to

throw around without having a clear idea of what is being said.

So far I have said that the ECT is an attempt to give a general functional explanation

of cognition. Functional explanations have received a lot of attention in the philosophical

literature, especially over the past 20 years or so. (For a collection of classic and recent

articles, see Allen, Bekoff and Lauder 1998) In this paper I will not discuss the many issues

that have arisen in these debates, as the present topic is not functional explanation in general

but evolutionary hypotheses about cognition. All that is important for present purposes is

the idea that functional explanations are attempts to describe, in a shorthand way, the

processes of mutation and natural selection that were responsible for the origination and

maintenance of biological structures. Functional explanations are attempts to isolate the

effects or dispositions of a structure which were responsible for the natural selection of that

structure. So functional claims are “teleological” only in this specific Darwinian sense.

In the first section I said I would be using a broad and undemanding sense of the

term “cognition.” But how broad? What exactly is cognition? What is the set of organisms

in which it is found?

I understand cognition as a collection of capacities which, in combination, allow

organisms to achieve certain kinds of coordination between their actions and the world. This

collection typically includes the capacities for perception, internal representation of the

world, memory, learning, decision-making and the production of behavior. This set of

capacities, according to the ECT, has the function of making possible patterns of behavior

which enable organisms to effectively deal with complex patterns and conditions in their

environments.

So although the ECT is expressed as a connection between cognition and

environmental complexity, it really embeds two separate claims. One is the claim that the

immediate role of cognition is to control behavior. The other is the claim that the point of

this control of behavior is to deal with environmental complexity.

________________________________________

Page 6

6

Think of cognition here not as a single type of process, but as a biological “tool-kit”

used to direct behavior. There is no single list of tools found across all the organisms with

cognitive capacities; different organisms have different collections of tools, according to

their circumstances and history. So when I listed “perception, internal representation of the

world, memory, learning (etc.)” above, it should not be thought that this has to be a

description of a set of recognizable and distinct “modules” common to all cognitive

systems. Rather, I am referring to a set of capacities which are realized in very different

ways in different organisms, a set of capacities which shade into each other and shade off

into other, non-cognitive parts of the biological machinery. Also, the capacity for “internal

representation of the world” is one about which there is enormous disagreement within

philosophy and cognitive science (Stich and Warfield 1994). Here again, some people

understand this ability in a very demanding way, perhaps requiring language, while others

view a very wide range of mechanisms for registering external events as representational in

some sense. Here I assume a very loose and undemanding view of representation, but the

issues surrounding that concept are too complex to go into here.

In the list of basic cognitive capacities given above, some are more fundamental than

others. Perception is very fundamental; learning is somewhat less so. It is true that all

macroscopic animals are thought to be able to “learn,” in at least a minimal sense. Bees have

been shown to have quite impressive ways of learning the location of food sources, and fruit

flies have been conditioned to exhibit avoidance behaviors in some of the same sorts of

ways found in more celebrated learners such as rats and pigeons. The neural basis for the

most minimal kinds of learning is often studied using sea slugs. Still, learning is not in

principle essential to cognition. A behavioral pattern which is completely insensitive to any

refinement through learning, but which does involve the coordination of actions with

perceived environmental conditions, does display a minimal type of cognition.

But if learning is not essential, where does cognition stop? Do plants have it? What

about bacteria? By any normal standard, plants and bacteria do not have minds and do not

exhibit cognition. But my suggestion is that cognition shades off into other kinds of

biological processes. There is not much point in trying to draw an absolute line. Plants and

bacteria do exhibit some capacities for flexible response to environmental conditions, using

environmental cues to control development and metabolism. These are low-level cases of the

same types of capacities that, in more elaborate cases, do constitute cognition.

Many bacteria can adjust in adaptive ways to changing circumstances around them.

Dretske (1986) discusses aquatic bacteria which use little internal magnets to track the

distinction between north and south, enabling them to move towards water with their

required chemical properties. Bacteria also make use of external cues to adjust their

________________________________________

Page 7

7

metabolic activity. A famous case is the lac operon system in e. coli bacteria. These bacteria

can respond to a change in local food type through processes in which the availability of a

nutrient affects the regulation of genes which code for enzymes able to digest that nutrient.

Plants are able to direct a range of their activities with the aid of cues from the

external world. Within “activities” here I include some cases which fall naturally under the

headings of growth or individual development, and others which might be distinguished

from development and considered genuine behavior. Silvertown and Gordon (1989) argue

at length that plants can behave, but they use an extremely broad conception of behavior. I

suggest a narrower construal; one rough way to distinguish plant behavior from plant

development is to say that behavioral changes must be reasonably rapid and also reversible.

Then there will be a large range of cases where plants adaptively control individual

development with environmental cues, and a smaller range of cases where they control

behavior. Genuine plant behaviors include the behaviors of Venus Fly Traps, and a large

range of reversible responses to local light conditions.

Some of the ways in which plants use environmental cues are quite sophisticated.

For example, many plants can determine not just that they are being shaded, but that they are

being shaded by other plants. This is done by detecting the wavelength properties of

reflected light. The plants respond to shading by growing longer shoots (Silvertown and

Lovett Doust 1993 pp. 11-12). Lest I leave out the least glamorous biological kingdom,

some soil fungi have a reflex which enables them to trap and digest tiny wandering worms.

Though it makes sense to distinguish control of plant behavior from control of plant

development, for many theoretical purposes these can be seen as similar capacities. From

the point of view of theoretical modeling, control of developmental processes and control of

behavior have much in common (see section 4 below). And the temptation to use intentional

and cognitivist terms when describing control of plant development in more informal ways

can be strong. David Attenborough begins his book The Private Life of Plants as follows:

“Plants can see. They can count and communicate with one another” (1995 p. 7). Most of

the phenomena Attenborough is referring to here involve control of growth and

development, such as tactile exploration by a young vine, looking for a tree to climb.

I said that plants use a lot of their “smarts” for controlling growth and development,

rather than behavior. In fact this phenomenon is not restricted to plants, but is found in

vertebrate animals as well. In certain fish, capacities for perception and information-

processing are used in directly regulating central aspects of development. These fish

determine whether they will develop as male or female via perception of their relative size

within the population (Francis and Barlow 1993). So when I said earlier that the ECT links

cognition first to behavior, and then to environmental complexity, this was a slight

________________________________________

Page 8

8

oversimplification. Even in animals, sometimes cognition controls things other than

behavior.

I have spent some time discussing capacities for flexible response which do not

constitute genuine cognition on any normal standard. My aim in discussing these cases is to

suggest that cognition shades off into other kinds of biological processes; even though

plants do not exhibit cognition and people do, there is no single scale between them and us

with a threshold marking a transition to genuine cognition. Rather, all or practically all living

organisms have some capacities for responding to environmental changes and conditions.

Sometimes environmental cues are used to control metabolic processes or development,

sometimes they are used to consciously choose where to plant crops. In ordinary talk and in

theoretical discussion, we habitually pick out only some of these capacities as “intelligent”

or “cognitive,” and the decision to do so can be guided by a mixture of criteria. Complexity

and flexibility play a role, but so does the time-scale at which responses occur. There is

nothing wrong with that; my point is just that there are some fundamental similarities

between real cognition and much simpler capacities for control of biological processes, and

there is no reason to seek a sharp cut-off between the two classes.

