Content moderation

Content moderation determines the type of data displayed on platforms. Although this type of work is conducted online without interpersonal interactions, it does not remain emotionless. This article presents findings from a longitudinal qualitative study of how content moderation is conducted on a UK-based platform that publishes patients' stories of their health experiences. Drawing upon the literatures on emotion work in organizations and on content moderation, this paper addresses the questions: what is the nature of emotion work that is undertaken during content moderation and how does emotion work play out with control on online platforms? Although management studies have made extensive contributions to platform work, content moderation remains under-theorized, despite its significance in determining the type of information platforms exhibit. The paper contributes theoretically and empirically to this field in three ways. First, it categorizes the work practices moderators undertake on a health platform, which encompass application of rules, quantification, objectification, verification, and care. Second, our paper shows that content moderation is constitutive of emotion work and goes on to identify various types of emotion work, including managed and unmanaged authentic emotions. Third, our paper offers a model that illustrates how emotion work constitutes a sense-making mechanism that allows moderators to exercise discretion and undertake care work, bypassing the prescriptive regulatory environment of content moderation, and sourcing occasional autonomy.

Background: Over the past two decades, online platforms in Europe have shifted from being relatively passive intermediaries to powerful actors that influence information flows, public opinion, and digital markets. Early voluntary initiatives failed to deliver consistent transparency or accountability, which led the EU to introduce binding frameworks-the Digital Services Act (DSA) and Digital Markets Act (DMA). Both build on the General Data Protection Regulation (GDPR) and national laws such as Germany's NetzDG, with the shared goal of tackling societal risks and limiting market dominance. Their effectiveness, however, depends not only on legal provisions but also on whether users can interpret and apply the information these rules require platforms to provide. Materials and Methods: The study combines analysis of EU legislation, regulatory guidance, and academic literature. It traces the transition from self-regulation to binding legal regimes and examines how the DSA and DMA approach transparency, user rights, and systemic risk management. The assessment also considers overlaps with other legal instruments, national enforcement disparities, and the role of digital literacy in making transparency measures effective. Results: The DSA and DMA introduce stronger safeguards-including clearer explanations of algorithmic processes, opt-out possibilities, and structured risk assessments. Yet practical barriers remain: national enforcement capacity varies, compliance costs weigh more heavily on smaller platforms, and technical disclosures often fail to reach or engage the average user. Without communication formats tailored to different literacy levels, transparency may end up as a formal requirement rather than a tool for user empowerment. Conclusion: By replacing voluntary commitments with enforceable standards, the DSA and DMA mark a turning point in EU platform governance. Their long-term success will rely on ensuring that transparency is accessible and actionable, supported by consistent oversight and targeted education so that all users-not only the digitally skilled-can benefit from the protections these laws intend to provide.
This work was supported by the National Research, Development and Innovation Fund, provided by the Hungarian Ministry of Culture and Innovation, under Grant NKKP STARTING_24 (STARTING 150399) programme.

