Value-based decision making by populations of probabilistic neurons
Value-based choices can vary strongly across time, thus violating axioms of transitivity. Classical behavioral choice models from psychology and economics subsume this variability in some unspecific noise term that has no clearly defined mental or behavioral basis. In this talk, I propose that the variability of value-based choices emerges naturally from the probabilistic nature of value computations instantiated by neuronal populations in the ventro-medial prefrontal cortex. I will first show that distributed patterns of neural activity in the vmPFC, as measured with fMRI, encode probability distributions over stimulus values and that this probabilistic information can be used to derive estimates of both the preferences themselves and of the associated uncertainty. I will then present a biologically-grounded model of these computations that allows organisms to access and exploit the uncertainty associated with their preferences in order to optimally guide value-based choices. This model accurately predicts the outcome and response speed of preference-based decisions and is able to explain the emergence of framing-related choice biases. Taken together, these results support the idea that subjective preferences are encoded as probability distributions derived from the activity profile of value-coding neuronal populations. This proposed coding scheme makes it possible for humans to optimally combine multiple sources of information for decisions and may pave the way for mechanistic explanations of puzzling distortions often observed in economic choices.
Date:
19 May 2016, 13:00 (Thursday, 4th week, Trinity 2016)
Venue:
Tinbergen Building, South Parks Road OX1 3PS
Venue Details:
C113 Weiskrantz Room
Speaker:
Professor Christian Ruff (University of Zurich)
Organising department:
Department of Experimental Psychology
Host:
Dr Matthew Apps (University of Oxford)
Booking required?:
Not required
Audience:
Members of the University only
Editor:
Janice Young
