The mechanisms underlying biases in decision making in humans are described in a new model published online in Nature Communications. This behavioural study highlights the strong links between reward-dependent learning, decision making and attention in human reasoning.
Humans usually combine multiple lines of evidence when making decisions, but are sometimes biased in their judgment about which choice or option is associated with most reward in any given situation. The neural mechanisms that accompany these decision making processes are not fully understood.
Alireza Soltani and colleagues combine behavioural experiments and modelling in order to understand the mechanisms involved in decision making bias when the outcomes of a given choice are associated with different probabilities. In the study, 37 undergraduate students completed a task in which combinations of up to four shapes were presented and associated with different outcomes (rewards). The participants were asked to choose between two targets (represented by the colours red or blue) for each shape combination, each target being associated with a high or low reward. The authors find that the participants made choices that maximized their likelihood of being rewarded and, over the course of 120 trials, learnt the probabilities of reward associated with each shape. However, when asked to guess the likelihood of a reward for each of the shapes, those shapes that rarely delivered rewards were assessed as more valuable by the participants. This is a previously recognized judgment bias known as ‘base-rate neglect’, whereby humans tend to attribute too much predictive power to cues preceding rare outcomes, when presented with multiple cues simultaneously.
In order to explain how this bias occurs, the authors construct a biophysically inspired neural model - incorporating reward-based learning, decision-making and attention processes - and show that the interaction of all of these processes takes place and is essential to explain these counterintuitive behavioural findings.