Background: The rapid expansion of social media has reshaped global information consumption and dissemination. In countries like Iran, where traditional media are often state-controlled and public trust in official sources may be limited, platforms such as Telegram, Instagram, and X (formerly Twitter) have become alternative hubs for news, opinion, and social discourse. However, this shift has introduced a significant challenge: the widespread circulation of fake news and misleading information, often disguised as credible content. As users increasingly rely on these platforms for real-time updates, distinguishing fact from fiction becomes more difficult, accelerating the spread of misinformation, which can influence public opinion, behavior, and even social stability. Objective: This study has two primary objectives: (1) to identify and categorize the structural features distinguishing fake information from real information on Persian-language social media, and (2) to examine how these features influence Iranian users' sharing behavior. Moving beyond content verification, the study explores how form, presentation, and rhetorical structure shape user perceptions and actions, particularly in emotionally or politically charged contexts. Methodology: A mixed-methods approach was employed. First, a qualitative content analysis was conducted on a corpus of verified fake and real social media posts to identify recurring structural elements, including headline style, emotional tone, source attribution, visual use, narrative coherence, and rhetorical strategies. Second, a survey was distributed among a diverse sample of Iranian social media users to quantitatively assess how these structural features influence sharing behavior. The survey incorporated scenario-based questions, self-report scales, and demographic variables (e.g., age, education, media literacy) to account for contextual factors. Findings: The analysis revealed distinct structural differences between fake and real information. Fake content frequently employs emotional triggers, hyperbolic language, fear-based appeals, and pseudo-expert citations to create an illusion of credibility. It tends to be narrative-driven, lacking verifiable data, yet remains persuasive due to its affective tone and simplified storytelling. In contrast, real in-Quarterly Journal of Virtual Space and Social Media Summer 2025 10 ‫اجتماعی‬ ‫های‬ ‫رسانه‬ ‫و‬ ‫مجازی‬ ‫فضای‬ ‫فصلنامه‬ formation prioritizes factual integrity, source credibility, consistency, and a neutral or balanced tone. However, structural sophistication alone does not ensure user skepticism. Users were significantly more likely to share fake content if it aligned with their emotions, preexisting beliefs, or perceived urgency-even when lacking credible support. This pattern highlights the role of cognitive bias and affective influence in digital sharing behavior. The study demonstrates that the structural composition of information-its presentation, tone, and rhetorical form-plays a critical role in users' engagement and sharing decisions. This influence often outweighs factual accuracy, particularly in fast-paced, emotionally charged digital environments. Addressing these structural cues is essential for developing effective strategies to combat misinformation. Implications: The findings offer valuable insights for media policymakers, platform developers, educators, and fact-checking organizations. They underscore the need for structural awareness training in digital literacy programs and suggest that content moderation should extend beyond fact-checking to identifying high-risk structural features. Additionally, the study recommends that social media platforms implement design-based nudges (e.g., contextual warnings) to alert users to emotionally manipulative content, fostering more deliberate sharing practices.

ምልከታ ላይ የተመሰረቱ ዘዴዎች (Observational methods) -ተዓማኒነት እና ትክክለኛነት (Reliability and validity) -ውስጣዊ እና ውጫዊ ትክክለኛነት (Internal and external validity) -የፍላጎት ባህርያት እና የሃውቶርን ውጤት (Demand characteristics and the Hawthorne Effect) -የምርምር ሥነ-ምግባር (The ethics of research) 3 ፐርሰፕሽን እና ትኩረት -የእይታ ፐርሰፕሽን ንድፈ-ሐሳብ (Visual perception theory) -የጌስታልት የፐርሰፕሽን ንድፈ-ሐሳብ (The Gestalt Theory of Perception) -የእይታ ትኩረት (Visual attention) -የተመረጠ ትኩረት እና መስማት (Selective attention and hearing) -የህመም ፐርሰፕሽን (Perception of pain) 4 ትውስታ (Memory) -ኮድ ማድረግ (Encoding) -ማከማቸት (Storage) -ማስታወስ (Retrieval) -የትውስታ ዓይነቶች (Types of memory) -የትውስታ ውድቀቶች (Memory failures) -የምላስ ጫፍ ክስተት (Tip-of-the-tongue phenomenon) • በትውስታ ላይ ያሉ ተጽዕኖዎች፡ ሽታ፣ ስሜቶች እና ውጥረት • የሐሰት ትውስታ (False memory) መማር (Learning)

The need for Content moderation is indispensable as publishers can't allow inappropriate or illegal content to be published through or on their platforms. Also, every publisher has the responsibility of checking works to ensure that the content of the results is ethical for readers, and viewers can get clean and explicit content. Machine moderators may be the all-important response to the growing need for moderators. The push for machine moderators is often justified as a necessary response to the scale of work supposed to be edited. The use of machines, Artificial Intelligence (AI) will make content editing faster and possibly more accurate than human moderators. This work looks into the meaning of content moderation, how machines are used for content moderation and the advantages of using machines in content moderation. It also highlights the significant innovations in the use of machine moderators in content management. The simulation of Human Intelligence in machines programmed to think like humans and mimic their actions is referred to as artificial intelligence (AI). The term may also be used for any machine or computer that exhibits characteristics resembling the human mind, such as problem-solving and learning. The optimal feature of artificial intelligence is defined as the ability to rationalize, learn and take actions that have the highest potential to be most beneficial for the intended goal and objectives.

