Protean Primates: The Evolution of Adaptive Unpredictability in Competition and Courtship



Download 112.27 Kb.
Page1/3
Date conversion12.05.2016
Size112.27 Kb.
  1   2   3
Protean Primates:

The Evolution of Adaptive Unpredictability

in Competition and Courtship

Geoffrey F. Miller

ESRC Research Centre for

Economic Learning and Social Evolution (ELSE)

University College London

Gower St., London WC1E 6BT, England



geoffrey.miller@ucl.ac.uk
Published as:
Miller, G. F. (1997). Protean primates: The evolution of adaptive unpredictability in competition and courtship. In A. Whiten & R. W. Byrne (Eds.), Machiavellian Intelligence II: Extensions and evaluations (pp. 312-340). Cambridge University Press.

Abstract
Machiavellian intelligence evolves because it lets primates predict and manipulate each others’ behavior. But game theory suggests that evolution will not stop there: predictive capacities tend to select for unpredictability in counter-strategies, just as many competitive games favor “mixed” (stochastic) strategies. For example, prey animals often evolve “protean” (adaptively unpredictable) evasion behavior to foil the predictive pursuit tactics used by their predators. The same adaptive logic should apply to more abstract social tactics, but protean social behavior remains overlooked in primatology and psychology, because complex order rather than useful chaos has been considered the hallmark of evolved adaptations. This chapter reviews the notions of psychological selection from evolutionary theory, mixed strategies from game theory, and protean behavior from behavioral ecology. It then presents six possible types of social proteanism in primates, and develops a model of how sexual selection through mate choice could have elaborated primate social proteanism into human creative intelligence.
1 Introduction: Unpredictability, animacy, and psychology
Nature cloaks herself in many modes of unpredictability. Science advances in part by recognizing and distinguishing these modes (see Kruger, Gigerenzer, & Morgan, 1987). Statistical mechanics modelled the complexity of fluids using stochastic principles. Quantum theory accepted the noisiness of elementary particles. Chaos theory revealed that many dynamical systems show extreme sensitivity to initial conditions. Evolutionary theory showed how random variation plus cumulative selection could yield organic complexity. Such progress in physics and biology has not been matched by psychology. Although unpredictability is a hallmark of animal behavior, it has been the bane of the behavioral sciences. Variation in behavior, whether across species, situation, space, or time, has usually been attributed either to adaptation or to error, with adaptation narrowly defined as systematic (if complex) correspondence between environmental conditions and behavioral tactics, and error narrowly defined as raw behavioral noise. Psychology’s favorite statistical shibboleth, analysis of variance, assumes that behavior can be explained by the interaction of environmental determinants and random, nonadaptive noise.
This chapter examines a type of behavior that is both adaptive and noisy, both functional and unpredictable, and that has therefore been overlooked by most behavioral scientists. The difficulty of predicting animal behavior may be much more than a side-effect of the complexity of animal brains. Rather, the unpredictability may result from those brains having been selected over evolutionary history to baffle and surprise all of the would-be psychologists who preceded us. To appreciate why psychology is hard, we have to stop thinking of brains as physical systems full of quantum noise and chaos, or as computational systems full of informational noise and software bugs. We have to start thinking of brains as biological systems that evolved to generate certain kinds of adaptive unpredictability under certain conditions of competition and courtship.