As some terminology might be useful here, I will say that plants and bacteria have a

number of “proto-cognitive” capacities. These are capacities for controlling individual

growth, development, metabolism and behavior by means of adaptive response to

environmental information. The term “development” refers here only to processes within an

individual lifetime; evolutionary change is not classified as proto-cognitive. Complex multi-

cellular organisms like ourselves also contain “subpersonal” systems with some of the

proto-cognitive capacities of simpler whole organisms. The vertebrate endocrine and

immune systems are examples. In this paper I will not discuss the very difficult questions

raised by the attribution to proto-cognitive capacities to higher-level systems, such as ant

colonies.

In stressing that cognition shades off into other proto-cognitive biological processes,

I am asserting a version of what is sometimes called a “continuity” assumption about

cognition. The simplest biological capacities that we might consider proto-cognitive are

cases of flexibility in behavior or development (etc.) controlled by a fixed response to a

physically simple environmental cue, but where the nature of the response is not determined

directly by the physical properties of the cue. (There has to be some “arbitrariness” in how

the cue affects the system, to use a term due to Levins 1968.) As we add different types of

flexibility of response, and different kinds of inner processing of the output of perceptual

mechanisms, we reach clearer and clearer cases of cognition. But there is no single path that

takes us from the simplest cases to the most elaborate. There are various ways of adding

________________________________________

Page 9

9

sophistication to the mechanisms of behavioral control, ways which will be useful to

different organisms according to their circumstances. The ability to expand or contract the

range of stimuli coupled to a given response is one important sophistication (Sterelny

1995). The ability to learn through reinforcement is another. Yet another is the ability to

construct a “cognitive map” of spatial structure in the environment (this case will be

discussed further below). It is an error to try to describe a single hierarchy of cognitive

skills, from simplest to most complex. Here as elsewhere, there are many distinct kinds of

complexity.

Compare two imaginary organisms which both have good spatial memory. One is

“more sophisticated” than the other because it can remember more features of the

environment, and can use its knowledge to find novel routes to where it wants to go. But this

first organism can only acquire this spatial information by first-hand experience, by

laboriously traveling and remembering the terrain. The second organism has a more limited

capacity to remember features and to manipulate its internal model of the world, but it can

acquire its knowledge in a richer variety of ways. It can infer spatial structure from the

behavior of other organisms. In some respects the first organism is smarter, but in other

respects the second is.

As we add sophistication to the tool-kit of behavior-guiding capacities, we eventually

reach clear, unmistakable cases of cognition. And these clear cases do not all involve

humans. Some birds, such as Clark’s nutcrackers in the south-west of the US, hide stores of

food when supplies are good, and retrieve it in times of scarcity. To do this requires a

sophisticated combination of perceptual abilities and spatial memory. In the sense in which I

am using the term “cognition,” there is nothing marginal about such cognitive capacities.

So I claim that we reach uncontroversial cases of cognition before we reach

language use. And I have left out all mention of the “qualitative,” first-person, “how it feels”

side of mental life. My overall position is that we do have reasonably good evidence to posit

rich qualitative states in non-human animals. But that is a separate point which does not

matter to the task at hand.

To use a very broad sense of “cognition,” as I do here, does not require postulation

of fundamental similarity in the cognitive processes in all these diverse cases. Indeed, I have

been stressing the opposite — the diversity of ways in which cognitive and proto-cognitive

capacities are realized. Many people have given general overarching theories of how all

cognitive processes work. Examples include the general theories of learning that dominated

psychology in the middle decades of this century. If one has an overarching theory of this

kind, then one will want to have a broad term like “cognition” to capture what one is

generalizing about. (Although many of the behaviorist psychologists who defended general

________________________________________

Page 10

10

theories of learning would not have liked the term “cognition.”) I am skeptical about those

overarching theories of cognitive processes and mechanisms, and I use a broad sense of

“cognition” for what might be called more “ecological” reasons. There is a certain kind of

job that the collection of processes I am calling “cognitive” performs. For one reason or

another, organisms acquire capacities for behavior and machinery to control this behavior.

The behavioral machinery acquired is diverse, and so are the mechanisms used to control

this behavior. Our sophisticated human mental abilities are one instance of this evolutionary

phenomenon, but the abilities of bees and jaguars are as well. Understanding the evolution

of cognition is understanding this whole domain of evolution’s products.

3. On Complexity

The ECT claims that the function of cognition is to enable organisms to deal with

environmental complexity. But what exactly is environmental complexity? There is a good

deal of unpacking to do here (see also Godfrey-Smith 1996.)

I suggest that the most useful concept of complexity here is a simple one.

Complexity is heterogeneity. Complexity is variety, diversity, doing a lot of different things

or having the capacity to occupy a lot of different states.

There are many different kinds of heterogeneity, hence many kinds of complexity. It

is not just unnecessary, but positively mistaken, to try to devise a single scale to order all

environments from the least to the most complex. Rather, any environment will be

heterogeneous in some respects, and homogeneous in others. Environments can be

heterogeneous in space and in time, and spatial and temporal heterogeneity exists at many

different scales. An environment with a large number of different possible states which

come and go over time is a complex environment, in that respect. So is an environment

which is a patchwork of different conditions across space. The heterogeneity property is not

the same in these two cases, but in both cases heterogeneity can be opposed to

homogeneity. A complex environment is in different states at different times, rather than the

same state all the time; a complex environment is different in different places, rather than the

same all over. Whether a particular type of complexity is relevant to an organism will

depend on what the organism is like — on the organism’s size, physiology, needs and habits.

The heterogeneity properties of environments are objective, organism-independent

properties, but among the countless ways in which an environment is structured and

patterned, only some will be relevant to any given organism. (See Levins 1968 for a classic

discussion of some of these issues.)

________________________________________

Page 11

11

In the ECT I said that cognition enables agents to “deal” with environmental

complexity. That terminology suggests that environmental complexity is seen as posing

problems for organisms. Often this is so, but I do not want to put too much weight on the

concept of a “problem” here (Lewontin 1983). In some cases, environmental complexity

provides what would normally be called an “opportunity” rather than a problem. A

population might be located in a fairly benign set of circumstances, but one where tracking

and adapting to environmental complexity makes it possible for some individuals gain a

reproductive advantage over others. Natural selection works in a comparative way; the

absolute level of hardship is in general not important in understanding evolutionary

processes within a population. So while I will often write of the “problems” posed by

environmental complexity in this paper, occasionally I will use the term “opportunities” as

well. The distinction between the terms is mostly an everyday one which should not be

taken too seriously in this context.

If we think of complexity just as heterogeneity, this concept of complexity can be

applied to organisms as well as to environments. An organism is complex to the extent that

it is heterogeneous. Here again, there are different kinds of heterogeneity; an organism can

be heterogeneous in many different respects. (For different concepts of organismic

complexity, see McShea 1991.)

