by Richard Dawkins
In the pioneering days of radio, my grandfather’s job was to lecture to young engineers who were joining Marconi’s company. To illustrate that any complex wave form can be broken down into summed simple waves of different frequencies (important in both radio and acoustics), he took wheels of different diameters and attached them with pistons to a clothesline. When the wheels went round, the clothesline was jerked up and down, causing waves of movement to snake along it. The wriggling clothesline was a model of a radio wave, giving the students a more vivid picture of wave summation than mathematical equations could ever have done.
This was my first exposure to models in the ordinary scientific sense: a model resembles the real thing in some important respects, although it may not necessarily look, to the human eye, like a replica of the real thing. A child’s train set is a model, but so also is a railway timetable. Engineers build airplane models for testing in wind tunnels; weather forecasters make use of very elaborate, dynamic computer models of the earth’s weather.
Biologists, too, use models to express what they think is going on inside organisms and in ecosystems. But I want to say something altogether more radical. An animal is a model. Any organism is a model of the world in which it lives. One way to understand this is to imagine a zoologist presented with the body of an animal she has never seen before. If allowed to examine and dissect the body in sufficient detail, a good zoologist should be able to reconstruct almost everything about the world in which the animal lived. To be more precise, she would be reconstructing the worlds in which the animal’s ancestors lived. I say this because an animal can never be strictly adapted to its present environment. It is always adapted to a sum of past environments in which its ancestors survived. More strictly still, the sum is a weighted sum, with the weights diminishing as we go back in time.
All these claims rest upon the Darwinian assumption that animal bodies are largely shaped by natural selection. If Darwin’s theory is correct, an animal is the inheritor of attributes that enabled its ancestors to be ancestors. If they hadn’t had those successful attributes, they would have been not ancestors but the childless rivals of ancestors.
So, what are the attributes that make for success as an ancestor, the attributes that we should expect to find in the body of our animal when we inspect it? The answer is anything that helps the individual animal to survive and reproduce in its own environment—not just one or two attributes, but hundreds, thousands of them. This is why, if you present an animal’s body, even a new species previously unknown to science, to a knowledgeable zoologist, she should be able to “read’ its body and tell you what kind of environment it inhabited: desert, rain forest, arctic tundra, temperate woodland, or coral reef. She should also be able to tell you, by reading its teeth and its guts, what it fed on. Flat, millstone teeth indicate a herbivore; sharp, shearing teeth, a carnivore. Long intestines with complicated blind alleys indicate a herbivore; short, simple guts suggest a carnivore. By reading the animal’s feet and its eyes and other sense organs, the zoologist should be able to tell how it found its food. By reading its stripes or flashes, its horns, antlers, or crests, she should be able to tell something about its social and sex life.
But zoological science has a long way to go. By reading the body of a newly discovered species, we can now come up with only a rough verdict about its probable habitat and way of life—rough in the same way as a pre-computer weather forecast was rough. The zoology of the future will computerize many more measurements of the anatomy and chemistry of the animal being read. More importantly, it will not take the teeth, guts, and chemistry of the stomach separately. It will perfect techniques of combining sources of information and analyzing their interactions, resulting in inferences of enormous power The computer, incorporating every thing that is known about the body of the strange animal, will construct a model of the animal’s world to rival any model of die earth’s weather. This. it seems to me, is tantamount to saying that the animal, any animal, is a model of its own world or the world of its ancestors. And its genes are a coded description of the worlds in which its ancestors survived.
In a few cases, an animal’s body is a model of its world in the literal sense. A stick insect lives in a world of twigs, and its body is a precise replica of a twig. A fawn’s pelage is a model of the dappled pattern of sunlight filtered through trees onto the woodland floor. A peppered moth is a model of lichen on the tree bark that is the moth’s world when at rest. But models, as we have seen, do not stop at replicas.
Models can be static or dynamic, and sometimes both. A railway timetable is a static model, while a weather model in a computer is dynamic: it is continually—in advanced systems continuously—being updated by new readings from around the world. (Even with the help of sophisticated computers and updated information from satellites, balloons, ships, planes, and weather stations, accurate forecasting is possible only for a few days ahead, at best.) Some aspects of an animal’s body are a static model of its world—the millstone slab of a horse’s tooth, for instance. Other aspects are dynamic. Sometimes the change is slow. A Dartmoor pony grows a shaggy coat in winter and sheds it in summer. The zoologist presented with a pony’s pelt can read not only the kind of place it inhabited but also the season of the year in which it was caught. Many animals of high northern latitudes, such as arctic foxes, snowshoe hares, and ptarmigans, are white in winter and brownish in summer.
