Who models the modelers? Empirically investigating consequences of computational systems

Momin M. Malik - Carnegie Mellon University

Feb. 23, 2018, 2:30 p.m. - Feb. 23, 2018, 3:30 p.m.


Hosted by: Derek Ruths

Applying models from machine learning and statistics to social systems comes with several caveats. One is that if we do not test and interpret models correctly, for example in making claims on the basis of cross-validation schemes that do not meaningfully simulate real application settings, we might find worse than expected performance or even unexpected outcomes. Another is that model applications may have impacts, and even produce expected outcomes, without necessarily correctly working in "technical terms." I discuss, with examples, how models can impose their own logic on social systems regardless of their success in technical terms, and how we can use careful, reflective modeling to identify and combat instances of mismatches between model claims and outcomes. Lastly, I offer suggestions for how to approach modeling and the design of computational systems in more responsible, robust, and effective ways. 

Momin is a PhD candidate in Societal Computing at the School of Computer Science, Carnegie Mellon University, with backgrounds in history of science and in social science of the internet. His research uses statistical modeling to investigate social science critiques around digital trace data.