Cognitive capacities themselves are complex, so the ECT can be seen as claiming

that one kind of organic complexity has been produced by evolution to enable organisms to

deal with environmental complexity. Dealing with complex problems by means of

perception and action can be seen as a special case of a more general phenomenon: dealing

with environmental complexity by means of flexibility.

This way of looking at the ECT is illustrated by the “proto-cognitive” capacities that

were discussed in the previous section of this paper. When a plant has the ability to

adaptively alter its development to suit its environment, this is a case of complexity in the

plant’s developmental capacities which enables the plant to adapt to heterogeneity in its

environment. Similarly, why do e. coli bacteria have their lac operon system of gene

regulation? The preferred food of e. coli bacteria is glucose, but sometimes glucose is not

available while other sugars are. The variability in the availability of different sugars is one

type of environmental complexity faced by bacteria. Metabolic machinery is expensive, and

e. coli have apparently been selected to economize in their production of enzymes. So the

enzyme needed to digest lactose is not produced in the absence of lactose. Instead, the

production of the enzyme is controlled by an environmental cue. Here the cue used is the

presence of lactose itself (and also the amount of glucose available to the cell — see Lodish

et al. 1995 pp. 421-22). The system of gene regulation used by the bacteria here constitutes

________________________________________

Page 12

12

one kind of complexity in these organisms, and this complex mechanism has a functional

explanation, of the strong type discussed earlier. The function of the lac operon system is to

enable e. coli bacteria to deal effectively with one type of environmental complexity –

variation in the availability of different sugars.

As I said earlier, I do not claim that bacteria exhibit cognition; this is at most a case

of proto-cognition. However, the ECT claims that the explanations for more complex and

genuinely cognitive capacities tend to have a similar general shape as this explanation for a

property of bacteria. The point of acquiring complex systems for behavioral control is to

enable the organism to deal with variation in what the environment confronts the organism

with, and variation in the opportunities the environment offers.

Environmental complexity figures in evolutionary processes that give rise to cognition.

But where does environmental complexity itself come from? And what should we make of

cases where environmental complexity is itself the product of organisms and their activities?

Environmental complexity itself has many sources. For the purposes at hand I will

make a loose distinction between two main categories. One source is the class of physical

processes which are more or less independent of the activities of the organisms under

consideration. Seasonal cycles provide an obvious example. And many resources that are

relevant to an organism’s well-being will be scattered through space in a way that is largely

independent of the organism’s own actions and properties.

When some type of organism acquires, through evolutionary processes, a way of

tracking and dealing with environmental complexity of this first kind, the explanatory

pattern described by the ECT has a straightforward causal directionality. But there are other

cases in which the situation is more complicated. These are cases where the environmental

complexity that organisms must deal with is either a causal product of, or is constituted by,

the activities of other organisms within the same population. Then we have a situation that

can exhibit feedback, or a “coupling” of organism and environment. (Lewontin 1983,

Odling-Smee 1988).

The most graphic examples are probably those that involve competitive interactions

between animals. If the only way for you to obtain and hold a resource is by winning

contests with other individuals in the same population, then these other organisms constitute

a key part of your environment. Their behavioral complexity constitutes part of the

environmental complexity you must deal with, so the behavioral capacities of organisms

similar to yourself are the source of a crucial kind of complexity in your own environment.

In behavioral ecology, contests of this kind are modeled with game theory (Krebs and

Davies 1987). Most mathematical game theory models only remain simple enough to be

comprehensible when many idealizations are made. Surrounding a few well-understood

________________________________________

Page 13

13

cases explicitly modeled with game theory, there is now a great deal of informal verbal

“modeling” (in scare quotes) and computer simulation of these sorts of interactions going

on. In fact, some have claimed that feedback processes of this kind are the key to

understanding the evolutionary transition to genuine human intelligence. Those suggestions

will be discussed in my final section below.

My present point is that these phenomena are not incompatible with the ECT. The

ECT need not be understood in a way in which the processes generating environmental

complexity are casually autonomous, or independent of the activities of the evolving

population in question. The ECT is compatible with the view that a centrally important

aspect of environmental complexity for many organisms is complexity that is made up by,

or caused by, the activities of other organisms of the same species. In those cases, the ECT

describes one part of a larger causal “cycle” — the part in which environmental complexity

puts selective pressure on organisms’ cognitive capacities. The other part of the “cycle” is

where the behaviors of organisms influence or determine the relevant patterns of

environmental complexity.

So some environmental complexity for a given organism is made up by the activities

of other organisms in that population. What about organisms from other species, which

constitute sources of food, or sources of danger, for the organisms we are concerned with? I

stress again that my two-way distinction here is rough and ready. To the extent that the

relevant activities of other species are causally influenced by the properties and activities of

the population under consideration, we have a case of the second “coupled” type. Predator-

prey interactions are a classic example. In general descriptions of ecological relationships,

people often stress that every species is connected to virtually ever other, through direct or

indirect causal chains. Clearly however this is a matter of degree. It is an error to over-

generalize about the richness of inter-specific connections, just as it is an error to treat

organisms as if all they ever have to deal with is an independent, causally autonomous,

physical environment.

I have been contrasting relatively simple cases in which organisms are responding to

environmental complexity that is causally independent of them, and more complicated cases

where the environmental complexity itself depends on the organisms in significant ways.

But even in the “simpler” class of cases, it should not be thought that I am suggesting that

the evolutionary processes themselves are simple and predictable. Much environmental

complexity is not relevant to any given organism, and the factors that contribute to some

aspect of complexity posing a problem are diverse and subtle. Suppose you move through

the world like a monkey, swinging from tree branches. Then the relevance of diversity in the

size and strength of these branches depends a great deal on your size. If you are small, most

________________________________________

Page 14

14

branches will support you and in any case a fall is unlikely to lead to serious harm. If you

are larger, paths through the forest must be chosen with care and pose a significant

information-processing problem because of the causal role of your own weight (Povinelli

and Cant 1995). Some branches will break, leading to a dangerous fall, while others will

bend, in ways that affect your possible next moves. Heterogeneity in the properties of tree

branches is thus relevant in different ways, and in different degrees, to differently sized

organisms.

The mere presence of environmental complexity that is relevant for a given type of

organism does not automatically generate cognition, or even natural selection for it. The

consequences of relevant environmental complexity also depend on many other features of

the organism and its ecology. The fact that the ECT is expressed as a simple generalization

should not be taken to downplay the role for “architectural constraints” in explaining why

evolution takes the course it takes in a particular case. (These constraints are famously

discussed in Gould and Lewontin 1979.) For some organisms, getting smart is not really an

evolutionary option, as a consequence of their basic biological lay-out, their characteristic

developmental sequence, or their overall ecology. Even for those that could, in principle, start

to respond to environmental variation by tracking and behaviorally adapting to it, the

appropriate genetic variation has to arise, and there will be costs associated with the

machinery required to take a smart approach.

