hide
Free keywords:
-
Abstract:
In this work we provide evidence that shifting preferences upon observing the choices of others largely reflects
Bayesian inference. Inference is performed on the parameters defining one’s own utility function. These parame-
ters can in turn be said to describe the decision-maker’s individual tastes. Temporal discounting parameters are
a paradigmatic example of such tastes, very important in computational psychiatry and economics. High values
of discounting parameters are associated with several psychiatric disorders, lower IQ and poverty. Temporal dis-
counting tasks have good psychometric properties, leading to a well-established hyperbolic model. Somewhat
surprisingly, individual preferences shift in the face of observing the preferences of others even if this shifting is
not itself rewarded. The computational basis of this is unclear. We propose a new model of tastes as (uncertain)
Bayesian beliefs, allowing for one’s tastes to be updated through observing choices made by other, epistemically-
trusted people. Such random tastes may form an important part of random-utility based choices for the individual.
If uncertainty is thus reflected in choice variability, then a key signature of our account is that baseline choice
variability should correlate with the magnitude of apparent preference change. We examined discounting in a
novel community study of 740 young people who made choices between a smaller but immediate, versus a larger
but delayed, reward. They did this both before and after learning about the preferences of a ’partner’. We found
that participants displayed considerable choice variability. The degree of preference shift upon learning about the
partner was correlated with baseline choice variability, lending support to our Bayesian account. Younger people
were influenced by others more than older ones, and this was explained by the former being less certain about
their own preferences. These findings raise the possibility of tastes being subject to Bayesian inference in other
important domains.