Computational and Neural Mechanisms Underlying Decision-Making in Humans

How do we make economic decisions in everyday life? How do we make decisions in the face of uncertainty regarding the statistics of the environment? These are the questions that played a pivotal role in the formation of the field "Decision Neuroscience". In each chapter of this thesis, we investigated the computational and neural mechanisms to tackle these questions using behavioral and neural data acquired through fMRI experiments. In the first chapter, we investigated the computational and neural basis of economic decision-making in a binary choice task between two food items. We analyzed behavioral and neural data in a task where participants conducted a sequence of binary choices under the manipulation of fixation-based attention. We developed and calibrated a computational model based on evidence sampling and accumulation to show that the model not only accurately captured basic properties such as choice and reaction time (RT) but also the effect of attentional manipulation in participants’ behavior. We found that the evidence accumulation process predicted by the model to drive a decision was implemented in the areas of frontoparietal network including dmPFC and IPS. These regions also exhibited increased functional connectivity with the activity in vmPFC during choice period where sampled evidence was represented. Our results suggest the involvement of these areas in value-based binary choice. In the second chapter, we examined the computations involved in the decision making under uncertainty. In particular, we aimed to pin down the computations related to temporal change detection. Temporal change detection is the capacity to detect change in the statistics that govern the timing of occurrence of events. We analyzed behavioral data from a novel task where participants observed a sequence of images presented at irregular timings and tasked to detect a change in the frequency of image presentations. We developed and compared computational models from Bayesian to heuristic models and found that all the models captured quantitative aspects of participants’ behavior equally well despite the difference in their computational complexity. Thus, we could not distinguish computations involved in temporal change detection solely from the behavioral data. In the third chapter, we aimed to elucidate the computations involved in temporal change detection from the perspective of neural implementation using fMRI data by leveraging the computational models examined in the previous chapter. We found that the key variable to guide a decision derived from a computationally frugal heuristic model correlated with the activity of the frontalparietal network including dlPFC and IPS, while similar variables derived from more computationally taxing Bayesian models did not show significant correlation with any of the brain regions. Our results suggest that humans might be relying on a simple heuristic model to implement temporal change detection.