Elsevier

Cognition

Volume 178, September 2018, Pages 109-117
Cognition

Original Articles
(Ir)rational choices of humans, rhesus macaques, and capuchin monkeys in dynamic stochastic environments

https://doi.org/10.1016/j.cognition.2018.05.019Get rights and content

Abstract

Human and animal decision-making is known to violate rational expectations in a variety of contexts. Previous models suggest that statistical structures of real-world environments can favor such seemingly irrational behavior, but this has not been tested empirically. We tested 16 capuchin monkeys, seven rhesus monkeys, and 30 humans in a computerized experiment that implemented such stochastic environments. Subjects chose from among up to three options of different value that disappeared and became available again with different probabilities. All species overwhelmingly chose transitively (A > B > C) in the control condition, where doing so maximized overall gain. Most subjects also adhered to transitivity in the test condition, where it was suboptimal, but ultimately led to negligible losses compared to the optimal, non-transitive strategy. We used a modelling approach to show that differences in temporal discounting may account for this pattern of choices on a proximate level. Specifically, when short- and long-term goals are valued similarly, near-optimal decision rules can map onto rational choice principles. Such cognitive shortcuts have been argued to have evolved to preserve mental resources without sacrificing good decision-making, and here we provide evidence that these heuristics can provide almost identical outcomes even in situations in which they lead to suboptimal choices.

Introduction

Organisms make thousands of choices every day, choosing where to go, what to eat, with whom to mate, and whether to run away or fight. How can individuals make good decisions, particularly when survival depends on it? In real-life situations, it can be difficult to assess what the best available option is, particularly when the outcomes of our actions are uncertain. For example, should a female mate with the current male or wait for a better opportunity, which risks going empty-handed? Should an animal expend energy moving toward a known but almost depleted food patch or look for a new one? Understanding why and under what circumstances we make adaptive decisions – or fail to do so – is a fundamental question to human and animal behavior.

Given the complexity of natural environments, it is not surprising that individuals often do not act entirely optimally, but deviate from rational expectations in certain contexts, e.g., switching their preferences depending on their past experiences, the current choice set, how options are presented, or their energetic state (Kahneman et al., 1982, Rosati and Stevens, 2009, Thaler, 1992). While the optimal solution is sometimes obvious and does not require much thought, other times it may be difficult to discern or may even appear irrational. In this study, we explored how subjects would respond in the latter case. We tested whether three primate species would employ seemingly irrational decision rules in environments in which doing so was actually optimal.

Rational choice theory (Neumann von & Morgenstern, 2007) provides a useful starting point to predict decision makers’ choices based on the assumption that individuals assign absolute values for each option and use them to form consistent and ordered preferences in order to maximize their long-term gain (and minimize their long-term loss). Two important principles that follow from this assumption are the transitivity axiom (if an individual prefers Option A given the choice {A, B} and Option B in choice {B, C}, then it should prefer A in choice {A, C}) and the principle of independence of irrelevant alternatives (adding or removing inferior options to a choice set should not influence decisions; if an individual prefers Option A given the choice {A, B, C}, then it should also prefer A in choice {A, B}).

However, both humans and animals are known to sometimes violate even these most basic principles of rational choice, particularly when options vary along multiple dimensions (Bateson et al., 2002, Bateson et al., 2003, Latty and Beekman, 2010, Parrish et al., 2015, Shafir, 1994, Shafir et al., 2002, Waite, 2001). For example, hoarding grey jays consistently violated transitivity when choosing among three foraging options that placed them at different risks of predation (Waite, 2001): The jays preferred one raisin at low risk over two raisins at intermediate risk (A > B) and preferred those two raisins over three raisins at high risk (B > C), but preferred three raisins at high risk over one raisin at low risk (C > A). Similarly, hummingbirds’ preferences between two artificial flowers that varied in volume and sugar concentration changed when a lower-quality decoy was introduced, violating the assumption of independence of irrelevant alternatives (Bateson et al., 2002, Bateson et al., 2003). Such studies suggest that decision makers often assign relative rather than absolute values to the options available in their environment, which can lead to inconsistent choices. They do not, however, explain why we act irrationally under some circumstances but not others (e.g., Schuck-Paim & Kacelnik, 2002), and what characterizes the circumstances in which we do.

