Hanna Wallach
This talk will be structured around four talking points -- intended to prompt discussion -- that lie at the heart of fairness and transparency in machine learning: data, questions, models, and findings. In discussing these talking points, I will touch upon limitations of the most prevalent definitions of big data; the need for true collaborations with social scientists when analyzing social data; data-driven research vs. question-driven research; convenience data vs. carefully selected data; transparency and algorithmic accountability reporting; models for exploration/explanation vs. models for prediction; representing and maintaining uncertainty; error analysis; intuition and bias in interpreting findings; and, finally, the importance of scientific communication.