Gaming personality - a dual systems approach to reinforcement learning
When doing my master’s thesis, I was fascinated by learning that algorithms of reinforcement learning (RL) have neurobiological correlates in the brain’s dopamine system. I would like to continue this line of research and attempt to provide new pieces to the puzzle.
In the last year we have seen significant improvements in RL agents’ performance. This is impressive but still missing human abilities like higher-order reasoning as shown by , where fMRI measurements during a task with higher-order structure are better explained by a Bayesian learning component than basic RL.
Using the task from  as our starting point, we propose to construct a cognitive model and train it to replicate their results. The model would be composed of two main systems, where basic relationships are learned by RL and higher order structure by a Bayesian component. Biologically, these systems can be framed as the basal ganglia and cortex.
We will then attempt to improve the model by expanding to human subjects, measuring individual differences using self-report questionnaires (i.e. Big5) to investigate relationships between personality and learning performance similar to  & .
This would open the possibility of creating game AI personalities based on human data. If we find an interested commercial partner, the experimental results from earlier may be compared with player behavioral data and thus infer personality types. We could also have players complete a short personality inventory before playing, which can both be related to their choices during gameplay and be used to personalize the game experience.
Adding EEG measurements would allow us to investigate FRN (feedback related negativity). Using these signals could further improve our model, with the potential for future game developments with player psychophysiological recordings being used within the game environment.
The proposed project would hit several points from IGGI’s research themes and also build upon Professor Pickering’s work in this area. For example, extraversion has been connected with variation in signal strength of FRN which in turn is an indirect measure of reward prediction error in dopamine neurons in the anterior cingulate cortex.
 Hampton, A. N., Bossaerts, P., & O’doherty, J. P. (2006). The role of the ventromedial prefrontal cortex in abstract state-based inference during decision making in humans. The Journal of Neuroscience, 26(32), 8360-8367.
 Stankevicius, A., Huys, Q. J., Kalra, A., & Seriès, P. (2014). Optimism as a prior belief about the probability of future reward.
 Pickering, A. D., & Pesola, F. (2014). Modeling dopaminergic and other processes involved in learning from reward prediction error: contributions from an individual differences perspective. Frontiers in human neuroscience, 8.
Henrik got hooked on video games in the early ‘90s when he caught his dad playing Tetris instead of working. His interest for the human mind began when Henrik was a teenager and became fascinated with dreams, especially lucid dreaming. He intended to learn more through the combination of mathematics and natural science at Lund University, but later went for a more holistic view of the brain as the better approach. Armed with a BSc in Psychology, an MA in Cognitive Science and a life time of playing video games, Henrik has now come full circle and is using the IGGI PhD programme to launch his quest for world domination through game playing AI based on cognitive modelling of human behaviour.
Home institution: Goldsmiths
Supervisor: Professor Alan Pickering
Ready to apply?
Once you have identified your potential supervisor, we would encourage you to contact them to discuss your research proposal.Learn More