Essay Review Evolution and Apparent Irrationality Rory Smead Department of Philosophy & Religion Northeastern University 371 Holmes Hall 360 Huntington Avenue Boston, MA 02115 E-‐mail address:
[email protected] Evolution and Rationality: Decisions, Co-‐operation and Strategic Behavior. Samir Okasha and Ken Binmore eds.; Cambridge University Press, Cambridge, 2012, pp. 296, Price $90, hardcover, ISBN 978-‐1-‐107-‐00499-‐3. On the surface, evolution and rationality are very different notions. The theory of rational choice was developed in economics and is used to understand the decisions of rational agents. Evolutionary theory, on the other hand, studies the changes of biological populations and pertains to organisms that have no mind at all. A closer examination of the two theories begins to reveal some interesting and deep connections. The central notion of each theory is one of maximization. Rational individuals will make choices that maximize their utility. Evolution will select for behaviors that are fitness maximizing. Consequently, it is not surprising that the tools and methods of rational choice theory have been fruitfully applied to evolutionary settings and vice versa (see e.g. Hoffbauer and Sigmund, 1998; Sandholm, 2010). This methodological parallel raises several interesting questions surrounding evolution and rationality. Does utility (roughly) correspond to
biological fitness and if not, why not? Are humans rational decision makers? Will evolution lead to behaviors that are close to rational? These are some of the central questions addressed in Evolution and Rationality. Samir Okasha and Ken Binmore have assembled a very nice collection of works that spans several disciplines and presents a range of different perspectives on the connections between evolution and rationality. This diversity is important due to the interdisciplinary nature of the central questions. However, this same diversity, perhaps unsurprisingly, means that there is not a clear unified message regarding the connections between evolution and rationality. On several key points there seems to be disagreement among the contributed chapters. Moreover, the focus of several chapters is not on rational behavior but on providing explanations of apparently irrational behavior. This review will attempt to summarize what general messages there are in the book as well as offer some critical reflections on the points in contention. I will focus on two central themes. The first is the methodological parallel between rational choice theory and evolutionary theory, which I consider in the context of evolutionary game theory. The second is whether evolution should be expected to produce outcomes that correspond to rational behavior and, in particular, possible evolutionary explanations of apparently irrational behavior. I will also discuss a connection between evolution and rationality that is much less prominent in the book: the evolution of the cognitive mechanisms that underlie decision-‐making. The methodological connection between rational choice theory and evolution is nicely highlighted by the relationship between rational game theory and
evolutionary game theory. Game theory, historically, is an extension of the classical theory of rational choice to strategic situations. The primary method of rational game theory involves equilibrium analysis, where equilibria are often defined in terms of rational self-‐interest. The central solution concept in game theory is the Nash equilibrium: a set of strategies such that each player’s strategy is maximizing her expected utility given the strategies of all the other players. The equilibria are taken to represent the expectation for how rational individuals will behave when playing a game. Evolutionary game theory applies the formal tools of game theory to (possibly arational) biological populations with the aim of determining how they will evolve when they are facing a fitness affecting game. Given this focus, the natural method of analysis would presumably involve evolutionary dynamics: how does a population change over time when interacting in a game? Indeed, models of evolutionary dynamics are part of early work in evolutionary game theory (Taylor and Jonker, 1978). However, much of the methodology of evolutionary game theory involves static equilibrium analysis similar to that of its rational choice counterpart. Moreover, despite the focus on populations of arational individuals, the traditional “rational” solutions to games, such as the Nash equilibria, can be used to gain insight into how a population will evolve. One reason for this is that Nash equilibria form rest points under a broad class of evolutionary dynamics (Weibull, 1995). However, placing too much of an emphasis on Nash equilibria or other “rational” solutions would be a mistake in evolutionary contexts. In Chapter 1 of Evolution and Rationality, Peter Hammerstein argues that considering the