It is also well understood that some kinds of environmental complexity can be

effectively dealt with by buffering it or blocking it out. One can respond to a threat by being

smart, but also by becoming impervious to it, via a strong shell or via sheer size. Some

organisms, including many insects, deal with certain kinds of environmental complexity with

an “r-selected” strategy for reproduction, in which there is massive reproduction in good

times, and little activity in bad times. To take this strategy it is necessary to be able to

produce huge numbers of quickly-maturing offspring when times are good. All this makes

for a lack of cognitive machinery in r-selected organisms.

So whether evolution takes a lineage of organisms down a path towards increased

cognitive capacity is contingent on a great range of factors, many of them having to do with

the “raw materials” that evolution has to work with in that particular case. But this fact does

not make the ECT false or naive. The ECT, when it applies to some particular case, is one

part of a more complicated and detailed explanation.

________________________________________

Page 15

15

4. The ECT as a Component in Many Evolutionary Scenarios

So far this paper has discussed the ECT in extremely general terms. In this final section I

will look at some specific models and programs of empirical work. I will discuss four

examples, each focused on understanding a specific type of cognition (or proto-cognition). I

suggest that the ECT is one component in many diverse scenarios that have been discussed

in connection with the evolution of cognition.

(i) Phenotypic plasticity

I have said several times in this paper that cognition, understood in my broad way, shades

off into other biological processes, especially those that use signals (from the environment

or elsewhere) to control adaptive responses. One important class of cases in the category I

have been calling “proto-cognitive” is the phenomenon of phenotypic plasticity, especially in

plants.

In the paradigm cases, phenotypic plasticity is a phenomenon in which a single plant

genotype can produce variety of forms (phenotypes) or can take a variety of developmental

paths, where the “choice” is determined by an environmental cue transduced by the plant

(Bradshaw 1965, Sultan 1987, Schlichting and Pigliucci 1998). A plant might have a wet-

environment and a dry-environment phenotype, for example, or might alter its form

according to altitude and accompanying climatic conditions, as in the classic experiments of

Clausen, Keck and Hiesey (1948). So in these cases the plant has some mechanism for

transducing an environmental cue, and of controlling development as a function of the state

of the cue. The cue, the plant phenotype, and the environmental variables that are being

adapted to might be discrete or continuous. (When the organism’s response is a discrete

choice this is sometimes called “polyphenism,” but I will not make that terminological

distinction here.)

No nervous system is involved in these cases, and in general it is growth and

development, rather than behavior, that is being controlled. But this type of phenomenon is a

useful “zero order” case for discussions of models of adaptive response to environmental

conditions. There are formal similarities between these capacities and cases of real

cognition. Indeed, in the 1990s two mathematically identical evolutionary models were

published independently (Moran 1992, Sober 1994). One was presented as a model of the

advantages of learning (Sober), while the other was presented as a model of the advantages

of plastic control of development (Moran).

How do the models and theoretical discussions look? Let us be very abstract.

Assume that an organism confronts an environment which has a range of alternative

possible states. The organism itself has a range of possible developmental options. The

________________________________________

Page 16

16

alternative environmental states have consequences for the organism’s chances of surviving

and reproducing, and the best developmental option for one environmental state is not the

best choice for another. The organism receives imperfect information about the actual state

of the environment, as a consequence of correlations between environmental conditions

which matter to it and environmental conditions which directly affect its periphery. There are

several ways to respond to the problem. One way is to be able to buffer out the

environmental variation — perhaps by being big, or strong, or restricting exposure in some

way. Another way is to adapt to the most common or the most critically important

environmental state. But yet another way is to use a flexible strategy — to use environmental

information to determine the organism’s phenotype in accordance with how the environment

is perceived to be.

For example, Drew Harvell (1986) investigates defences against predators produced

by colonial marine invertebrate animals called “bryozoans,” or sea moss. The bryozoans

Harvell studies are able to detect the presence of predatory sea slugs, making use of a water-

borne chemical cue. When sea slugs are around, the bryozoans produce spines. The spines

have been shown to effectively reduce predation, but also to incur a significant cost in terms

of growth, so they are detrimental when sea slugs are not around.

Here we have a rudimentary form of perception. The bryozoans show sensitivity to an

environmental cue which is not itself practically important, but which carries information

about a more important state of the world, the presence of predators. The organism use a

cue to produce an adaptive response to a more important “distal” environmental state.

In cases like these, a complete explanation for the organism evolving a “smart” or

proto-cognitive capacity includes a description of the problem posed by environmental

complexity — the fact that predators are sometimes, but not always, present. But the

explanation includes much more as well. We need to also know the reasons for a proto-

cognitive response being favored over buffering, adaptation to the most common condition,

or some other “dumb” strategy.

When will the proto-cognitive strategy be favored? There is no simple answer, but

many models developed by biologists and others can be pieced together to give a partial

answer (Godfrey-Smith 1996, chapters 7-9). Some parts of the story are intuitive. To use a

proto-cognitive strategy, the organism needs a suitable signal from the environment. If there

is no way of tracking the relevant states of the environment, it is better to produce a single

“cover-all” phenotype, or to adapt to the single most common or important environmental

state. Parts of information theory and signal detection theory can be used to describe exactly

what sorts of properties an environmental cue must have, in order for it to be worth using.

When do you want to choose a flexible strategy over an inflexible one? Only when your

________________________________________

Page 17

17

environmental cue is good enough so you don’t make too many of the wrong kinds of

errors. Even you can track the world with some reliability, if one type of wrong decision is

sufficiently disastrous, it may be best never to behave in a way which risks this error.

Principles like those are close to common sense. But these models also have a number of

more subtle features. For example, it can matter a great deal how payoffs from individual

encounters or “trials” are related to each other in their effects on overall fitness — whether

payoffs are summed or multiplied (Levins 1968, Seger and Brockman 1987). The principles

discussed by these models of plasticity cast light on both proto-cognition and genuine

cognition as well. The models describe the first step towards cognition — opting for a

flexible response to a heterogeneous environment.

(ii) The evolution of associative learning.

The second example I will discuss is a computer simulation of the evolution of associative

learning, due to Todd and Miller (1991). This simulation explores the evolution of the

architectural properties of simple networks of neurons, using what is known as the “genetic

algorithm.” The aim is to see when evolution will select for organisms that exhibit one of the

simplest kinds of learning — classical conditioning.

The neural networks can usefully be imagined as embodied in simple marine

animals which are born in the open sea, but which settle down to an immobile life feeding

on passing food particles. Once an individual has settled, its only problem is to decide

whether to feed or not feed, when presented with each item of possible food. The

environment contains both food and also inedible or poisonous particles, in equal

proportions. When food is eaten the organism gains an energetic benefit, and when poison

is eaten the organism pays a cost, though the error is not fatal.

Particles of possible food have two sorts of properties that the organisms can

perceive — color and smell. Food smells sweet and poison smells sour, but in this turbulent

environment smells can mislead. The probability of a sweet smell, given the presence of

food, is 0.75. The probability of a sour smell, given poison, is also 0.75.

The color of food is not affected by turbulence, but color is unpredictable in a

different respect. In half of the population’s environment food is red and poison is green,

but in the other half the colors are reversed. Within each of these two micro-environments,

color is 100% reliable.

