Sarah Dunnett (s4589335) Assessment 1: 'The Cultural Internet' Essay Automated Moderation Systems and the Online Experience Since it's origination in 1983, and the birth of the first social networking site in 1997, the internet, and the way in which it is used, has transformed dramatically. In contrast to early social media; which consisted of small communities with large degrees of anonymity, and operated with only single-mode textual features; modern social media now consists of enormous communities with large degrees of ID transparency, and operate with multimodal features[ CITATION Cam19 \l 3081 ] . These online platforms, that create a space for users to create and share content, have become an intrinsic part of today's society (Gandhi, 2019). In fact, society is uploading and consuming content more than ever before; with over 1.2 trillion photos being shared online in the year 2017. However, with this staggering amount of user generated content being shared online, platforms have becoming progressively more aware of an increase in harmful content being posted by users as a result (Gandhi, 2019). This brought about the need for content moderation, which is the process of monitoring and controlling information shared by applying a set of rules that establishes what is acceptable online behaviour [ CITATION Cam19 \l 3081 ]. Across different platforms, content moderation has formed various systems - such as pre-moderation, reactive moderation, post moderation, and distributed moderation (Papegnies, Labatut, Dufour, & Linares, 2017). Yet, due to the deluge of content being monitored by the human moderators each day, platforms have recently begun developing automated moderating systems through the use of artificial intelligence; as to both look out for the mental well-being of moderators, and to strive towards a cheaper and more efficient model of moderation. Platforms such as Tumblr, Facebook, and Instagram have adopted this model to some degree. However, automated moderation has also proved to be unreliable and subject to algorithmic bias (Delort, Arunasalam, & Paris, 2011). Therefore, despite its few strengths, it can be said that automated moderation has a negative impact on the online experience. Firstly, to understand the detriments of automated moderation, one must first understand the basic mechanics behind it, as to understand how they contribute to the weaknesses of automated moderation. Automated moderation utilises artificial intelligence; constituting of a machine learning algorithm that automatically filters content (Papegnies, Labatut, Dufour, & Linares, 2017). The algorithm is programmed with a universal "blacklist" of words, images, content etc. that are deemed inappropriate. All constructs of communication are then filtered
through this - with all content being process being deemed inappropriate or appropriate according to this standard (Papegnies, Labatut, Dufour, & Linares, 2017). Automated moderation generally takes place in the pre-moderation stage, prohibiting the deemed "inappropriate" content from being shared altogether. However, despite its effectiveness in moderating a specific sector of information, due to the nature processing information through an algorithmic blacklist, automated moderation systems are unable to deduce context and, therefore, are extremely subject to unreliability (Delort, Arunasalam, & Paris, 2011). Pursing this