evolutionary process explicitly is important for determining the relevance of certain solution concepts. Hammerstein specifically considers the relevance of the subgame perfect equilibrium, a refinement of the Nash equilibrium, concluding that it is relevant to specific biological problems but that it should be used with caution. Similar points could be made for the connections between Nash equilibria and evolutionary dynamics. Although there may be no evolutionary change at the Nash equilibria of games, these points may not be stable under evolutionary dynamics— evolution could lead nearby states away from Nash equilibria. One refinement of the Nash equilibrium that was intended specifically for evolutionary settings is the evolutionarily stable strategy or ESS (Maynard Smith and Price, 1973; Maynard Smith, 1982). The ESS represents a behavior, which, if adopted by the population, no mutant behaving otherwise would be able to invade. The ESS, when it exists, is strongly stable under a broad class of evolutionary dynamics (Weibull, 1995) and, consequently, is often the central solution concept for evolutionary games. In Chapter 4, Simon Huttegger and Kevin Zollman discuss the ESS and its use in biology. They argue that the methodology of focusing on ESSs as the explanatorily relevant solutions is flawed when we explicitly consider evolutionary dynamics. The reason is that there are examples of evolutionary outcomes that arise and persist, yet do not correspond to any ESS. Furthermore, these outcomes may be more likely, and hence more significant, than the outcomes that correspond to an ESS of the game. It is important to note that a dynamical analysis is also important for considering rational actors as well. No real agents embody the extreme idealizations
of full economic rationality as is often assumed in equilibrium analysis. If real agents are rational they are only boundedly rational. Consequently, modeling the way that agents learn over time may be just as important for the economic theory of rational choice as evolutionary dynamics are for evolutionary theory. Indeed, much of the development of evolutionary game theory has taken place within economics and has been aimed at how boundedly rational agents will learn to interact (e.g. Fudenberg and Levine, 1998; Young, 2004). Hence, there is a sense in which the issue of static vs. dynamic methodology is orthogonal to the relationship between rationality and evolution. Nevertheless, the general lesson from Hammerstein and Hutteger & Zollman goes beyond the fact that the outcomes of evolutionary dynamics do not always coincide with particular equilibrium concepts. A dynamic methodology brings evolutionary change to the forefront of the analysis whereas traditional equilibrium analysis focuses on static solutions to games. Consequently, the dynamical approach can shed light on topics static equilibrium analysis cannot. As Huttegger and Zollman point out, equilibrium analysis only tells us what happens when populations or players are already close to the equilibrium and cannot tell us how players can reach an equilibrium if they are not near one. The underlying dynamics of behavior change, whether of individuals learning over time, or evolving populations, is arguably more important that the static equilibria. I think these arguments are correct and suspect that many of the book’s contributors would agree. However, despite the recognized importance of evolutionary dynamics, other chapters of Evolution and Rationality have little
discussion of dynamical models and often proceed with static equilibrium analysis. While, this presents an apparent conflict of methodology throughout the book, I think the dual approach is justified for two reasons. First, much of the previous important work that has been done on evolution and rationality involves static equilibrium analysis. Second, even if one holds a dynamical analysis as privileged, this does not mean that we ought to abandon the methods of equilibrium analysis. The rational equilibria in games can be informative of the evolutionary possibilities and constraints. Many of the static equilibrium concepts such as ESS or Nash equilibrium have certain properties that hold under a wide range of possible dynamics. Consequently, equilibrium analysis can be of use if we are ignorant of (or if we wish to remain agnostic about) the underlying evolutionary dynamics. Furthermore, the formal tools from decision theory and economics for identifying and analyzing the equilibria of a game are well developed and are often relatively easy to use compared dynamic methods. This means that equilibrium analysis may be a prudent place to begin analysis even if we believe that a dynamical model will have the final say. Beyond the methodological parallels between studies of rationality and evolution, there are also possible connections between some of their central notions. Does utility correspond to fitness? If natural selection can, in some way, act on our preferences, one should expect the evolution of fitness tracking preferences. An organism with utilities that correspond to fitness will, ceteris paribus, do better than one with utilities that do not. Consequently, our decision-‐making should, at least roughly, be maximizing evolutionary fitness subject to whatever constraints
are present. In Chapter 10, Herbert Gintis presents a similar argument in the context of a broader discussion on how to unify the economic, sociological, psychological and biological models of decision-‐making. Gintis adds the caveat that individuals may not be consciously maximizing utility or anything else, but their decisions may nonetheless be captured by optimization models. The standard decision-‐theoretic definition of utility is used throughout the book. Utility is a numerical representation of preferences, which are revealed by an agent’s choice behavior (see von Neumann and Morgenstern, 1944, Binmore, 2009). Fitness is understood in terms of relative ability to produce offspring. With this in mind, the argument above is fairly convincing: the most evolutionarily successful organisms will be those whose behaviors maximize fitness, which will happen if fitness and utility coincide. Given the plausibility of this view, the most interesting aspects of the utility-‐ fitness connection naturally involve cases where preferences (and so, utilities) do not seem to match fitness. These are cases where individuals behave in apparently irrational ways—“irrational” in the evolutionary sense that behavior that does not approximately maximize fitness. We know that humans sometimes behave irrationally and this is to be expected even if fitness and utility generally do coincide; we may make mistakes or find ourselves in unfamiliar circumstances where we simply do not know the best way to behave. But, it is systematic and persistent apparent irrationalities that demand an explanation. One explanation for irrationality points to the constraints and limits on what evolution can produce. In Chapter 2, El Mouden, Burton-‐Chellow, Gardiner and West