Each generation contains a large number of individuals of different types, which

settle at random in the two different micro-environments. At the end of a fixed period they

reproduce (sexually) according to their accumulated fitness, with the possibility of mutation

________________________________________

Page 18

18

and recombination of genes. The new generation then floats about and settles in the

environment at random and the cycle begins again.

The neural networks placed in this scenario are constrained to have three “units,” or

nodes, only. These nodes are like idealized nerve cells. Despite these limited resources there

are lots of possibilities for the networks’ architectures, and these architectures evolve in the

model by natural selection across generations. Units can function as input devices of

various kinds (red-detectors, green-detectors, sweet-detectors or sour-detectors). There is

just one type of output unit (eating), and a “hidden” unit, which mediates between a detector

and a motor unit, is also a possibility. The range of units an individual has, and which ones

are connected to which, are determined by its genetic make-up.

Connections between units can be hard-wired with either an excitatory or inhibitory

one-way connection, or they can be plastic and altered by the individual’s experience. If the

genotype specifies a plastic connection, then the connection is shaped over time by what is

known as a “Hebbian” learning rule. If those two units tend to fire at the same time in the

individual’s experience, they acquire a positive connection between them — one unit comes

to have a (one-way) excitatory connection to the other. If they do not tend to fire together,

the connection becomes negative or inhibitory. The question the model is intended to

address is: when and in what ways will individuals with the ability to learn evolve in the

population? The question is interesting because Hebbian learning is discussed a good deal

by neuroscientists, and they hunt for Hebbian learning in the synapses of the brain. But

from the evolutionary point of view, it is often not obvious what use Hebbian learning has.

If two neurons tend to fire together, what is the point of also making one excite the other?

At the start of a “run” of the Todd and Miller simulation, the population consists of

randomly configured individuals, most of which do not fare well. For example, some will

not have a motor unit and will never eat, or will have a motor unit connected to an input unit

which has the wrong setting — it might tell the organism to always eat when the present food

particle smells sour. Another type of miswiring might be called “the academic.” An

individual can have two input units and a motor unit, but only learnable connections between

all three units, connections which are initially set at zero. Suppose such a creature lands in a

patch where food is red. Then it will learn the statistical association between redness and a

sweet smell — the red-color input unit will tend to be on at the same time as the sweet-smell

unit. But nothing is inducing the individual to eat. The motor unit will never be turned on,

and its knowledge of the world will not do the individual any good, as far as nutrition is

concerned.

Two kinds of wiring do work well for the organism though. One has a fixed positive

connection between a sweet-smell sensor and a motor unit, and nothing else which

________________________________________

Page 19

19

influences behavior. This organism will generally eat when there is food present — in the

present case, it will make the right decision 75% of the time. After a short period, these

individuals tend to proliferate in the population.

The best possible wiring is a variant on this one, which has a fixed connection

between a sweetness sensor and a motor unit (as above), but also a learnable connection

between a color sensor and the motor unit. From the start this individual will tend to eat

when there is food, as the smell sensor is controlling the motor unit. But in addition there

will be a correlation between eating and some particular state of the color sensor. If the

micro-environment is a red-food one, then when the organism eats it will also tend to be

seeing red. This correlation establishes a connection between the color sensor and the motor

unit, and (given the right initial settings) this connection will eventually be strong enough to

control the motor unit by itself. Then the eating behavior will be controlled by a 100%

reliable cue for the remainder of the individual’s life. Typical runs of the simulation begin

with the fairly rapid evolution of the simple, hard-wired smell-guided networks, and some

time afterwards learners appear and take over.

This simulation illustrates the advantages associated with two kinds of behavioral

complexity. Consider first the contest between individuals which always eat every particle

that drifts by, and individuals which use smell as a cue. If everything in the environment was

food, there would be little point in controlling behavior with perception, especially with an

only partially reliable environmental cue. But the environment used by Todd and Miller is

one where food and poison drift by with equal frequencies. This is one type of

environmental complexity, and it has the consequence that a permanently-eating architecture

will have low fitness when compared to an individual that uses smell as a cue. A different

type of environmental complexity, and a different reliability relationship between a cue and

the world, explains the evolution of learning. The total environment in which these

organisms live is spatially heterogeneous — in half the environment food is red and in the

other half food is green. If food was red in the whole environment, it would not be worth

taking the time to learn that food is red, and a network with a red-detector hard-wired to the

output unit would be optimal. But Todd and Miller use an environment which is

heterogeneous in this respect as well. And although the color of food is not predictable in

advance, the past experience of an individual is a good guide to the future. That is what is

needed for learning to be more useful than an inflexible behavioral program.

(iii) Spatial memory and cognitive maps

For a mobile animal, one very important kind of environmental heterogeneity is

heterogeneity in the distribution of resources, dangers and other factors in space. Spatial

________________________________________

Page 20

20

structure plays a very different role for a plant, of course, or for an animal like a clam which

does not move around the world. But once an organism is on the move, as most terrestrial

animals must sometimes be, spatial structure in the environment is of prime importance.

Here as in general, some of this environmental heterogeneity can be dealt with by

various forms of buffering. But evolutionary responses to the problem of dealing with space

have produced some impressive and complicated forms of cognition, even in small and

otherwise behaviorally simple animals. The mechanisms associated with bee dances, which

direct workers from the hive to sources of food, are one famous example, but it has turned

out that bees as individuals also show good spatial skills. They can learn to reliably

associate a source of food with either single landmarks or with geometrical structure in a set

of landmarks. In recent years a lot of attention has been directed on spatial memory in food

storing birds, such as the Clark’s nutcracker in the south-western US, and marsh tits in

England. Clark’s nutcrackers hide thousands of pine seeds as a food source for the winter.

It appears that these birds have specialized spatial memory abilities which do not extrapolate

(as far as has been determined) to superiority over other birds in non-spatial memory tasks.

(See Roberts 1998 chapter 7 for the bee and bird examples in this paragraph.)

Back in 1948, E. C. Tolman suggested that both animals and humans make their

way through space by using “cognitive maps,” or rich internal representations of spatial

structure in the environment. After initial controversy and some decades of neglect during

the heyday of strict forms of behaviorism, the concept of a cognitive map is again being

used by ethologists and comparative psychologists (Tolman 1948, Thinus-Blanc 1988,

Roberts 1998). The concept of a cognitive map is controversial in a number of respects.

First, there is a good deal of vagueness and ambiguity in how it is applied by different

researchers (Bennett 1996). Some use the term to refer specifically to postulated internal

structures which work in psychological processing in ways reminiscent of ordinary, external

maps. In this narrow sense, the hypothesis that an animal has a “cognitive map” requires, at

a minimum, a capacity to devise novel detours and shortcuts in response to obstruction of

more familiar paths. But the term is sometimes used more broadly, to refer to almost any

kind of spatial memory. Tolman, for example, distinguished “strip maps” and

“comprehensive maps” within the more general category of cognitive maps. Strip maps

represent only a path to a goal; they are dependent on the starting point of the animal.

