Too good to be ruled: AI’s Transformative impact on content moderation regulation in the mirror of the rules/standards distinction

This research aims to illuminate an expanding gap between the way online platforms publicly present their content moderation policies (as Rules), and the fashion in which advanced NLP (Natural-Language-Processing) technologies de-facto enforce these platforms’ policies (as Standards). This gap generates several pressing ramifications and highlights severe governance and legitimacy flaws that should be tackled. The Rules vs. Standards debate enjoyed much attention within the modern philosophy of law. Rules were often understood as clear, detailed orders that distinguished the permissible from the impermissible, while Standards were perceived as more general principles that more loosely outlined what is right and what is wrong. The choice between Rules and Standard was understood to also have consequences regarding allocation of power, costs of articulating the legal substance and of litigation, legal certainty, and behavior’s guidance capacity. Many of these considerations are also valid in the digital sphere. During the last few years, online platforms’ content policies have demonstrated a shift towards more Rule-oriented provisions. This trend is reflected, inter alia, in a more detailed and nuanced description of the forbidden content and the sanctions that will be applied to users who violate their policies. Advanced NLP methods are, however, often based on neural network embeddings, and extract the semantic meaning from data in a more abstract fashion, that reflects a Standard-oriented regulation. We aim to explore the transformative power of advanced NLP methods and cutting-edge LLMs (Large-Language-Models) vis-à-vis the nature of platforms’ policies, through the lens of the Rules vs. Standards distinction.