2 Genuine unpredictability is the best defense against predictive mind-reading
The Machiavellian Intelligence hypothesis suggests that apes and humans have evolved special cognitive adaptations for predicting and manipulating the behavior of other individuals (Humphrey, 1976; Byrne & Whiten, 1988; Whiten & Byrne, 1988). These adaptations are postulated to include a “Theory of Mind” module for attributing beliefs and desires to others, to better predict their behavior. (see Leslie, 1994; Baron-Cohen, 1995; Dennett, 1988). Suppose these hypotheses are right. Would evolution stop there, with everyone able to predict and manipulate each other’s behavior, or would counter-strategies also be expected to evolve? In a society of Machiavellian psycho-analysts, individuals that are harder to predict and manipulate must have selective advantages.
In their classic paper on mind-reading and manipulation, Krebs and Dawkins (1984) identified only two defenses an animal might use against having its actions predicted by a hostile “mind-reader”: concealment (of telltale intention cues), and active deception (by generating false cues). They overlook the classic third option, familiar to all military strategists, sports coaches, and game theorists, who routinely confront the problem of stopping an enemy from predicting and preparing for their next move: randomness. Genuine unpredictability. The kind that submarine commanders used in World War II when they threw dice to determine their zigzagging paths during dangerous patrols against surface ships. Thus, resistance to mind-reading may take several major forms: (1) hiding intentions (the Poker Face Strategy), (2) disinformation and deceit (the KGB Strategy), and (3) adaptive unpredictability (the Protean Strategy). Because these strategies are useful under different circumstances at different times, we might expect that they will tend to evolve together in a repertoire of social defenses against mind-reading. However, the Protean Strategy has been much neglected compared to the Poker Face and KGB Strategies. And, while the Poker Face and KGB Strategies remain vulnerable to the coevolution of smarter intention-sensing and deception-foiling capacities, there is no real defense against genuine unpredictability. Thus, the Protean Strategy may be the only evolutionarily stable strategy in the arms race against Machiavellian Intelligence.
The Protean Strategy’s usefulness has been overlooked because evolution was widely assumed to produce deterministic mechanisms of animal behavior. Descartes wrote of animals as automata; ethologists wrote of sign stimuli and simple releasing mechanisms; sociobiologists wrote of genes for specific behaviors. Such determinism makes sense for behaviors that deal with inanimate objects, but was extended too easily to behaviors subject to mind-reading. For example, Krebs and Dawkins (1984, p. 384) suggest that “Natural selection itself will favour male sea otters whose behavior happens to take advantage of the lawfulness of female behavior. The effect is that the male manipulates the female in much the same way as he manipulates a stone .... animals respond in mechanical, robot-like fashion to key stimuli.” Further, their notion of mind-reading was based on using statistical laws to exploit the supposed predictability of animal behavior: “For an animal, the equivalent of the data-collection and statistical analysis is performed either by natural selection acting on the mind-reader’s ancestors over a long period, by some process of learning during its own lifetime” (Krebs & Dawkins, 1984, p. 386).
This view of animals as intuitive statisticians suggests some obvious counter-measures. Anything that psychologists try to eliminate from their laboratory experiments can be useful against intuitive psychologists in the wild. Skew their distributions. Dehomogenize their variances. Bias their samples. Add confounds. Regress to the mean. Introduce order effects, practice effects, fatigue effects, maturation effects, expectancy effects, prestige biases, interviewer biases, and social desirability biases. Confound their reliability and validity. But these are just ways to delay enemies from discovering the real determinants of your behavior. The best protection is to undermine the determinants themselves to some degree: increase “residual variance” in one’s behavior to erode the validity of an opponent’s correlations, ANOVAs, MANOVAs, and path analyses. Squirt some noise into your behavior and their intuitive statistics will suffer.
This argument may seem silly. The notion of cognition as intuitive statistics, though once popular (e.g. Peterson & Beach, 1967), is problematic (Gigerenzer & Murray, 1987). But the accuracy of perception and cognition should be undermined by noise in the input, whether one models cognition as intuitive statistics, cognitive psychology flowcharts, neural networks, knowledge-based systems, or dynamical systems theory. Genuine unpredictability is an objective, information-theoretic feature of behavior, so would affect any information-processing system that tries to perceive and predict the behavior, no matter what cognitive metaphor one prefers. Thus, the Protean strategy should often prove useful, especially in primate social behavior. But to understand why, in more detail, we must review three things: the notion of psychological selection from evolutionary theory, the notion of mixed strategies from game theory, and the notion of protean behavior from ethology.
