Motivation and learning in reward-guided decision-making
Lead Research Organisation:
University of Oxford
Department Name: Experimental Psychology
Abstract
My research investigates the mechanisms of reward-guided decision-making. As part of understanding decision making, one of my projects seeks to understand the mechanisms of intrinsic motivation in humans and non-human primates. Specifically, why do they sometimes suddenly disengage from an activity despite unchanging reward incentives? Likewise, what mechanisms allow them to re-engage afterwards? To answer both questions, I will reanalyse a large selection of datasets in which macaques perform various rewarded learning and decision-tasks. The aim is to identify the behavioural and neural similarities associated with collapses and rebounds of motivation across these different tasks. By building a computational model that accounts for intrinsic motivational states and fluctuations, I hope to capture a side of motivation that so far has not received much attention. This is because it is hard to study experimentally as it, by definition, occurs despite a task design that is meant to prevent disengagements. However, luckily for us, the non-human primates were more prone than most humans to give up when the task didn't seem worth it anymore. Thus, using the large amount of fMRI data, it is possible to look for robust patterns across a variety of tasks that are associated with task (dis)engagement. As such, these datasets provide a unique opportunity to study a particular kind of motivation.
I hope nonetheless that I can also translate these findings to human performance by developing a novel task that can further be used to test the model predictions. In particular, I plan on using online experiments to gather large amounts of data. Combined with manipulations of intrinsic and extrinsic motivation as well as possibilities to "opt-out" of the task, I hope to capture aspects of the same motivational collapse behaviourally.
The last part of my research addresses a related question in the reward domain. Rather than testing fluctuating motivation, I want to find ways for participants to consciously change motivational/reward components in their decision making by using our decision models. Specifically, two of the most important quantities in reward learning are primary reward signals and prediction errors, as well as the resulting rate at which errors lead to learning. I will attempt to change either learning rate or reward valuation, which in standard tasks is not directly observable but can be made explicit using a model that is either applied to participants behaviour or directly to their neural activity. I will then use this estimate to change how participants implement future experience. I expect to change both participants behaviour and the neural mechanisms dynamically by either behavioural or neural feedback or both combined. Recent work in computational psychiatry has shown that the aforementioned variables differ between healthy and clinical populations. It is thus important to investigate the utility of such findings for potential therapies. Additionally, to any therapeutic applications, it is vital to test to what degree these quantities are fixed, but also how participants are capable of consciously altering them, if given appropriate feedback
I hope nonetheless that I can also translate these findings to human performance by developing a novel task that can further be used to test the model predictions. In particular, I plan on using online experiments to gather large amounts of data. Combined with manipulations of intrinsic and extrinsic motivation as well as possibilities to "opt-out" of the task, I hope to capture aspects of the same motivational collapse behaviourally.
The last part of my research addresses a related question in the reward domain. Rather than testing fluctuating motivation, I want to find ways for participants to consciously change motivational/reward components in their decision making by using our decision models. Specifically, two of the most important quantities in reward learning are primary reward signals and prediction errors, as well as the resulting rate at which errors lead to learning. I will attempt to change either learning rate or reward valuation, which in standard tasks is not directly observable but can be made explicit using a model that is either applied to participants behaviour or directly to their neural activity. I will then use this estimate to change how participants implement future experience. I expect to change both participants behaviour and the neural mechanisms dynamically by either behavioural or neural feedback or both combined. Recent work in computational psychiatry has shown that the aforementioned variables differ between healthy and clinical populations. It is thus important to investigate the utility of such findings for potential therapies. Additionally, to any therapeutic applications, it is vital to test to what degree these quantities are fixed, but also how participants are capable of consciously altering them, if given appropriate feedback
Organisations
People |
ORCID iD |
Matthew Rushworth (Primary Supervisor) | |
Jan Grohn (Student) |
Publications
Grohn J
(2020)
Multiple systems in macaques for tracking prediction errors and other types of surprise.
in PLoS biology
Moeller M
(2021)
An association between prediction errors and risk-seeking: Theory and behavioral evidence
in PLOS Computational Biology
Jahn CI
(2023)
Neural responses in macaque prefrontal cortex are linked to strategic exploration.
in PLoS biology
Holton E
(2024)
Goal commitment is supported by vmPFC through selective attention.
in Nature human behaviour
Grohn J
(2024)
General mechanisms of task engagement in the primate frontal cortex
in Nature Communications
Studentship Projects
Project Reference | Relationship | Related To | Start | End | Student Name |
---|---|---|---|---|---|
MR/N013468/1 | 30/09/2016 | 29/09/2025 | |||
2108371 | Studentship | MR/N013468/1 | 30/09/2018 | 29/09/2021 | Jan Grohn |