All models are wrong, some are dangerous

On the one hand, complex adaptive systems (people, social systems and behaviours) are dynamic, messy, ambiguous, non-predictable and non-determinable. On the other hand, machine prediction of such systems has become common place. In this talk I discuss how machine learning models not only are incapable of capturing complex adaptive phenomena but can also be harmful to individuals and communities at the margins of society. I argue that unless we actively work to mitigate it, ML models and large scale datasets inherently embed historical and social stereotypes and injustices.

Click here to join the meeting