3 Psychological selection: How minds guide evolution
One of Darwin’s greatest achievements was to naturalize the role that mind plays in guiding evolution. He discarded grandiose religious ideas about God as Cosmic Designer and philosophical ideas about Reason willing itself into existence (e.g. Hegel, Schopenhauer, Spencer, Lamarck), and explained simply how animal perceptual systems can act as selective forces to shape the fantastic forms and varieties of flowers (Darwin, 1862), domesticated animals (Darwin, 1868), and courtship traits (Darwin, 1871). Particularly in his analysis of female choice, Darwin (1871) started to develop a general theory of ornaments based on regularities of animal perception such as sensitivity to color, symmetry, repetition, and novelty (see Miller, 1993). However, his incipient theory of what I have called “psychological selection” (Miller, 1993) was not taken forward by anyone at first, largely because Wallace (1889) and others proved so skeptical about the possibility of “aesthetic choice” by female animals (see Cronin, 1991). Perception was viewed mainly as a selective force that operates between species to shape morphological adaptations, such as the appearance of fruit, flowers, camouflage, warning coloration, and mimicry (see Wallace, 1870, 1889; Morgan, 1888; Cott, 1940). This hostility towards Darwin’s ideas about the role of minds as selective forces within species, affecting behavior and not just morphology, probably delayed the development of the Machiavellian intelligence hypothesis by about a century, from Darwin (1871) to Humphrey (1976).
Recently though, there has been an explosion of interest in psychological selection — going under a variety of terms such as “sensory drive” (Endler, 1992), “sensory exploitation” (Ryan, 1990), “signal selection” (Zahavi, 1991), and “the influence of receiver psychology on the evolution of animal signals” (Guilford & Dawkins, 1991). However, most such theory continues to emphasize how minds shape bodies, not how minds shape other minds. A few exceptions are some analyses of communication (Dawkins & Krebs, 1978), deception (Krebs & Dawkins, 1984; Byrne & Whiten, 1988), self-deception (Trivers, 1985), and animate motion perception (Miller & Freyd, 1993).
We lack a general theory of how minds can select other minds within a species. This is a major gap in evolutionary theory, because cognition can guide evolution in such powerful and surprising ways. For example, the evolutionary dynamics that arise when mate choice interacts with natural selection may lead to much faster evolutionary innovation, optimization, and diversification (Miller, 1994a; Miller & Todd, 1993, 1995; Todd & Miller, 1993). Developing a useful theory of psychological selection will require identifying fundamental regularities in perception and cognition that emerge repeatedly through convergent evolution, and which could shape the evolution of behavior within or across species. The Machiavellian intelligence hypothesis offers one such regularity: animals living in complex social groups should regularly evolve mental adaptations for social perception, prediction, manipulation, and exploitation. This regularity in turn sets up reliable selective pressures favoring counter-measures such as intention-hiding, tactical deception, and social proteanism. But to understand just how these pressures operate, we must turn to game theory.
4 Differential game theory: Mixed strategies in pursuit and evasion
The idea of a mixed strategy from game theory is best introduced with an example. In the game of Matching Pennies, two players each have a coin. Every turn, each player secretly turns her coin heads-up or tails-up. Then the coins are revealed. If the first player, in the role of “matcher,” has turned up the same side as her opponent (e.g. both coins are heads), then she wins a dollar from her opponent. If the coins mismatch (e.g. one is heads, the other tails), then she must pay a dollar to her opponent. Players can repeat this game turn after turn, producing long sequences of heads and tails, until one player goes broke, or, as more often happens, becomes lividly frustrated.
The roles of “matcher” and “non-matcher” seem different, but their goals are fundamentally the same: predict what the opponent will do, and then do whatever is appropriate (matching or not matching) to win the turn. All that matters is to find out the opponent’s intentions. The ideal offensive strategy then is to be the perfect predictor: figure out what the opponent is doing based on her past behavior, extrapolate her strategy to the next move, make the prediction, and win the turn. But there is a remarkably easy way to defeat this prediction strategy, by playing unpredictably:
“In playing Matching Pennies against an at least moderately intelligent opponent, the player will not attempt to find out the opponent’s intentions but will concentrate on avoiding having his own intentions found out, by playing irregularly ‘heads’ and ‘tails’ in successive games” (Von Neumann & Morgenstern, 1944, p. 144).
If a player picks heads with probability 1/2 and tails with probability 1/2, then no opponent, no matter how good a predictor they are, can do better than break even in this game. This half-heads, half-tails strategy is an example of a “mixed strategy,” because it mixes moves unpredictably.

