Guided Unlearning of Cognitive Pitfalls in Georeferenced Social Sensing

点击翻译按钮获取中文摘要

Social media platforms have accelerated the prosperity of social sensing and made a substantial contribution to the generation and dissemination of big geospatial data for the benefits of science, business, politics and society. At the same time, social media platforms have also become the breeding bed for abuse, manipulation, fake or conspiracy theory. The current research focus is mainly set on learning new knowledge rather than reflecting upon flaws in data and in human minds. This proposal addresses bias-induced cognitive pitfalls in social sensing. For selected application scenarios of georeferenced social media data about Covid-19 pandemic and Climate Change, an interactive platform for guided unlearning of cognitive biases will be developed and prototypically implemented. Unlearning is a radical method of learning. Unlike conventional learning or knowledge accumulation, which is based on the addition of what is new to the learner, an unlearning process is based on the conscious subtraction of something undesirable that already exists, either innately or learned. It is often counter-intuitive and therefore cognitively demanding. Neither a ready-made general recipe nor an empirically feasible technical pathway is available. The project has three objectives: (1) To improve the transparency with regard to the value chain of georeferenced social sensing. Ethically permissible mechanisms will be developed to explain how social media data about a given phenomenon for a given purpose is created, georeferenced, processed, disseminated and shared, and who is involved at which stage of the data flow from upstream to downstream. (2) To support users’ holistic understanding of cognitive biases in georeferenced social sensing. A guided unlearning approach along with a set of visual analytical tools will be designed and tested with conceptualized georeferenced data flows. It allows users to get a comprehensive picture of where and which cognitive biases may creep in and how some seemingly beneficial mental shortcuts for the everyday life may become misleading pitfalls in uncertain and complex situations.(3) To assess users’ capacity of critical reasoning after the training of guided unlearning. A number of guided unlearning experiments will be designed and implemented with biased data for selected real-world scenarios. We will collect preliminary findings and/or raise new questions, relying on two kinds of comparison: between untrained user solutions and benchmark solutions, and between untrained and trained user solutions.