• Home
  • University of Queensland
  • Connectivity and culture
  • Content Moderation and Algorithmic Bias in Social Media

Content Moderation and Algorithmic Bias in Social Media

COMU1130 Lecture 2 Content Moderation & Algorithmic Bias Characteristics of Early Social Media: Anonymity (fake names, handles) Single communities (one-one; one-few) Single mode (text based) Limited regulations/rules (lack of consequences) Characteristics of Modern Social Media Real IDs; anonymity Massive communities (many-many) Multimodal (text-audio-video-image-character) Platform-dependent regulations/rules (lack of consequences?) CONTENT MODERATION: method of restriction to internet experience, a set of rules to determine what is and is not acceptable. PRE MODERATION: when a user submits content that is then placed in queue to be checked by moderator before it is published. Slow and time consuming. REACTIVE MODERATION: a 'safety net' in which community members flag content that is inappropriate, undesirable, or in violation of the community guidelines. Subject to apathy, still visible, have to wait for action for it to be taken down. POST MODERATION: all content is published straight away, but replicated in a queue for a moderator to approve or delete afterwards. Slow and time consuming. DISTRIBUTED MODERATION: relies on a rating system which members of the community use to vote on how content aligns with community expectations or the rules of use. Self- moderation. Lots of gaps in moderation, due to differing of interpretation and opinion on what is acceptable. BRIGADING: purposefully invading a topic, thread or communication messaging system by a group of individuals who have organised themselves online with the purpose of manipulating content or its visibility. Secret/undercover communication. Moderation customisations to avoid being taken down. PLATFORM GOVERNANCE: platforms beings hosts vs. platforms being mediators US Communications Decency Act: Section 230 Online intermediaries that host or republish speech are protected against a range of laws that might otherwise be used to hold them legally responsible for what others say/do. Australian Content Laws: "sharing of abhorrent violent material (2019)" New offences for content service providers and hosting services that fail to notify the Australian federal police about or fail remove videos depicting "abhorrent violent conduct" within a reasonable time. Jail time = 3 years AUTO MODERATION: algorithm that monitors and/or filters content. When you remove humans from deciding what's right and wrong, creates a large grey area. "How do you make something so subjective as human culture and make it so objective that its able to fit into a mass algorithm?" ALGOITHMIC BIAS: when (seemingly) inoffensive programming takes on the prejudices either of its creators or the data it is fed. Human language changes so rapidly, so how can algorithms predict people or make decisions about us and how we communicate? Algorithms need to be developed so that they are able to be changed quickly as humans change. Who is in control?