present a convincing case that humans can often be described as striving to maximize their inclusive fitness, though they are imperfect and sub-‐optimal predictable ways. The apparent irrationalities of humans, on their view, are expected given that there are many constraints and complexities in real evolutionary settings. For example, natural selection prefers the “quick and cheap” solutions to problems that rely on heuristics rather than precise optimization. These can lead to good but not optimal behavior in ordinary circumstances and may lead to misapplications when individuals are faced with unusual or novel circumstances. Furthermore, El Mouden et al. argue, it can be difficult to assess true fitness effects of behavior as well as the extent that behavioral traits can be molded by natural selection. Deeper differences between utility and fitness may be expected when we consider the oddities of modern human culture relative to our evolutionary past. In Chapter 11, Kim Sterelny argues that cultural transmission of ideas is a way for us to acquire preferences and values that are detrimental to our fitness. Furthermore, this happens by learning within an individual’s lifetime and may not be closely correlated with genetic relationships. Consequently, culturally acquired maladaptive behaviors will not necessarily be eliminated by selection. When this occurs, fitness and utility can become “decoupled.” The explanations of irrational behavior offered by El Moulden et. al. and Sterelny take irrationality to be an unselected consequence; that it is due to constraints or complexities in the evolution of decision-‐making mechanisms. In Chapter 3, Alasdair Houston distinguishes this variety of explanation for apparently
irrational behavior from a second type: that irrationality is selected for during evolution. Of course, these two types of explanation need not be viewed as competing and both may be applicable to instances of irrational behavior. However, the idea that irrationalities are somehow evolutionarily beneficial is an intriguing possibility. Houston’s discussion focuses on the case of intransitive preferences. Transitivity of preference is commonly assumed to be a requirement of rationality. Yet, there are cases of laboratory experiments where animals and humans appear to exhibit intransitive preferences. It is possible that natural selection leads to intransitivity of choice, but when this occurs, Houston argues, the intransitivity are often only apparent. A correct view of the decision situation would reveal that, considered from a broader context, the choice is not intransitive. The general idea is that irrationalities seem to occur because theorists take too narrow a view of the decision-‐making problem that needs to be solved. On reflection, there is something rather obvious about this explanation of apparent irrationalities (in the biological sense). Any trait that is selected must have some evolutionary advantage, for this is part of what selection means. However, when assessing whether or not a decision or behavior is one that maximizes fitness it is important to specify a measure of fitness. This turns out to be more complicated than it may seem. Behaviors or decisions can be viewed from many different scopes. On the one hand we may view a decision from a narrow “local” perspective where we consider a particular setting and take everything else to be fixed or constant. On the other hand, we may view a decision from a broader “global” perspective and how this decision may relate to many different settings across time. The best thing
to do in a particular context, holding everything else fixed, may not be the best thing to do all things considered. The time-‐scale in which we measure fitness is also important: an action that maximizes gene frequency in the next generation may not be maximizing gene frequency two generations from now. Evolutionary explanations of irrational behavior tend to assume that the behavior is only apparently irrational. That is, the behavior is irrational in a local sense, but not some more global sense. This is largely a point of consensus throughout the book: cases of irrationality are either not products of selection or are only apparent irrationalities and generate some benefits in a broader context. Despite this, selection can lead to systematic and striking cases of local irrationalities. This is particularly true in the context of social interactions, where the distinction between strategic and non-‐strategic situations is important for the potential connections between evolution and rationality. In strategic situations, having preferences that do not directly track evolutionary fitness can result in indirect fitness benefits. In Chapter 6, Berninghaus, Güth and Kliemt discuss “indirect evolution,” where individuals make decisions according to their preferences but the preferences undergo evolution. In these models, it is assumed that individuals have some set of preferences, have information about the preferences of others, and then behave rationally given the preferences of all involved. Having preferences that do not match evolutionary fitness can influence the way that others interact with you, and thereby, it is possible to reap indirect evolutionary benefits from such preferences.
