The Turn to Artificial Intelligence in Governing Communication Online

Abstract

Presently, we are witnessing an intense debate about technological advancements in artificial intelligence(AI) research and its deployment in various societal domains and contexts. In this context, media andcommunications is one of the most prominent and contested fields. Bots, voice assistants, automated (fake)news generation, content moderation and filtering – all of these are examples of how AI and machinelearning are transforming the dynamics and order of digital communication.On 20 March 2018 the Alexander von Humboldt Institute for Internet and Society together with the non-governmental organisation Access Now hosted the one-day expert workshop “The turn to AI in governingcommunication online”. International experts from academia, politics, civil society and business gathered inBerlin to discuss the complex socio-technical questions and issues concerning subjects such as artificialintelligence technologies, machine learning systems, the extent of their deployment in content moderationand the range of approaches to understanding the status and future impact of AI systems for governingsocial communication on the internet.This workshop report summarises and documents the authors’ main takeaways from the discussions. Thediscussions, comments and questions raised and responses from experts also fed into the report. The reporthas been distributed among workshop participants. It is intended to contribute current perspectives to thediscourse on AI and the governance of communication.

Human-Machine Collaboration for Content Regulation

“…One potential solution to minimize such costs is to automate content regulation. Recognizing this, some scholars have begun developing automated (often, ML based) solutions to automate certain aspects of content regulation [28,42,45,68,94,108]. For example, researchers have proposed computational approaches to identify hate speech [18], pornography [98] and pro-eating disorder content [15].…”

Re-humanizing the platform: Content moderators and the logic of care

“…For insights on future developments, we interviewed representatives from three Finnish companies offering AI services for content moderation. With the mounting public pressure to prevent and remove problematic content, companies are turning toward greater automation of content moderation and AI services routinely present their technologies as a catch-all solution to detect and filter hate speech and misinformation (Gollatz et al, 2018). We have closely followed Utopia Analytics, a Finnish company that specializes in automating content moderation, a service which has, for instance, been utilized in the largest Finnish online forum, Suomi24 (Finland24), since 2017.…”

“…Yet, the machine is treated by moderators in a remarkably different way to that suggested by AI promoters, who sell the service of optimizing moderation tasks by delegating them to machines. Moderators are extremely critical of AI systems adopted with an economic logic to reduce human labor, emphasizing, like experts in the field, that the integration of such systems requires carefully designed and implemented human–machine collaboration (Gollatz et al, 2018). Time and again, moderators emphasize that AI systems cannot operate on their own, and that the final responsibility for moderation tasks should always be human-led.…”

“…With the aid of AI, more possibilities are emerging for messages never to be published and undesirable content to be swiftly rejected. Services that seek removal at scale, proactively analyzing all uploaded content pre-publication, are bound to generate false positives, raising concerns about erroneous decision-making and over-policing of content (Gollatz et al, 2018).…”

“…Yet, as an open-ended process, the logic of care is better equipped to deal with unpredictable events, flaws, and uncertainties (Mol, 2008: 95). This kind of open-endedness is particularly important in the current situation, characterized as it is by unresolved legal, political, and cultural tensions, with automated systems transforming the field of content moderation and introducing new kinds of human–machine associations (Gollatz et al, 2018).…”