Monitoring harmful language on social media to ensure the democratic processes in developing countries

Based on practical experience of two recent presidential elections in African countries, the intention of this talk is to share the learnings from this novel work carried out in collaboration with election commissioners – breaking new ground for the continent’s electoral processes. Thereby, we will focus on the following:  The local context - characterising the electoral processes of interest and the political communication within the media landscape in particular relating to the use of social media. The actual language to look for - to identify online behaviour that is disruptive to the conduct of free and fair elections, it is first necessary to determine what these disruptions are and how particular kinds of language can contribute to it. This is the vantage point in defining the concept of harmful speech. The appropriate principles and tools for social media monitoring - principles should not just guide the use of AI but also the application of monitoring social media in the context of elections. With a fast-growing array of AI-based tools on offer, we will discuss how to build a safe and sustainable monitoring tool that fits the needs of local election manager bodies serving an electorate that communicates using languages that are under-represented in many language technologies. What do we see during elections - countries are different and what matters during elections varies substantially. Hence, the language and the local online behaviour must be understood in relation to the local context. We will share examples and stories showing how important this work is and how it can help develop countermeasures that also need to be locally adjusted and flexible.  Call to Action - ultimately, we want to use this session to show how under-explored this space is and where civil society could focus in on to ensure that election transparency can develop in a responsible and structured manner. Equally, this approach does not need to be limited to elections but could take into account wider democratic processes and offer insight into how society at large could benefit from developing critical skills in the use and interpretation of social media. This talk will be facilitated by speakers who can comment on the design of a bespoke AI-based monitoring tool as well as on its roll-out in the countries and the development of the countermeasures. We will use a mix of conversational content as well as graphics and pictures to ensure that audiences with varying levels of technical backgrounds are accommodated. The final part of the session-the call to action-is an interactive component that allows for the audience to engage in the conversation through easily accessible polls and activities.