Ecological rationality is one framework that has been put forward to explain such irrational behavior (e.g., Houston, 1997, Houston and McNamara, 1999, Houston et al., 2007a). This view emphasizes that decision rules evolve in, and adapt to, complex natural environments. By this view, an individual that acts optimally (i.e., acts as if to maximize fitness) should violate rational choice principles in some environments. Here, we refer to such violations as economically irrational behavior that arises from inconsistent preferences. However, such behavior may be ecologically (or biologically) rational when the decision rule fits the structure of the environment (Kacelnik, 2006). For example, animals may fail to behave in a way that maximizes their short-term gains, acting as if conditions were worse than they actually are. If the environment fluctuates between mild and harsh conditions, then a bias toward behavior that is adaptive in harsh conditions may be favored (even if it is suboptimal in mild conditions) because doing well in poor environments often has a bigger impact on fitness outcomes in the long run (McNamara, Trimmer, Eriksson, Marshall, & Houston, 2011). Thus, economically irrational behavior is not necessarily suboptimal but can instead reflect optimal choices under specific sets of conditions. This approach has been useful to demonstrate that cognitive biases such as pessimism and contrast effects can be adaptive (The Modelling Animal Decisions Group, Fawcett, Fallenstein, et al., 2014), particularly in uncertain environments that vary over space and time (spatiotemporal heterogeneity), but tend to be more similar the closer they are in space and time (positive autocorrelation). This is because the options that an animal currently faces can provide information about likely future conditions in the environment, which can – and indeed should – affect the best strategy for the current choice. Indeed, both violations of transitivity and independence of irrelevant alternatives can be optimal under such conditions (e.g., Houston et al., 2007b, Trimmer, 2013).

Of course, this concept of rationality only becomes meaningful with respect to the specific properties of an environment. That is, the same decision rule may be ecologically rational in one setting but ecologically irrational in another. Models of ecological rationality can explain why animals with such evolved decision rules may violate economic principles in situations in which doing so is suboptimal (e.g., in deliberately simplified lab tasks), but they also allow us to construe environments in which animals should violate economic principles to behave optimally. Testing this explicitly is important in order to establish whether and how animals’ proximate decision rules match those that are optimal at an ultimate level.

The purpose of this study was to empirically test the predictions of one such model by McNamara, Trimmer, and Houston (2014) on three primate species: humans, rhesus macaques (Old World monkeys), and capuchin monkeys (New World monkeys). These species are suitable to such a task because their decision-making strategies are likely to be adapted to environments with spatiotemporal, autocorrelated fluctuations (e.g., as a result of the monkeys’ reliance on fruit; MacLean et al., 2014, Oates, 1986). Moreover, primates have demonstrated various degrees of statistical learning and successfully solve cognitive tasks in which the outcomes of their decisions were determined probabilistically (e.g., Conway and Christiansen, 2001, Parrish et al., 2014, Proctor et al., 2014).

In their model, McNamara et al. (2014) demonstrate that an individual that maximizes its long-term gain can exhibit violations of rational choice in a dynamic environment in which options are not always available. Instead, each of three foraging options varies over time and will probabilistically disappear if it is currently available, or reappear if it is currently unavailable. An individual is assumed to take some time to process the option that it chose and cannot make another decision during this time (but options continue to appear and disappear). That is, there is an opportunity cost associated with an individual’s choice because better options may appear or unchosen options may disappear while it handles the current option. McNamara et al. identified specific cases in which the optimal decision rules either favor economically rational or irrational choices. We chose this model because it predicts violations of economic rationality regardless of energetic state and therefore lent itself to implementation in a computerized task with our populations. (An animal’s state, such as whether it is hungry or satiated, is known to influence decision-making (e.g., Caraco, 1981), but is often difficult to measure, manipulate, or control (but see Schuck-Paim, Pompilio, & Kacelnik, 2004).)

For this study, we derived two different environments based on McNamara et al.’s (2014) model to assess whether primates would actually behave so as to maximize their gain and flexibly follow or break the rational choice principles. Subjects were tested with distinct option parameters that call for them to either violate (test condition) or follow (control condition) the transitivity principle to maximize long-term gain. (Note that the model also predicts violations of independence of irrelevant alternatives; however, because this principle is violated whenever transitivity is violated (McNamara et al., 2014; see also Fig. 4), we focus on transitivity alone.) If individuals act optimally, they should flexibly violate or adhere to the principle of transitivity in a way that is consistent with the statistical structure of the task setting. Thus, we predicted that subjects would adopt intransitive decision rules in the test condition, but transitive decision rules in the control condition.

We did not originally employ a modelling approach ourselves. However, to provide an initial test of a hypothesis that emerged from our results (see below), we subsequently extended McNamara et al.’s (2014) model to include a temporal discounting parameter to systematically assess how decision rules may change when future rewards are discounted (a well-known cognitive constraint on decision-making, Stevens, 2014). Specifically, we expected transitive decision rules to become optimal over a wider range of parameters when future rewards are valued less, relative to immediate rewards.