Comprehensive maps are richer representations of the overall spatial structure in some

domain, so they can used despite variation in starting points, new obstacles and so on. But it

can be argued that the sort of behavior associated with “strip maps” is easily explained

without talking of inner “maps” at all; the animal is just executing a sequence of behaviors

________________________________________

Page 21

21

in response to a set of cues or landmarks. Some sophistication in memory is clearly

involved, but there is no need to postulate an inner map-like structure.

However, there are experimental results which do justify a richer interpretation of

some animals’ inner processing of spatial information. A simple and striking case is found

in an experiment by Tolman and Honzik (1930). Rats were trained in a maze that has three

different paths to a single supply of food. Path 1 is shorter than path 2, and path 2 is shorter

than path 3. The rats were easily able to learn to prefer the best available path. After a few

days of training, path 1 was almost always chosen first. If path 1 was blocked (at an early

point), they would go back and take path 2. If path 2 was also blocked, they would settle for

path 3. So far, this only shows a fairly routine (but very useful) type of reinforced learning.

The impressive behavior resulted when path 1 was blocked in a novel way for the first time.

Path 1 has its final section in common with path 2, but path 3 reaches the food

independently of this common section. So when this final part of path 1 is blocked, that has

the effect of making path 2 useless as well. What will the rats do when path 1 is blocked in

this novel way? Their history of conditioning has taught them that when path 1 is blocked,

path 2 is the next choice. But if the rats are smart enough to realize the consequences of this

novel way of blocking path 1, they should choose path 3 directly and not waste time on path

2. In Tolman and Honzik’s experiment, a large majority of rats, on encountering the novel

obstruction on the later part of path 1 for the first time, returned to the junction point of the

three paths and immediately chose path 3. They did not follow their failure on path 1 with

an attempt at path 2; they had somehow been able to represent the new obstacle as rendering

path 2 useless as well.

Tolman and Honzik interpreted this as showing “insight” on the part of the rats,

following the gestalt psychologist Köhler, who had found similar results with chimps.

Tolman did not, for some reason, use this experiment in the famous 1948 discussion which

introduced the concept of a “cognitive map.” Instead he used results which, to my mind,

were a good deal less convincing than his 1930 “insight” experiment. (Thinus-Blanc 1988

erroneously reports Tolman 1948 as actually discussing the “insight” experiment in support

of the cognitive map concept, an interesting case of wishful thinking, or post-hoc

improvement of Tolman’s paper!) Perhaps Tolman did not think of the “insight” experiment

as showing specifically spatial cognitive skills, but a more general capacity to draw

conclusions and reason beyond the immediate lessons of conditioning. However the

“cognitive map” concept is used, this 1930 experiment does appear to show a capacity to

construct and manipulate some sort of internal model of spatial structure in the environment.

In qualification of this, I should note that Tolman and Honzik found it quite tricky to devise

a maze in which most of their rats would consistently show this spatial “insight.” For

________________________________________

Page 22

22

example, insight was consistently shown only when the maze was made of elevated tracks,

not tunnels.

Another kind of experiment designed to investigate innovative behavior based on

representation of spatial structure studies the use of short cuts. Both dogs and chimps can

be shown hidden pieces of food in an environment, with the order of their exposure to the

food corresponding to an inefficient path from food item to food item. Once released, the

animals in both cases are able to find the food items and move from item to item using a

path that is new and more direct and efficient than the one they were trained on. (In effect,

the chimps do a reasonable job at what mathematicians refer to as a “traveling salesman

problem.”) Here again, the behavior produced does not correspond to any motor routine or

action pattern that the animal was trained on, and these experiments also control for such

possibilities as olfactory detection of the food (Roberts 1998, Menzel 1997).

All of these experiments are associated with some controversy. For example, it can

be argued that to the extent that animals are simply moving from one memorized landmark

(or point specified by its relation to a set of landmarks) to another, there is justification for

attributing memory to the animal but not a cognitive map (Bennett 1996). Menzel (1997)

found deviations from “traveling salesman” optimality when macaques were tested on

whether they chose the optimal path from item to item, or simply went towards the closest

piece of food at each decision point. But it is noteworthy that the best of this work, such as

the Tolman and Honzik “insight” experiment, does involve behavior which appears to justify

the postulation of internal representation of the world, and fairly complex use of the

representations in guiding behavior, without the animal having a capacity for public

language. There is a tradition in philosophy of denying that any animal which lacks

language can properly be said to “think” or “represent.” In recent years, Davidson (1975)

has been the most influential defender of this view. It was held in a different form by Dewey

(1929) and is often associated with Wittgenstein (1953) and his followers. Such views

struggle to make sense of the skills in animal path-choice discussed above. Whether or not

these animals are able to “represent” the world in the richest, most philosophically loaded

sense of “represent,” they do seem to be doing some kind of representation or mapping of

the spatial structure of their environment (see also Allen and Bekoff 1997). Problems and

opportunities associated with spatial structure in environments have apparently generated a

range of sophisticated cognitive skills.

(iv) Social intelligence models

It has usually been thought not too hard to explain the evolution of such capacities as

learning (Example (ii) above) and the ability to represent spatial structure in an environment

________________________________________

Page 23

23

(Example (iii)). But it is a different matter to explain the highly developed and distinctive

cognitive capacities of human beings and non-human primates. Over 20 years ago, Nicholas

Humphrey suggested that primates seem to have too much brain-power to be explained by

the demands of such activities as foraging for food. He suggested that the problems

primates are using this intelligence to deal with stem from the social complexity of their

environments (Humphrey 1976). In particular, much primate life is concerned with the

formation and maintenance of alliances, the policing of dominance hierarchies, and a variety

of other social tasks that involve a mixture of competition and cooperation. Humphrey’s

suggestion (which had been partially anticipated by others) was that high primate

intelligence evolved in response to the problem of dealing with this kind of complexity. As

each individual primate comprises part of the environment for the others in a population, we

have the ingredients here for a process of feedback, in which each increase in intelligence

produced by evolution adds to the complexity of the social environment that individuals

face.

This idea has come to be known as the “Machiavellian intelligence hypothesis”

(Byrne and Whiten 1988, Whiten and Byrne 1997). Here I use the term “social

intelligence” rather than “Machiavellian intelligence.” As Whiten and Byrne say, the term

“Machiavellian” should not be taken to suggest that all the behaviors involved in the

hypothesis are manipulative and deceitful. They mean to include “cunning cooperation”

which contributes to individual reproductive success (Whiten and Byrne 1997 pp. 12-13).

Whiten and Byrne do mean to exclude, however, any suggestion of natural selection

operating at the level of groups rather than individuals — they mean to exclude the idea that

intelligent cooperation could evolve “for the good of the group.” As I do think the term

“Machiavellian intelligence” continues to mislead, suggesting the darker side of social

behavior, and I also do not want to rule out some role for group selection in these processes,

I prefer to use the term “social intelligence hypothesis.” Like Gigerenzer (1997), I reserve

“Machiavellian” as a narrower term, specifically for behaviors involving exploitation rather

than cooperation.

