Adapting to Change: The Role of Priors, Surprise and Brain Damage on Mental Model Updating by Alexandre Leo Stephen Filipowicz A thesis presented to the University of Waterloo in the fulfillment of the thesis requirement for the degree of Doctor of Philosophy in Psychology Waterloo, Ontario, Canada, 2017 © Alexandre Leo Stephen Filipowicz 2017 Examining Committee Membership The following served on the Examining Committee for this thesis. The decision of the Examining Committee is by majority vote. External Examiner Dr. Suzanna Becker Professor Supervisor(s) Dr. James Danckert Professor Dr. Britt Anderson Associate Professor Internal Member Dr. Stephanie Denison Assistant Professor Internal-external Member Dr. Paul Marriott Professor ii Author’s Declaration I hereby declare that I am the sole author of this thesis. This is a true copy of the thesis, including any required final revisions, as accepted by my examiners. I understand that my thesis may be made electronically available to the public. iii Abstract To make sense of the world, humans build mental models that guide actions and expectations. These mental models need to be receptive to change and updated when they no longer accurately predict observations from an environment. Although ubiquitous in our everyday lives, research is still uncovering the factors that guide mental model building and updating. A significant challenge arises from the need to characterize how mental models can be both robust to noisy, stochastic fluctuations, while also being flexible to environmental changes. The current thesis explores this trade-off by examining some of the main components involved in updating. Chapter 2 proposes a novel task to measure the influence of prior mental models on the way new information is integrated. Chapter 3 tests the role of unexpected, ‘surprising’ events on our ability to detect changes in the environment. Chapter 4 measures the strategies used to explore new mental models, after a change has been detected, and how specific forms of brain damage influence these strategies. The results from this thesis provide novel insights into the behavioural and neural mechanisms that underlie mental model updating. The last chapter situates these results in existing literature, and suggests directions for future research. iv Acknowledgements I would like to begin by thanking my supervisors, Dr. James Danckert & Dr. Britt Anderson, for their tremendous support and guidance throughout my graduate studies. You have both been instrumental in developing my skills as a researcher and I could not have asked for better academic mentors. I would also like to thank the members of my defense committee, Dr. Suzanna Becker, Dr. Paul Marriott, and Dr. Stephanie Denison, for offering to read and comment on my thesis work. Your time and effort are deeply appreciated. To the present/past members of the DAAG Lab and Brittlab, thank you for your years of comments, advice, and camaraderie. I would particularly like to thank Nadine Quehl for all of her help with patient recruitment, and Dr. Elisabeth Stöttinger (my unofficial third supervisor), Dr. Derick Valadao, and Syaheed Jabar for the stimulating (and often challenging) discussions that were crucial in developing to the work presented in this thesis. To my parents, Christine and Stephen, thank you for always encouraging my interests and providing me with an enriching environment in which to grow. To my brother Josef, thank you for keeping me in check. To my children William and Madeleine, thank you for lighting my life and being there when I needed you most. And finally and most importantly, to my wife Candice, thank you for your incredible love and support throughout these often trying times. None of this would have been possible without you. v For June. vi Table of Contents List of Tables ...................................................................................................................... x List of Figures .................................................................................................................... xi Chapter 1: Introduction ....................................................................................................... 1 Chapter 2: The Influence of Priors on Mental Model Building .......................................... 4 2.1. Introduction .............................................................................................................. 4 2.2. Experiment 2.1: Methods ......................................................................................... 8 Participants .................................................................................................................. 8 Plinko task environment ............................................................................................. 9 Experimental conditions ........................................................................................... 11 Measuring accuracy .................................................................................................. 12 Measuring learning rates ........................................................................................... 13 Characterizing participant prior mental models ........................................................ 14 2.3 Experiment 2.1: Results .......................................................................................... 15 Participants exhibit heterogeneous prior mental models .......................................... 15 Prior group performance when exposed to Wide Gaussian distributions ................. 17 Prior group performance when exposed to Narrow Gaussian distribution ............... 22 2.4. Experiment 2.1: Discussion ................................................................................... 24 Chapter 3: The Role of Surprise in Change Detection ..................................................... 27 3.1. Introduction ............................................................................................................ 27 3.2. Experiment 3.1: Methods ....................................................................................... 30 Participants ................................................................................................................ 30 Experimental conditions ........................................................................................... 30 vii Computing surprise ................................................................................................... 31 3.3. Experiment 3.1: Results ......................................................................................... 35 Updating accuracy is worse with no breaks .............................................................. 35 Participants can represent a number of distinct distributions ................................... 40 Participants in the continuous condition demonstrate more hysteresis .................... 40 Surprise positively predicts updating in the continuous condition ........................... 41 3.4. Experiment 3.1: Discussion ................................................................................... 44 3.5. Experiment 3.2: Methods ....................................................................................... 45 Participants ................................................................................................................ 45 Experimental conditions ........................................................................................... 45 3.6. Experiment 3.2: Results ......................................................................................... 49 Updating is worst for low surprise shifts .................................................................. 49 High surprise shifts do not always lead to better updating ....................................... 53 3.7. Experiment 3.2: Discussion ................................................................................... 63 3.8. Discussion .............................................................................................................. 65 Chapter 4: The Influence of Brain Damage on Exploration ............................................. 67 4.1. Introduction ............................................................................................................ 67 4.2. Experiment 4.1: Methods ....................................................................................... 73 Participants ................................................................................................................ 73 Lesion tracing and analysis ....................................................................................... 78 PROBE Task ............................................................................................................. 79 Computational Models .............................................................................................. 84 PROBE model ........................................................................................................... 84 viii Reinforcement Learning (RL) Model ....................................................................... 87 Model fitting methods ............................................................................................... 87 4.3. Experiment 4.1: Results ......................................................................................... 88 Demographic differences .......................................................................................... 88 Behvavioural Performance ........................................................................................ 89 Updating Performance in Recurrent vs Open Sessions ............................................ 92 Good and Poor Updaters ........................................................................................... 96 Model fits of patient behaviour ............................................................................... 104 Model parameter differences between participant groups ...................................... 108 Lesion analyses ....................................................................................................... 109 4.4. Experiment 4.1: Discussion ................................................................................. 113 Chapter 5: General Discussion ........................................................................................ 122 References ....................................................................................................................... 128 Appendix 1 ...................................................................................................................... 143 Model descriptions ...................................................................................................... 143 PROBE model description ...................................................................................... 143 Reinforcement Learning model description ............................................................ 149 ix List of Tables Table 2.1. Mean performance parameters for prior distribution groups. ………………..20 Table 3.1. Experiment 3.2 estimated performance parameters. …………………………39 Table 3.2. Comparison of expected surprise measures for each switch type. ………......48 Table 3.3. Experiment 3.2 estimated performance parameters. …………………………52 Table 3.4. Influence of trial and surprise on likelihood of updating. ……………………62 Table 4.1 Patient demographics for RBD group. ………………………………………..75 Table 4.2 Patient demographics for LBD group. ………………………………………..75 Table 4.3 PROBE and RL Model Parameter Fits. ……………………………………..105 x