Colloquium

Joshua Peterson, Postdoctoral Research Associate, Department of Computer Science, Princeton University

Title:  What Can More Data Offer to Psychology 

Abstract:  Crowdsourcing platforms have greatly expanded our access to large participant populations, allowing us to collect more data than ever before. I argue that scaling data in psychology results in not just quantitative changes, but qualitative changes in the questions we can ask. I explore two projects where this is the case. In the classic domain of decision making under risk, scaling data collection by three orders of magnitude allows us to apply machine learning to obtain better prediction of human behavior than any current theory. We then iteratively constrain the algorithm to uncover what it's learning about human behavior that isn't captured in current theories, yielding a new theory. In a second classic domain, similarity and categorization, leveraging image feature representations learned by deep neural networks along with large human behavioral datasets allows us to evaluate competing theories with naturalistic stimuli and produces different results from previous studies. Ultimately, these new technologies can be used to accelerate scientific discovery and create more generalizable models, just as high-throughput methods have transformed other scientific disciplines.