Perhaps the most important and interesting result from Von Neumann and Morgenstern (1944) was that every two-player, zero-sum game of incomplete information with multiple saddle points (which, in technical terms, covers most of the interesting games you could play against someone) has an optimal strategy that is mixed rather than pure. The utility of mixed strategies has also been shown for many situations of pursuit and evasion studied by “differential game theory” (Isaacs, 1965; Yavin & Pachter, 1987; for review see Miller & Cliff, 1994a). For example, game theorists have designed “electronic jinking” systems to generate unpredictable flight paths for aircraft so they can evade guided missiles, by analogy to gazelles jinking erratically to avoid a predator (Forte & Shinar, 1988).


Evolutionary game theory (Maynard Smith, 1982) has also recognized the optimality of mixed strategies in many contests between animals. But mixed strategies are usually assumed to evolve as behavioral polymorphisms across a population rather than as unpredictable behavior within an individual. Also, evolutionary game theory has focused mostly on single-step games (such as sex-ratio determination or the Hawk-Dove game: see Maynard Smith, 1982) and discrete-step games (such as the iterated prisoner’s dilemma: see Axelrod, 1984). The literature on differential pursuit-evasion games has been strangely overlooked despite its obvious relevance to predator-prey interactions, dominance contests, sexual harassment, and play behavior. Dave Cliff and I have tried to fill this gap by developing simulations of co-evolution between pursuit and evasion strategies, implemented by genetically specified neural networks with noise parameters that evolve to implement proteanism (Miller & Cliff, 1994a, b; Miller & Cliff, submitted; Cliff & Miller, submitted).
5 Protean behavior theory: Unpredictable evasion by animals
A striking historical coincidence: four years after Michael R. A. Chance co-authored one of the foundational papers in Machiavellian intelligence (Chance & Mead, 1953), he became one of the first biologists to recognize the adaptive significance of unpredictable behavior in animals, with a paper titled “The role of convulsions in behavior” (Chance, 1957; see also Chance & Russell, 1959). Researchers had long been puzzled by “audiogenic seizures” in laboratory rats: when lab technicians accidentally jangle their keys, some lab rats go into bizarre convulsions. But Chance (1957) found that if the rats are provided with hiding places (little rat-huts) in their cages, they simply run and hide when keys are jangled; thus, the convulsions may be facultative defensive behaviors rather than pathological oddities. Convulsions would make it much more difficult for a predator to catch and hold the convulsing animal. Shortly after, Roeder (1962) found that moths tumble and loop unpredictably when hit by bat ultrasound (signalling a predator’s approach); Roeder and Treat (1961) found such tumbling much more effective at bat-evasion than passive tumbling or predictable fleeing (see May, 1991, for recent review).
Humphries and Driver (1970) termed this sort of adaptively unpredictable behavior “protean behavior”, after the mythical Greek river-god Proteus, who eluded capture by continually, unpredictably changing form. Their book Protean behavior: The biology of unpredictability (Driver & Humphries, 1988) presents a detailed theory and many ethological observations. Though they did not cite game theory, they made analogies between protean behavior in animals, unpredictable feints in human sports, and randomizing methods in military strategy.
The adaptive logic of proteanism is simple. Animals generally evolve perceptual and cognitive capacities to entrain, track, and predict the movements of other biologically-relevant animals such as prey, predators, and potential mates (Camhi, 1984; Freyd, 1992; Miller & Freyd, 1993; Premack, 1990). Such predictive abilities mean that unpredictable behavior will often be favored in many natural pursuit-evasion situations. For example, if a rabbit fleeing from a fox always chose the single apparently shortest escape route, the very consistency of its behavior would make its escape route more predictable to the fox, its body more likely to be eaten, its genes less likely to replicate, and its fitness lower. Predictability is punished by hostile animals capable of prediction. Thus, the effectiveness of almost any behavioral tactic can be enhanced by endowing it with characteristics that cannot be predicted by an evolutionary opponent (Driver & Humphries, 1988). Evolutionarily recurring pursuit-evasion contests will usually result in arms races between perceptual capacities for predicting animate motion, and motor capacities for generating protean behavior (Miller & Freyd, 1993).
Along with directional fleeing, protean escape behaviors are probably the most widespread and successful of all behavioral anti-predator tactics, being used by virtually all mobile animals on land, under water, and in the air. Driver and Humphries (1988) review ethological observations from hundreds of species, including insects, fish, birds, and mammals. Human proteanism is obvious in any competitive sport: good boxers use unpredictable feints and attacks, and good rugby players use unpredictable jinks. Predators can also exploit unpredictability to confuse prey, as when weasels do “crazy dances” to baffle the voles that they stalk, or when Australian aborigine hunters do wild dances to mesmerize the kangaroos that they hunt (Driver & Humphries, 1988). Of course, proteanism is typically used at one level of behavioral description (e.g. the trajectory through the environment), and is consistent with maintenance of orderly behavior at other levels (e.g. posture, locomotor gait, obstacle avoidance, perceptual scanning). A possible exception is convulsive “death throes,” when prey use wild, desperate, unpredictable movements to escape from the clutches of predators.