The fact that inaccurate preferences (in a local sense) may generate benefits in a broader context is not a new idea (Frank, 1987). A similar approach is explored in Chapter 7 where Wolpert and Jamison explore what they call “persona games.” The idea is that individuals may adopt different personas when playing different games, which changes the way we behave in those games. More specifically, the personas that individuals may adopt correspond to different sets of preferences when making choices in a game. Apparently irrational behavior can be a result of rationally adopting (from a global perspective) certain personas when facing a game. Wolpert and Jamison apply this idea to provide a possible explanation for apparently irrational cooperation in the Prisoner’s Dilemma. Cooperative behavior in social dilemmas, such as the Prisoner’s Dilemma, has historically received much attention in both evolutionary biology and rational choice theory. This is also true in Evolution and Rationality. This topic serves as the focal point in Chapter 8 where Jack Vromen discusses reciprocity in cooperative situations and whether or not humans are strong reciprocators (having the predisposition to cooperate and to punish non-‐cooperators even if it is costly). Cooperation in social dilemmas is also a central focus in Chapter 9 where Natalie Gold discusses the notion of “team reasoning” in games. The idea of team reasoning is that agents may approach an interaction by thinking about themselves as part of a group and aim to do what is best for the group. If individuals reason about a game in this way it is possible to see behaviors that are not rational from the perspective of egocentric individuals. Gold also discusses a number of possible explanations for
how team reasoning may have evolved but does not commit to any specific explanation. The significance of the way that people reason about their situation is related to another issue that does not receive much attention: the evolution of the cognitive mechanisms that underlie decision-‐making (see e.g. Hammerstein and Stevens, 2012). The common sense or psychological notions of rationality involve the process of reasoning and not simply the behavioral result. However, Evolution and Rationality has relatively little discussion of the way we might expect evolved agents to think or reason. This is because the classical theories of rationality are not aimed at actual thinking or decision-‐making, but rather the decision-‐making of idealized agents: “…few defenders of the rational economic agent framework think of it as a model of thinking” (Sterelny p.254). The idea is that we should not expect evolved organisms to be fully rational economic agents, but we will expect them to behave as if they were rational. Similar sentiments are expressed by many of the book’s contributors. However, being rational is not simply making the right decisions, but also making them for the right reasons. Deliberation, learning and the synthesis of information are also important aspects of rationality. One brief discussion of the cognitive processing aspect of rationality is by Herbert Gintis in Chapter 10. Gintis highlights studies that have successfully applied the methods and tools of rational choice theory to cognitive science as well as potential relationships between topics in cognitive science to game theory and evolution. Certain decision-‐making
processes seem to involve explicit and not just “as if” maximization. However, the evolution of cognitive mechanisms is not discussed in detail. Even if we expect evolution to lead to fitness maximizing behavior, there is no guarantee that the agents will have cognitive mechanisms designed to maximize fitness (or anything else for that matter). What is important for natural selection is fitness of behavior within the current ecological and social settings. Certain parts of a decision procedure that maximizes fitness can be “out sourced” to others or to the environment as long as those aspects of the environment remain relatively stable. What matters is the long-‐term impact of behavior on inclusive fitness. As long as (approximately) the right behavior is reliably generated, the cognitive mechanism generating that behavior is of little consequence. Consequently, there is no a priori reason to expect evolved cognitive mechanisms to be directly tracking fitness. In fact, there may be reason to think that the most successful cognitive mechanisms and