Paul Schrater grew up wasting time in southern California, but mysteriously left to pursue a PhD in Neuroscience at the University of Pennsylvania. At the University of Minnesota, he holds joint appointments in the departments of Psychology and Computer Science. His research combines sophisticated probabilistic modeling applied to problems in human behavior and artificial intelligence with experimental work to test hypotheses about human behavior. Schrater has published more than 75 papers in top venues like Nature and National Academy of Sciencesspanning many disciplines, including human perception, motor control, decision making, motivation, neuroscience, computer vision, robotics, trading agents, machine learning, statistical data analysis, and economics. Key results include explaining key biases in perception, motor control, and decision making in terms of rational statistical learning, which reinterprets puzzling human behavior as predictable and rational learning. His work on computer vision surveillance and trading agents has garnered press attention and the INFORMS ISS Design Science Award, 2012.
Over the last few years, Professor Schrater's research has revolved around providing new answers to what motivates people, treating motivation from the viewpoint of a probabilitistc adaptive control system. By integrating state-of-the-art Hierarchical Bayesian modeling with Reinforcement Learning he has developed a rational account for intrinsic motivation, including curiosity and boredom. These theories are applied to human behavior, both mined from video game play and from controlled experiments. He has given numerous talks, lectures and keynote addresses on this work, including at a White House summit on video games and Neuroscience. During 2012, he was a visiting professor at the University of Geneva, where he worked on developing neural models for key phenomena in motivation.
My group is interested in understanding people's motivations when they are not predicted by external incentives, which are termed intrinsic motivation. For example, It has long been recognized that preferences can dramatically devalue both spontaneously (boredom) or due to external factors. Spontaneous devaluation in music listening is ubiquitous, where yesterday’s hit song is today’s affliction. In active media like video games, devaluation occurs when game elements, performance or other users create frustration and negative affect. From a psychological perspective, devaluation can be viewed as a consequence of violating or failing to sustain intrinsic motivation. Loosely put, intrinsic motivation is the reason we do things without being rewarded. Theories of intrinsic motivation suggest we are internally rewarded by opportunities for improvement in our knowledge and control of the world. They predict that options that are in danger of being lost will increase in value, while options thrust upon them by external circumstance or by another individual will decrease . They also predict that options that are experienced too frequently will devalue due to lost opportunities for novel experiences [2-4]. These effects have direct impact on the overall value of a company or product like Pandora’s automated music selection systems or a video game like Dota2. Since one of the key reasons people play games is to experience empowerment and an increase in locus of control, events that cause a player to perform below their expectations will create frustration and characteristic behaviors associated with frustration including harassing other players, quitting and/or unsubscribing from a game (churning). Similarly, recommender systems that provide too much similarity violate people’s needs for increasing their range of experience, producing boredom and lost customers. The challenge to predicting devaluation is that people’s dynamic motivational state is not easy to directly measure, and the events we wish to capture are relatively rare, thus requiring lots of data. Advances in both mining big data and theories of intrinsic motivation make this problem ripe for solution. More generally, we are interested in predicting what sustains people's interests, including curiosity and the desire for mastery.
Green, C.S., Benson, C., Kersten, D., and Schrater, P. (2010) Alterations in choice behavior by manipulations of world-model. Proceedings of the National Academy of Sciences, 107(37) 16401-16406.
Acuna, D., and Schrater, P. (2010) Structure Learning in Human Sequential Decision-making. PLOS Computational Biology 6(12): e1001003.
Srivastava, N., Kapoor, K. and Schrater, P. (2011) A cognitive basis for theories of intrinsic motivation. Development and Learning and Epigenetic Robotics, 2011
Bavelier D, Green CS, Pouget A, Schrater P. Brain plasticity through the life span: learning to learn and action video games. Annu Rev Neurosci. 2012 Jul 21; 35:391-416.
Srivastava, N. and Schrater, P. (2012) Rational inference of relative preferences. In Proceedings Neural Information Processing Systems, 2012.