Designing human-machine collaborations – how to effectively interweave human and algorithmic activities
Drawing from my many years of experience of the open peer production system Wikipedia, I introduce various means of quality assurance mechanisms that effectively combine human and algorithmic activities. Over the last two decades, Wikipedia has developed a very effective systems consisting of bots, filters, and interfaces. By using one example of a machine learning-based, i.e., algorithmic, system, I will show possible sources of biases and explain how we can re-interprete these sources into points of participation, where humans are urgently needed to monitor, validate, configure these algorithmic systems. Based on that argument, I infer the concept of human-machine collaboration and introduce first design parameters: level of control, level of coordination, and level of interpretability. For each of these design parameters, I provide examples from our current research (annotation, ideation, visualization). I conclude this talk with an outlook on future research questions and reflect on the societal impact of my research.