decision rules will be those that do not directly track fitness. In Chapter 5, Brighton and Gigerenzer argue that the real world does not correspond to the “small worlds” considered in standard decision theory except in very artificial circumstances. When the possible states of the world (and their probabilities) are unknown, rational maximization is not possible and perhaps meaningless. Given the complexity and radical uncertainty of the “large worlds” simple heuristics can often outperform attempts at optimization. The study of the evolution of learning has produced results that may complicate the evolution-‐rationality connection, particularly in strategic settings (see e.g. Dubois et. al., 2010; Katsnelson et. al., 2012). The best way to learn can
depend on how others are learning, which means that it is important to consider the way that different learning mechanisms may interact. An example of this type of complexity comes from studies of the evolution of imitation learning. One recurrent result in these studies is that we should expect populations to be mixed with respect to the way they approach decision problems: some individuals will imitate, others will learn for themselves (e.g. Rogers, 1988; Boyd and Richerson, 2005). There is no reason to learn for oneself if the behavior of others is already a reliable indicator of the best thing to do. But, imitation is only useful when there are not too many other imitators. Hence, these models suggest that we should expect populations with different types of learners, some of which may exhibit “as if rationality” even if they do not have cognitive mechanisms one would consider rational. In all, Evolution and Rationality is a stimulating collection of work that should be of interest to philosophers, biologists, psychologists and economists alike. With the wide range of views and approaches covered in the book, there is also variation in accessibility: while many chapters are written for a general audience, some may be too difficult for undergraduates and others too technical for those without some mathematical training. Despite this, the book is an excellent and much needed contribution to an area that demands interdisciplinary attention. I think the editors will be successful in their stated goal to “promote a constructive dialogue” between the diverse approaches in the study of evolution and rationality. References Binmore, K. (2009). Rational Decisions. Oxford: Oxford University Press.
Boyd R. and Richerson P. J. (2005). Culture and the Evolutionary Process. Oxford: Oxford University Press. Dubois, F., Morand-‐Ferron, J., and Giraldeau, L. A. (2010). Learning in a game context: strategy choice by some keeps learning from evolving in others. Proc. R. Soc. B 277, 3609-‐3616. Frank, R. H. (1987). If homo economicus could choose his own utility function, would he want one with a conscience?. The American Economic Review, 77(4), 593-‐604. Fudenberg, D. and Levine, D. (1998). The Theory of Learning in Games. Cambridge: MIT Press. Hammerstein, P. and Stevens, J. R. (2012). Evolution and the Mechanisms of Decision Making. Cambridge: MIT Press. Hofbauer, J. and Sigmund, K. (1998). Evolutionary Games and Population Dynamics. Cambridge University Press. Houston, A.I., McNamara, J. M., and Steer, M. D. (2007). Do we expect natural selection to produce rational behaviour? Philosophical Transactions of the Royal Society B, 362, 1531-‐1543. Katsnelson, E., Motro, U., Feldman, M. W., and Lotem, A. (2012). Evolution of learned strategy choice in a frequency-‐dependent game. Proc. R. Soc. B, 279, 1176-‐1184. Maynard Smith, J. (1982). Evolution and the Theory of Games. Cambridge: Cambridge University Press. Maynard Smith, J. and Price, G. R. (1973). The logic of animal conflict. Nature, 246, 15-‐18.
Rogers, A. R. (1988). Does biology constrain culture? American Anthropologist, 90, 819-‐831. Sandholm, W. H. (2010). Population Games and Evolutionary Dynamics. Cambridge: MIT Press. Skyrms, B. (1996). Evolution of the Social Contract. Cambridge: Cambridge University Press. Taylor, P. and Jonker, L. (1978). Evolutionary stable strategies and game dynamics. Mathematical Biosciences, 40, 145-‐56. von Neumann, J. and Morgenstern, O. (1944). Theory of Games and Economic Behavior. Princeton: Princeton University Press. Weibull, J. W. (1995). Evolutionary Game Theory. Cambridge: MIT Press. Young, P. H. (2004). Strategic Learning and its Limits. Oxford: Oxford University Press.