But animals are dynamic on much faster time scales as well—time scales of seconds and fractions of seconds. These are the time scales of behavior, which can be seen as high-speed dynamic modeling of the environment. Think of a herring gull adroitly riding a sea cliff’s upcurrents. It may not be flapping its wings, but this doesn’t mean that its wing muscles are idle. They, along with the tail muscles, are constantly making tiny adjustments, sensitively fine-tuning the bird’s flight surfaces to every nuance, every eddy, of the air around it. If we fed information about the state of all these muscles into a computer, from moment to moment the computer could in principle reconstruct every detail of the air currents through which the bird was gliding. It would assume that the bird was well designed to glide, and on that assumption construct a model of the air around the bird. Again, it would be a model in the same sense as the weather forecaster’s. Both are continuously revised by new data. Both can be extrapolated to predict the future. The weather model predicts tomorrow’s weather; the gull model could “advise” the bird on the anticipatory adjustments that it should make to its wing and tail muscles in order to glide on into the next second.
Even if no human programmer has yet constructed a computer model that could advise gulls on how to adjust their wing and tail muscles, just such a model is almost certainly being run continuously in the brain of the gull and of every other bird in flight. Similar models, preprogrammed in outline by genes and past experience, and continuously updated by new sense data from millisecond to millisecond, are running inside the skull of every swimming fish, every galloping horse, every echo-ranging bat.
I should not wish, by using the metaphor of the computer, to imply that brains work like modern digital electronic computers. They probably don’t. The principle of getting information about the real world by simulating it internally is what I want to emphasize, and the digital electronic computer happens to be a familiar and powerful tool for simulation. But there are other conceivable tools that are not digital and not electronic, and the brain might well resemble them more. Before digital computers became readily available, engineers used a variety of devices to simulate reality. My grandfather’s clothesline was a simple example. Other such “analogue” devices were, and sometimes still are, used to solve serious mathematical problems. A mathematical function, for example, can be represented as a curve of a particular shape.
As recently as World War II, differential equations were solved by elaborate mechanical analogue computers consisting of concatenations of mathematically curved cams and rods sliding over one another. Even today, the simplest way of solving that mathematician’s chestnut—the “traveling salesman” problem (planning an optimal route for a salesman who has to visit a particular list of cities)—is by knotting bits of string together.
The same is true of some other optimization problems. The brain obviously doesn’t tie knots in string, but the psychologist and philosopher Kenneth Craik and the biologist John Maynard Smith have conjectured (not in these words) that brain models have more in common with knotted string than with digital computers. For our purposes here, it doesn’t matter. It is sufficient that the brain makes simulation models of the outside world. I think in terms of digital electronic computers because I am familiar with them, but neither their digitalness nor their electronicness is important to the analogy.
Can an animal’s mental model of its world free-run into the future and so simulate future events, as does the computer model of the world’s weather? Suppose we set up an experiment. Find a steep cliff in a mountainous area of Ethiopia inhabited by hamadryas baboons and place a plank so that it sticks out over the edge of the precipice, with a banana on its far tip. The center of gravity of the plank is just on the safe side of the edge, so that it does not topple into the gorge below, but if a monkey were to venture out to the end of the plank, it would be enough to tip the balance. Now we hide and watch what the monkeys do. They are clearly interested in the banana, but they do not venture out along the plank to get it. Why not?
We can imagine three stories, any of which might be true, to account, for the baboons’ prudence. In all three stories the cautious behavior results from a kind of trial and error, but of three different kinds. According to the first story, the baboons have an “instinctive” fear of precipitous heights. This fear has been built into their brains directly by natural selection. Their ancestors’ contemporaries that did not possess a genetic tendency to fear cliffs failed to become ancestors because they got killed. Consequently, since modem baboons are all descended, by definition, from successful ancestors, they have inherited the genetic predisposition to fear cliffs. There is indeed some experimental evidence that the newly born young of a variety of species have an innate fear of heights. In “visual-cliff” experiments, a sheet of glass lies on a table and projects over the edge. Newborn animals are then placed on the glass near the edge, to see whether they shy away from the edge or are indifferent to it. The first story, then, involves trial and error of the crudest and most drastic kind: Darwinian natural selection dicing with ancestral life and death. We can call this the Ancestral Fear story.