Patterns of animal play behavior reveal the importance of proteanism. Most animal play is play-chasing and play-fighting (Fagen, 1981), and includes intense practice in pursuit and evasion, prediction and proteanism, anticipation and violation of expectations. Judging by the relative play time devoted to learning different skills, foraging for plant foods and navigating through space is much easier than catching prey, escaping from predators, and fighting conspecifics. These latter skills are harder because they demand the robust, continuous, dynamic control of one’s own body in competition with the continuous, dynamic movements of a motivated, well-adapted opponent (Miller & Cliff, 1994a, b). Insofar as primates rehearse proteanism in juvenile play, they probably use it as adults to avoid predators, attack prey, and compete for dominance.
Unpredictability can be useful at many levels of biological organization. When threatened, octopi, cuttlefish, and sea pansies use “color convulsions” across the fast-response chromataphores on their skin, quickly going through different color patterns to defeat the search images (perceptual expectations) used by their predators (Driver & Humphries, 1988). Animals in groups use unpredictable movements, complex motion patterns, and confusing coloration (e.g. zebra stripes or shiny fish scales) to confuse predators. Selection for unpredictability can favor the evolution of large differences between individuals, as when animals within a species evolve “aspect diversity” (polymorphic coloration or behavior) through “apostatic selection” (Clarke, 1962) that favors low-frequency traits (e.g. because predators’ use of search images penalizes common appearances).
Co-evolution itself can be viewed as a pursuit-evasion contest between lineages rather than between individuals. From this perspective, sexual recombination makes sense as a protean strategy that unpredictably mixes up genes so as to “confuse” pathogens (Hamilton, Axelrod, & Tanese, 1990). Indeed, this proteanism argument is one of the leading explanations for the evolution of sex itself (Ridley, 1993). Despite proteanism’s importance, it has been long overlooked in biology, because complex order rather than useful chaos was assumed to be the defining feature of Darwinian adaptations (see Miller, 1993).
6 Can animals really randomize?
For decades, experimental psychologists have investigated whether humans can generate sequences of numbers, letters, or motions that obey various tests of mathematical randomness. Dozens of papers suggested that Reichenbach (1934) was correct in suggesting that humans tend to alternate too much (heads-tails-heads-tails) and don’t produce enough long runs (heads-heads-heads-heads). Tune's (1964) review concluded that “humans are incapable of generating a random series of selections from a finite number of alternatives”, and Wagenaar’s (1972) review concluded “Producing a random series of responses is a difficult, if not impossible task for humans, even when they are explicitly instructed”. Complex models were advanced to explain the “heuristics”, “biases”, or “cognitive constraints” underlying these failures of randomization (e.g. G. A. Miller & Frick, 1949; Kahneman & Tversky, 1972; Treisman & Faulkner, 1987).
However, most such studies were artificial in the extreme, typically requiring isolated subjects to write down a series of numbers on paper with instructions like “be as random as possible.” Recently though, Amnon Rapoport — a veteran game theorist (see Rapoport, 1966), and submitter of the “Tit for Tat” strategy that won Axelrod’s (1984) iterated prisoner’s dilemma contest — reasoned that randomization should be best in real social competition against a predictive opponent. Rapoport and Budescu (1992) found that sequences come much closer to genuine mathematical randomness when they are generated by subjects playing a real, face-to-face, strictly competitive game (“Matching Pennies”), than when they are generated by isolated subjects trying to write down “random sequences.” Even without explicit competition, other researchers have shown that animal and human subjects can learn to generate almost perfectly random sequences when given good feedback (Lopes & Oden, 1987; Neuringer, 1986; Neuringer & Voss, 1993). The randomization abilities of monkeys and apes could be tested by having them play a variant of the penny-hiding game, used by Baron-Cohen (1992) to show that autistics lacking a Theory of Mind are poor at randomization in two-person zero-sum games.
The recent skepticism about animals’ capacities for varied, unpredictable, novel behavior is ironic because such capacities were fundamental to Behaviorist theories of operant conditioning, which drew explicit parallels between learning and evolution. For example, Skinner (1974) and Campbell (1960) saw random exploratory behavior as analogous to genetic mutations, and reinforcement as analogous to natural selection. Without a reasonably unpredictable, varied set of initial behaviors for reinforcement to “shape,” the development of complex behavioral repertoires would be impossible. A classic volume titled Functions of varied experience (Fiske & Maddi, 1961) demonstrates the sophistication of Behaviorist reasoning about the importance of behavioral variation, before the computer metaphor and cognitive psychology conflated behavioral variation with noisy information and malfunctioning programs.
  1   2   3


The database is protected by copyright ©essaydocs.org 2016
send message

    Main page