The social intelligence hypothesis is compatible with the ECT; it is a specific

instance of the ECT which involves a special kind of environment. In the previous section I

argued that the ECT need not be understood in a way that requires the environment in

question to be independent of the population that is evolving. In any social animal, a key

part of an individual’s environment is made up of the other members of the social group,

with all their behavioral capacities. The same is true to a lesser extent with many non-social

animals. In cases like these, the ECT describes one particular explanatory arrow within a

larger explanatory structure, a structure which links cognitive capacities and environmental

________________________________________

Page 24

24

complexity in a “coupled” way. At any given time, the individuals in a social population face

environmental complexity in the form of the behavioral patterns of the other local members

of the population. This environmental complexity may (or may not) give more intelligent

individuals an advantage over less intelligent individuals, by some specific measure of

intelligence relevant to the situation. If intelligence is favored, and if this kind of intelligence

is inherited, then over time intelligence will increase in the population. If the intelligent

individuals themselves display more complex patterns of behavior than others in the

population, then this increase in intelligence will in turn entail an increase in the complexity

of the social environment faced by later individuals. This process may or may not have a

“runaway,” positive-feedback character. One should not assume that a runaway process is

the only outcome. For example, other constraints and costs might start to assume a larger

role once the individuals reach a certain level of intelligence.

The social intelligence hypothesis has been developed in a number of different

specific versions. Stronger versions claim that social complexity has been the key factor in

producing the high levels of primate intelligence; weaker versions see this as one

explanatory factor which might work in conjunction with others. For example, some suggest

a role for special problems associated with foraging for the ripe fruits favored by primates –

it might be that primates require especially sophisticated cognitive maps of their (non-social)

environments. (Whiten and Byrne 1997 contains discussions of a range of alternatives to

the social intelligence hypothesis.) Different versions of the hypothesis also stress different

aspects of social living — direct competition between males for mates, dealing with

dominance hierarchies, cooperative foraging and so on.

As Gigerenzer (1997) notes, the social intelligence hypothesis is sometimes

associated with the suggestion that (i) the overall degrees of complexity in social and non-

social environments (or social and non-social aspects of an environment) can be compared,

and (ii) social environments are more complex. As I said back in section 3 of this paper, I

am a skeptic about the project of giving overall measures of complexity across

environments; any environment is complex in some respects and simple in others.

Gigerenzer is similarly skeptical about these complexity measures. But where Gigerenzer

appears to think that this problem makes it pointless to generalize about the role of

complexity in the evolution of cognition, I think the environmental complexity thesis is a

useful general principle despite the absence of a unitary scale of environmental complexity.

As I understand them, most but not all of the specific hypotheses discussed under

the general category of “social intelligence” can be seen as applications of the ECT. The

versions that do fall under the ECT are those that stress the role of cognition in dealing with

the behavioral complexity of other individuals within a social group. An example of a

________________________________________

Page 25

25

hypothesis in this general area that does not fall under the ECT is one version of the

“protean behavior” hypothesis, discussed by Geoffrey Miller (1997). Miller suggests that

animals like primates have been selected to be able to produce genuinely unpredictable

behavior. Unpredictable behavior has fairly obvious advantages when an animal is escaping

from predators, and more subtle advantages in situations involving conflict and bluffing, as

discussed in game theory. In those cases, producing unpredictable behavior itself requires

no special cognitive sophistication. But Miller also suggests that capacities for novel,

creative behavior will help individuals attract mates, especially in situations of female choice,

and these behavioral capacities are facilitated by a large brain. There are several ways in

which this scenario might work, but consider one case. Suppose there has been selection for

behavioral novelty in males, and suppose these novel mating displays succeed by taking

advantage of a general feature of perceptual and cognitive mechanisms — the fact that

novelty attracts attention. Then any evolved increase in cognitive sophistication due to this

process cannot be seen as an application of the ECT. A complex social environment might

be created by male behaviors in this case, but cognition is not being selected as a way of

dealing with complexity. On the other hand, if females are being strongly selected for the

capacity to see through all this behavioral noise and make adaptive choices, there will be

selection, compatible with the ECT, for cognitive sophistication in females.

All these “sex-specific” hypotheses about the advantages of cognition in primates

have a problem stemming from the basic similarity between male and female brains in the

most intelligent primates (as Richard Francis stressed to me). If selection only favors

elaborate cognition in one sex, then the other sex will have much of its brain treated by the

theory as a mere byproduct. Big brains are too expensive to be treated like male nipples; big

brains as byproducts would be analogous to peacock tails on peahens. If both sexes are

being selected to be smart, but in very different ways, then the problem posed by the lack of

obvious sexual dimorphism is more subtle and hard to assess. In any case, I introduce these

speculations about unpredictable behavior not to endorse them, but to illustrate the fact that

while the ECT is very broad, it does not trivially encompass any possible explanation for the

evolution of cognition. The ECT only covers cases where environmental complexity

(whether social or nonsocial) creates a problem (or an opportunity) for some type of

organism, and the problem leads to natural selection favoring individuals with an ability to

use cognition to coordinate behavior with the state of the environment.

In earlier discussions of social intelligence, there was sometimes the appearance of a

sharp “either-or” characteristic to the debates about social and non-social complexity; either

primates became smart for social reasons, or they became smart for reasons having to do

with non-social aspects of their ecology. But as Byrne (1997) argues, there is plenty of

________________________________________

Page 26

26

room for mixed explanations, in which a number of factors have a role. Byrne himself

posits three distinct evolutionary transitions in the evolution of intelligence in monkeys, apes

and humans. In this scenario, the evolutionary branch containing monkeys and apes

(haplorhines) became smarter than its relatives because of selection for social intelligence.

But the “Great Apes” (chimps, bonobos, humans, gorillas, orangutans) branched off from

this group and became smarter because of selection for “technical intelligence,” which

involves planning and sophisticated tool-use in activities such as foraging. And then the

branch that led to modern humans was perhaps again the subject of selection for social

intelligence, in part because of larger group size. This scenario is very speculative, as Byrne

stresses, but it provides a good example of a “mixed” story about the evolution of human

cognition. It would be a mistake to only pursue “pure” social intelligence hypotheses, out of

an overly strong attachment to explanatory simplicity.

In closing, I will try to give a more “fleshed out” summary of what the ECT claims.

Environmental Complexity Thesis (more detailed version):

The basic pattern found in the evolution of cognition is a pattern in which individual

organisms derive an advantage from cognitive capacities in their attempts to deal with

problems and opportunities posed by environmental complexity of various kinds. Cognitive

capacities confer this advantage by enabling organisms to coordinate their behavior with the

state of the environment. Cognition itself should be thought of as a diverse “tool-kit” of

capacities for behavioral control, including capacities for perception, internal representation

of the world, memory, learning, and decision-making. These capacities vary across different

types of organism and are not sharply distinguished from other biological capacities, some

of which have a “proto-cognitive” character. The “environment” referred to in the ECT

includes the social environment, and there are some reasons to believe that problems posed

by social complexity have been very important in the evolution of primate and human

intelligence. Many specific evolutionary scenarios that have been discussed as possible

explanations of particular cognitive capacities are instances of the ECT, or have the ECT as a

part.

*

*

*

*

*

________________________________________

Page 27

27

Acknowledgment

Thanks to Richard Francis, Lori Gruen and Kim Sterelny for discussions and

correspondence on these issues.

