Thursday 27 August, 12pm-1pm AEST
Internet platforms with vast amounts of behavioral data commonly predict human behaviors.
Predictions are sold to third parties who utilize them for personalisation, targeting and other decision-making. Because better predictions translate into higher financial value, platforms are incentivised to reduce prediction errors. Beyond improving algorithms and data, platforms can stealthily achieve 'better' predictions by 'pushing' users' outcomes towards their predicted values, using behavior modification techniques, thereby demonstrating more certain predictions.
This strategy is absent from the machine learning and statistics literature. Professor Shmueli's team integrate causal inference notation with correlation-based prediction in order to formalise and analyse how behavior modification results in 'improved' prediction errors. Their discoveries should alert data scientists and purchasers of prediction products, and raise moral, and societal concerns.
About the speaker
Professor Galit Shmueli is Distinguished Professor and Director, the Institute of Service Science, College of Technology Management, National Tsing Hua University (NTHU), Taiwan. Prior to that she was the SRITNE Chaired Professor of Data Analytics and Associate Professor of Statistics & Information Systems at the Indian School of Business, and Associate Professor at University of Maryland's Smith School of Business. She is the editor-in-chief of INFORMS Journal on Data Science. Her research focuses on statistical and data mining methods for contemporary data structures, with a focus on statistical strategy – issues related to how data analytics are used in scientific research. Her main fields of application are information systems, electronic commerce, biosurveillance and healthcare, with a focus on human behavior. Originally from Israel, Professor Shmueli has worked in the US, Bhutan, India and now Taiwan.