Section snippets

Subjects

We recruited 30 undergraduate participants (26 female, 4 male, age: M ± SD = 21.33 ± 4.66, range: 18–40 years) from Georgia State University through an online system and gave them course credit for participation in the experiment. We also tested 16 capuchins monkeys (Cebus [Sapajus] apella, nine female, seven male, age: M ± SD = 14.06 ± 5.37, range: 7–28 years) and seven rhesus macaques (Macaca mulatta, all male, age: M ± SD = 19.00 ± 6.95, range: 12–32 years) at the Language Research Center of

Observed decision rules

In their 90-min sessions, humans made M ± SD = 299.60 ± 28.49 decisions in the control and M ± SD = 344.47 ± 23.39 decisions in the test condition. In the first 90 min, monkeys made M ± SD = 211.65 ± 77.02 decisions in the control and M ± SD = 241.83 ± 94.54 decisions in the test condition, with M ± SD = 2524.96 ± 769.59 total in the control and M ± SD = 3333.13 ± 1035.31 total in the test condition.

The frequency of subjects’ observed decision rules (Table 2) varied significantly across species

Modelling temporal discounting

We had no a priori assumptions about how subjects would develop their decision rules, or about the underlying processes that led to particular selections. We adopted an undiscounted MDP (see supplementary material) to derive the optimal decision rule in the two conditions, and decisions were based both on the immediate reward of an action as well as on an estimated value of the next state. However, because there was no temporal discounting parameter, decisions were modelled to consider events

Discussion

Successful decision rules readily emerged in the dynamic environments presented here, which implemented common stochastic characteristics of the real world. In the control condition, virtually all humans, rhesus macaques, and capuchin monkeys followed the optimal, transitive decision rule, which maximized both short- and long-term gains. In the test condition, only 5 of the 16 capuchins and none of the rhesus or humans found the optimal decision rule of violating transitivity, which maximized

Acknowledgements

We thank the veterinary and animal care staff at the Language Research Center for maintaining the health and well-being of the monkeys. Funding: This work was supported by the National Science Foundation (NSF) [grant numbers SES 1123897, SES 1425216] (to SFB) and the Duane M. Rumbaugh Fellowship from Georgia State University (to JW).

Ethical standard

This study was purely behavioral, non-invasive, and strictly adhered to the legal requirements of the United States of America. All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards. All applicable international, national, and institutional guidelines for the care and use of animals were followed. All

References (45)

  • M.J. Beran

    The comparative science of “self-control”: What are we talking about?

    Frontiers in Psychology

    (2015)
  • M.J. Beran et al.

    Sequential responding and planning in capuchin monkeys (Cebus apella)

    Animal Cognition

    (2012)
  • T. Caraco

    Energy budgets, risk and foraging preferences in dark-eyed juncos (Junco hyemalis)

    Behavioral Ecology and Sociobiology

    (1981)
  • T.A. Evans et al.

    An efficient computerized testing method for the capuchin monkey (Cebus apella): Adaptation of the LRC-CTS to a socially housed nonhuman primate species

    Behavior Research Methods

    (2008)
  • G. Gigerenzer et al.

    Reasoning the fast and frugal way: Models of bounded rationality

    Psychological Review

    (1996)
  • A. Hope

    A simplified Monte Carlo significance test procedure

    Journal of the Royal Statistical Society. Series B (Methodological)

    (1968)
  • A.I. Houston

    Natural selection and context-dependent values

    Proceedings of the Royal Society of London B: Biological Sciences

    (1997)
  • A.I. Houston et al.

    Models of adaptive behaviour

    (1999)
  • A.I. Houston et al.

    Do we expect natural selection to produce rational behaviour?

    Philosophical Transactions of the Royal Society of London. Series B

    (2007)
  • A.I. Houston et al.

    Violations of transitivity under fitness maximization

    Biology Letters

    (2007)
  • A. Kacelnik

    Meanings of rationality

  • D. Kahneman et al.

    Judgment under uncertainty: Heuristics and biases

    (1982)
  • Cited by (9)

    • Are the roots of human economic systems shared with non-human primates?

      2020, Neuroscience and Biobehavioral Reviews
      Citation Excerpt :

      In many cases, these apparently irrational tendencies are hypothesized to have evolved because they were adapted to a particular context (Houston & McNamara, 1999; Kacelnik, 2006) or because, on average, following a rule of thumb left the individual decision-maker no worse off and reduced the cognitive load required for each individual decision (i.e., see Simon, 1972; Gigerenzer & Goldstein, 1996; see also Section 1.2). Indeed, in some cases the payoffs between the rational and “irrational” decisions may be statistically indistinguishable (Watzek & Brosnan, 2018). The issue of detecting common decisional patterns under risk and uncertainty in human and non-human primates brings about the question of the relative importance of possibly shared cognitive decisional aptitudes and tendencies as sufficient and necessary building blocks for economic behavior in non-individual contexts, in which we can speak of economic behavior in a richer and perhaps more appropriate sense.

    • Economics and Epicycles

      2021, Perspectives on Psychological Science
    View all citing articles on Scopus
    View full text