Neural mechanisms of reward-guided learning and irrational decision-making

The ability to take effective decisions is fundamental for successful environmental adaptation and survival. In this thesis, I investigated situations in which decisions appear irrational, at least from certain standpoints. I conducted a behavioural decision-making experiment in two groups of macaques: controls and a group with ventromedial prefrontal cortex/medial orbitofrontal cortex (vmPFC/ mOFC) lesions. Some choices lead to compound outcomes composed of different constituent parts. Control macaquesâ decisions suggested their estimates of the value of the compound were biased away from the sum of the values of the constituents and towards their mean. Lesions of vmPFC/mOFC diminished the size of the effect so that macaques in some ways appeared to make more rational decisions. Based on the results of this experiment I devised a similar Functional Magnetic Resonance Imaging (fMRI) paradigm with the control animals. This demonstrated strong vmPFC/mOFC activity when similar decisions were made and suggested a value comparison process. In addition, I investigated the role of dopamine in learning using Fast-Scan Cyclic Voltammetry (FSCV), while rats performed a simple decision-making task. Theories about the role of dopamine in learning have focused on the possibility that it codes scalar reward value prediction errors. Less consideration has been given to the possibility that dopamine might reflect prediction errors about reward identities regardless of value. I measured dopamine in the nucleus accumbens when unexpected changes in reward value or identity occurred while rats executed a two-choice two-reward instrumental task. Dopamine levels in the nucleus accumbens reflected reward value prediction errors. In addition, however, they also reflected some information about reward identity under some circumstances. Further investigation suggested that this might be due to differences in the nutritional value of different reward types that did not have clear measurable impacts of behaviour in the tasks that I used.