The second story talks about the past experiences of the individual baboons. Each young baboon, as it grows up, experiences falling. Most likely, it will have enough encounters with small cliffs to learn that falls can be painful. (If it falls down a huge cliff, of course, the experience is its last.) Pain, in trial-and-error learning, is the analogue of death in natural selection. Natural selection has built brains with the capacity to experience as pain those very sensations that, in a stronger dose, would lead to the animal’s death. Pain is not only the analogue of death; it is also a kind of symbolic substitute for death if we think in terms of an analogy between learning and natural selection. Baboons build up in their brains, through experience of the pain of falling down small cliffs (perhaps through experience that the bigger the cliff, the worse the pain), a tendency to avoid cliffs. This is the second story, the Painful Experience story, of how the baboons have come to resist their natural tendency to rush out along the plank to seize the banana.
The third story is the one this is all leading up to. According to this story, each baboon has a model of the situation in its head, a virtual reality simulation of the cliff, the plank, and the banana, and it can run the simulation program on into the future. Just as the arcade computer simulates the racing car passing a tree, the baboon’s computer simulates his body advancing toward the banana, the model plank teetering, then toppling and crashing into the simulated abyss. The brain simulates it all and evaluates the results of the computer run. And that, according to our Simulated Experience story, is why the baboon doesn’t venture out in reality. It is a trial-and-error story, just like the Ancestral Fear and Painful Experience stories, but this time it is trial and error in the head, not in reality. Obviously, trial and error in the head, if you have a powerful enough computer there to do it, is preferable to trial and error in earnest.
Now, as you read these stories, I have little doubt that you had an imaginary picture of the scene. You “saw” the cliff, you “saw” the plank, and you “saw” the baboons. The details of all our imaginary pictures are, no doubt, very different. But we all set up a simulation of the scene, which was adequate the task of predicting a baboon’s future. We all know, from the inside, what it is like to run a simulation of the world in our heads. We call it imagination, and we use it all the time to steer our decisions in wise and prudent directions.
The experiment with the baboons and the banana has not been done. If it were, could the results tell us which of our three stories was true, or whether the truth was some combination? If the Painful Experience story were true, we should be able to find out by looking at the behavior of young or inexperienced baboons. One who had been sheltered all his life from falls should prove fearless when eventually confronted with an edge. If such a naive baboon turned out in fact to be fearful, this would still leave the other two stories open. He have inherited ancestral fear or he might have a vivid imagination. We could try to decide the issue by a further experiment. Say we place a heavy rock on the near end of the plank. Now we humans, at least, can see from our own mental simulation that it is safe to venture along the plank: the rock is obviously a secure counterbalance.
What would the baboons do? I don’t know. But I do know that, however certain I was from my mental model that the rock would be a staunch counterweight, I wouldn’t go out along the plank, not for a crock of gold. I just can’t take heights. The Ancestral Fear story sounds pretty plausible to me. What is more, so powerful is this fear that it enters into my Simulated Experience. When I imagine the scene, I experience a frisson of fear up my spine, however vividly I am able to simulate a ten-ton rock firmly clamped down on the plank. Since I know that all three stories are true for me, I could easily believe the same of baboons.
The imagination, the capacity to simulate things that are not (yet) in the world, is a natural progression from the capacity to simulate things that are in the world. The weather model is continually updated by information from weather ships and weather stations. To this extent it is a simulation of conditions as they really are. Whether or not it was originally designed to run on into the future, its ability to do so, to simulate things not only as they are but as they may turn out to be, is a natural, almost inevitable consequence of its being a model at all. An economist’s computer model of the economy of Britain is, so far, a model of things as they are and have been. The program hardly needs to be modified to take that extra step into the simulated future, to project probable future trends in the gross national product, the currency, and the balance of payments.
So it was in the evolution of nervous systems. Natural selection built in the capacity to simulate the world as it is because this was necessary in order to perceive the world. You cannot see that two-dimensional patterns of lines on two retinas amount to a single solid cube unless you simulate, in your brain, a model of the cube. Having built in the capacity to simulate models of thing as they are, natural selection found that it was but a short step to simulate things as they are not quite yet—to simulate the future. This turned out to have valuable consequences, for it enabled animals to benefit from “experience,” not trial-and-error experience in their own past or in the life and death experience of their ancestors, but vicarious experience in the safe interior of the skull.
And once natural selection had built brains capable of simulating slight departures from reality into the imagined future, a further capacity automatically flowered. Now it was but another short step to the wilder reaches of imagination revealed in dreams and in art, an escape from mundane reality that has no obvious limits.