Ambiguity aversion in rhesus macaques

Abstract

People generally prefer risky options, which have fully specified outcome probabilities, to ambiguous options, which have unspecified probabilities. This preference, formalized in economics, is strong enough that people will reliably prefer a risky option to an ambiguous option with a greater expected value. Explanations for ambiguity aversion often invoke uniquely human faculties like language, self-justification, or a desire to avoid public embarrassment. Challenging these ideas, here we demonstrate that a preference for unambiguous options is shared with rhesus macaques. We trained four monkeys to choose between pairs of options that both offered explicitly cued probabilities of large and small juice outcomes. We then introduced occasional trials where one of the options was obscured and examined their resulting preferences; we ran humans in a parallel experiment on a nearly identical task. We found that monkeys reliably preferred risky options to ambiguous ones, even when this bias was costly, closely matching the behavior of humans in the analogous task. Notably, ambiguity aversion varied parametrically with the extent of ambiguity. As expected, ambiguity aversion gradually declined as monkeys learned the underlying probability distribution of rewards. These data indicate that ambiguity aversion reflects fundamental cognitive biases shared with other animals rather than uniquely human factors guiding decisions.

Keywords: ambiguity; macaque; risk; risk aversion; risk seeking; uncertainty.

Figure 1

Task and stimuli. (A) Task design. Monkeys had 1 s to freely inspect the bars. When a small yellow fixation square appeared, they had to look at it and maintain fixation for 1 s. The square then disappeared, and they were free to shift gaze to one of the two targets. They then received the appropriate reward and waited through a short inter-trial interval (ITI). (B–F) Examples of stimuli used in this task. (B) Gray bar, certain stimulus, yields 200 μL juice. (C) Examples of risky options. Blue/red bars yield either 333 or 67 μL juice; probability can vary from 0 to 100%. In this example, probabilities of large reward are 50, 88, and 17%, corresponding to the size of the blue portion of the bar. (D) Example of risky option with partially covering occluder that did not render probabilities ambiguous. (E) Ambiguous options. The size of the occluder rendered the bar either low, medium, or high ambiguity (left to right). (F) Examples of stimuli used in triple bar control task.

Figure 2

Ambiguity aversion in monkeys. (A) Behavioral preference for risky over ambiguous options. Dots represent frequency of choosing risky option for each of the 100 individual risky probabilities on risky vs. ambiguous trials. Line is a best-fit cumulative Gaussian. The point of subjective equivalence (PSE, shown by vertical gray dashed line) indicates values at which subjects were indifferent to the two options. The PSE is smaller than 50, indicating that monkeys were ambiguity-averse. (B) Individually, all four monkeys showed ambiguity aversion (PSE significantly less than 50). (C) Ambiguity aversion varied parametrically with degree of ambiguity. Greatest aversion was observed for high uncertainty options.

Figure 3

Humans, like monkeys, preferred options with known probabilities. Conventions as in Figure 2A.

Figure 4

(A) Monkeys did not avoid occluders that did not obscure reward probability information. (B) Monkeys treated a new, magenta occluder the same as the previously encountered cyan occluder, demonstrating that preference do not reflect novelty aversion. (C) Monkeys learned reward probability distributions associated with ambiguous options over time. PSEs for the two most extensively tested monkeys, O (orange) and E (blue), plotted over the course of behavioral sessions. PSE indicates average level of aversion to ambiguous options for a session; values below 50% represent aversion, values at 50% (indicated by horizontal gray line) represent neutrality. Horizontal axis refers to recoding session number. Each session consisted of ∼1500–2500 trials.

Figure 5

Ambiguity aversion is not an artifact of poor stimulus discrimination or heuristic strategies. (A) Stimulus discrimination. The y axis gives the monkeys’ probability of choosing the target with the greater expected value (the bluer bar) on risk–risk trials, as a function of the difference in amount of blue (or red) between the two available targets (x axis). (B) Plot of the probability of choosing the rightward target, as a function of the size of the red and blue portions of the bars on the left and the right (as indicated in the legend to the right of the plot). Horizontal axis corresponds to size of either red or blue bar portion. Vertical axis indicates probability of choosing right-side target. Preferences vary roughly with both blue and red bar portions, indicating that monkeys attend to both in their decisions. Data is smoothed for presentation.