Voice of America: Curtailing Offensive Speech and the Fragility of Common Sense
Aug 20, 2018
Last week I had a very interesting conversation with Michelle Quinn from Voice of America on the struggle of tech companies to curtail offensive speech. Here is the article and the accompanying video. My primary contribution is in highlighting the challenge with real-world operationalizations of concepts like "bias" and "discrimination".
On the one hand, the bias against certain groups is so obvious as to be common sense. On the other hand, get outside of these "common sensical" categories and things get really controversial, really fast. Are rural Americans biases against? What about working class? Is there pervasive bias against caucasians?
I think most of us have heard these and similar claims. So the question I would like to pose is, can we validate these claims in some way? Even if we find some of these claims offensive, and the people making these claims odious, can we nevertheless test these claims rigorously, and either validate or falsify them in a fact-based, data-driven manner?
Since our forthcoming paper on the topic is still under embargo, I will leave things a bit vague here, and just say that I think we are beginning to develop tools that may finally answer these difficult questions. So, as lame as it might sound, stay tuned...
On the one hand, the bias against certain groups is so obvious as to be common sense. On the other hand, get outside of these "common sensical" categories and things get really controversial, really fast. Are rural Americans biases against? What about working class? Is there pervasive bias against caucasians?
I think most of us have heard these and similar claims. So the question I would like to pose is, can we validate these claims in some way? Even if we find some of these claims offensive, and the people making these claims odious, can we nevertheless test these claims rigorously, and either validate or falsify them in a fact-based, data-driven manner?
Since our forthcoming paper on the topic is still under embargo, I will leave things a bit vague here, and just say that I think we are beginning to develop tools that may finally answer these difficult questions. So, as lame as it might sound, stay tuned...