References

Attenborough, D. (1995). The Private Life of Plants. Princeton: Princeton University Press.

Allen, C. Bekoff, M. and G. Lauder (eds.) (1998). Nature’s Purposes: Analyses of Function

and Design in Biology. Cambridge MA: MIT Press.

Allen, C. and M. Bekoff (1997). Species of Mind. Cambridge: MIT Press.

Bennett, A. T. D. (1996). “Do Animals Have Cognitive Maps?” Journal of Experimental

Biology 199: 219-224.

Bonner, J. (1988). The Evolution of Complexity. Princeton; Princeton University Press.

Bradshaw, A. D. (1965). “Evolutionary Significance of Phenotypic Plasticity in Plants,”

Advances in Genetics 13: 115-55.

Byrne, R. W. and A. Whiten (eds.) (1988). Machiavellian Intelligence: Social Expertise and

the Evolution of Intellect in Monkeys, Apes and Humans. Oxford: Clarendon Press.

Byrne, R. W. (1997). “The Technical Intelligence Hypothesis: An Additional Evolutionary

Stimulus to Intelligence,” in Whiten and Byrne (1997) pp. 289-311.

Clausen, J., D. Keck, and W. M. Hiesey (1948). Experimental Studies on the Nature of

Plant Species III. Environmental Responses of Climatic Races of Achillea. Carnegie

Institution of Washington. Publication 581. Washington DC.

Davidson, D. (1975). “Thought and Talk,” reprinted in Essays on Truth and Interpretation.

Oxford: Oxford University Press, 1984, pp. 155-170.

Dewey, J. (1929). Experience and Nature (revised edition). New York: Dover, 1958.

Dretske, F. (1981). Knowledge and the Flow of Information. Cambridge MA: MIT Press.

Dretske, F. (1986). “Misrepresentation,” reprinted in Stich and Warfield (1994) pp. 157-73.

Francis, R. and G. W. Barlow (1993). “Social Control of Primary Sex Determination in the

Midas Cichlid,” Proceedings of the National Academy of Sciences, USA 90: 10673-

10675.

Gibson, J. J. (1966). The Senses Considered as Perceptual Systems. Boston: Houghton

Mifflin.

Gigerenzer, G. (1997). “The Modularity of Social Intelligence,” in Whiten and Byrne

(1997) pp. 264-288.

Godfrey-Smith, P. (1996). Complexity and the Function of Mind in Nature. Cambridge:

Cambridge University Press.

________________________________________

Page 28

28

Gould, S. J. and R. C. Lewontin (1979). “The Spandrels of San Marco and the Panglossian

Paradigm: a Critique of the Adaptationist Program,” Proceedings of the Royal Society,

London 205: 581-598.

Harvell. D. (1986). “The Ecology and Evolution of Inducible Defences in a Marine

Bryozoan: Cues, Costs, and Consequences”, American Naturalist 128: 810-23.

Humphrey, N. (1976). “The Social Function of Intellect,” Reprinted in Byrne and Whiten

(1988).

Krebs J. and N. Davies (1987). An Introduction to Behavioural Ecology. 2nd edition.

Oxford: Blackwell.

Levins, R. (1968). Evolution in Changing Environments. Princeton: Princeton University

Press.

Lewontin, R. C. (1983). “The Organism as the Subject and Object of Evolution,” reprinted

in R. Levins and R. C. Lewontin, The Dialectical Biologist. Cambridge MA: Harvard

University Press, 1985, pp. 85-106.

Lodish, H., D. Baltimore, A. Berk, S. L. Zipursky, P. Matsudaira and J. Darnell (1995).

Molecular Cell Biology, 3rd edition. New York: Freeman.

McShea, D. (1991). “Complexity and Evolution: What Everybody Knows,” Biology and

Philosophy 6: 303-24.

Menzel, C. R. (1997). “Primates’ Knowledge of their Natural Habitat,” in Whiten and Byrne

(1997) pp. 207-239.

Miller, G. (1997). “Protean Primates; The Evolution of Adaptive Unpredictability in

Competition and Courtship,” in Whiten and Byrne (1997) pp. 312-340.

Moran, N. (1992). “The Evolutionary Maintenance of Alternative Phenotypes,” American

Naturalist 139: 971-89.

Odling-Smee, F. J. (1988). “Niche-Constructing Phenotypes,” in H. C. Plotkin, (ed.)

(1988). The Role of Behavior in Evolution. Cambridge MA: MIT Press, pp. 73-132.

Povinelli, D. J. and J. G. Cant (1995). “Arboreal Climbing and the Evolution of Self-

Conception,” Quarterly Review of Biology 70: 393-421.

Roberts, W. A. (1998). Principles of Animal Cognition. Boston: McGraw Hill.

Schlichting, C. and M. Pigliucci (1998). Phenotypic Evolution: A Reaction Norm

Perspective. Sunderland MA: Sinauer.

Seger, J. and J. Brockman (1987). “What is Bet-Hedging?” Oxford Surveys in

Evolutionary Biology 4: 181-211.

Silvertown, J. and D. Gordon (1989). “A Framework for Plant Behavior,” Annual Review of

Ecology and Systematics 20: 349-366.

________________________________________

Page 29

29

Silvertown, J. and J. Lovett Doust (1993). Introduction to Plant Population Biology.

Oxford: Blackwell.

Sober, E. (1994). “The Adaptive Advantage of Learning Versus A Priori Prejudice,” in

From a Biological Point of View. Cambridge: Cambridge University Press, pp. 50-70.

Sterelny, K. (1995). “Basic Minds,” Philosophical Perspectives 9: 251-270.

Stich, S. P. and T. A. Warfield (eds.) (1994). Mental Representation: A Reader. Oxford:

Blackwell.

Sultan, S. (1987). “Evolutionary Implications of Phenotypic Plasticity in Plants,” in

Evolutionary Biology, Volume 21 (M. Hecht, B. Wallace and G.T. Prance, Eds.) New

York: Plenum, pp. 127-78.

Thinus-Blanc, C. (1988) “Animal Spatial Cognition,” in L. Weiskrantz (ed.) Thought

Without Language. Oxford: Clarendon Press, pp. 371-395.

Todd, P. M. and G. F. Miller (1991). “Exploring Adaptive Agency II: Simulating the

Evolution of Associative Learning,” in J.-A. Meyer and S. W. Wilson (eds.), From

Animals to Animats: Proceedings of the First International Conference on the

Simulation of Adaptive Behavior. Cambridge MA: MIT Press, 1991, pp. 306-15.

Tolman, E. C. (1948). “Cognitive Maps in Rats and Men,” Psychological Review 55: 189-

208.

Tolman, E. C. and T. H. Honzik (1930). “‘Insight’ in Rats,” University of California

Publications in Psychology 4: 257-275.

Whiten, A. and R. W. Byrne (eds.) (1997). Machiavellian Intelligence II: Extensions and

Evaluations. Cambridge; Cambridge University Press.

Wittgenstein, L. (1953). Philosophical Investigations. Translated by G. Anscombe. New

York: Macmillan.