Daniel Acuña, CU Department of Computer Science

Abstract: Detecting biases in artificial intelligence is challenging due to the impenetrable nature of deep learning. The central difficulty is linking unobservable phenomena deep inside models with observable, outside quantities that we can measure from inputs and outputs. Current techniques for detecting such biases are often customized for a task, dataset, or method, affecting their generalization. In this presentation, I will discuss a coherent intellectual framework to assess biases in A.I. based on a hundred-year-old field known as psychophysics. The framework effectively reproduces the results of custom-made methods while retaining the ability to import rich psychological literature into A.I. Using science of science, I will discuss the state of A.I. fairness itself and argue that it suffers from a few representational issues.

Bio: Daniel Acuña is a visiting associate professor in the Department of Computer Science at the University of Colorado Boulder. He leads the Science of Science and Computational Discovery Lab. He works in science of science, a subfield of computational social science, and A.I. for science. He writes papers and builds web-based software tools to accelerate knowledge discovery. Daniel has been funded by NSF, DDHS, Sloan Foundation, and DARPA through the SCORE project, and his work has been featured in Nature News, Nature Podcast, The Chronicle of Higher Education, NPR and The Scientist. Before joining CU Boulder, he was an associate professor in the School of Information Studies at Syracuse University and a postdoctoral researcher in neuroscience and mathematical psychology at Northwestern University and the Rehabilitation Institute of Chicago. He obtained his PhD in computer science at the University of Minnesota.