The computerization of public administration tasks is a reality. In contrast, the intelligence of public administration is shrouded in myths. For many decades, administrative science has contributed to the clarification of this distinction. Digital constitutionalism and technology-oriented administrative law doctrine have recently been added to this research. The basic regulations, proposed and adopted within individual states, in the European Union and in international organizations, whether it concerns the protection of personal data, cyber security, or artificial intelligence, do establish new tasks for public administration, but they affect methods rather than forms of administrative activity. Emerging technology raises concerns about the ability to understand artificial reasoning and its methods of classification, personalization, and prediction. It is questionable to assume that all actions can be quantified and thus everything becomes objective. Technology compounds the situation and has its own imperative.

Content moderation by social media platforms, once hailed as the holy grail for stamping out illegal content such as hate speech and disinformation, is failing in the Global South. Global conversations on content moderation and governance have primarily centred around an individualistic conception of human rights, largely dominated by responses grounded on international human rights law (IHRL). While IHRL is a dominant normative framework on content moderation, it often invisiblises (and ignores) the concerns of users in the Global South, mainly in Africa. By analysing insights from Africa and incorporating Third World Approaches to International Law (TWAIL), this Article explores the limitations of IHRL as a governing framework for content in Africa, which stem from its Eurocentric legal foundations, epistemic and linguistic blind spots. The Article ultimately advocates for an African approach to content moderation, anchored in a communal conception of human rights as articulated in African human rights law.

This article examines how artificial intelligence and social media algorithms have shifted from neutral conduits to active forces in shaping historical memory, driving a new wave of digital revisionism. Through case studies including Holocaust denial, synthetic deepfakes, and anti-Israel disinformation, we show how engagement-optimized platforms privilege sensational content, amplifying distortion and marginalizing authentic testimony. AI systems trained on biased or incomplete data fracture collective memory, rendering Jewish history a battleground where truth competes with algorithmic spectacle. Generative tools now fabricate plausible “evidence” and rewrite genealogies, undermining trust in archival integrity. The analysis underscores that safeguarding historical fidelity in the digital age requires interdisciplinary collaboration—uniting transparency in AI, ethical design, and grassroots activism—to resist the unseen agency of algorithms in manipulating the past.

This paper examines the evolving roles of states in content moderation and their complex relationships with social media platforms over time. As these platforms have become central to global communication, their interactions with governments have shaped content moderation in increasingly intricate ways. Despite the significance of these relationships, their chronological development and their implications remain insufficiently explored. This paper addresses this gap through an analysis of the annual reports of the United Nations' Special Rapporteurs (SRs) on freedom of opinion and expression from 1995 to 2024. It traces the progressive entanglement between states and platforms, highlighting how governments have increasingly relied on platforms to enforce content moderation through both formal and informal mechanisms. The paper makes two key contributions: first, it provides a chronological account of how state involvement in content moderation has evolved; second, it categorizes types of state-platform relationships, distinguishing between formal and informal modes of influence as well as top-down, bottom-up, and collaborative interactions. Finally, it touches upon the geopolitical issues relating to state-platform relationships in a context where it is unclear whether platforms and governments share the same values of freedom and democracy anymore. By offering an empirical and chronological perspective, this study contributes to ongoing debates on platform governance and the issue of social media regulation.

As AI started to produce, curate, and moderate content for an ever-expanding class of digital environments, the principles of fairness, transparency, and accountability, hallmarks of traditional due process, were thrust into the public spotlight. Driven by increasing regulatory scrutiny from the GDPR and the upcoming EU AI Act, alongside public demand for ethical AI systems, this paper seeks to mitigate the urgent need for bias in AI-generated content. This covers already known frameworks and methodologies to detect and correct such biases in processing and postprocessing data, including pre-processing. Furthermore, the paper also touches upon the application of Explainable AI (XAI) in content moderation. It reflects upon how the employability of interpretable models can lead to trust and allow audits on automated decisions. Building on the aforementioned fairness frameworks and XAI tools, this work presents a dual-layered approach to ethical AI, delivering equitable outcomes and transparent justifications. Grounded in both technical and policy-oriented approaches, the study lays a roadmap for achieving trusted, accountable, human-centric AI systems.

To address the limitations of self-regulation and the need to combat online misinforma- tion, domestic policies are increasingly imposing platform liability for content moder- ation. This study employs qualitative comparative analysis to examine five key national legislations: Germany’s NetzDG, France’s Law No. 2018-1202, Brazil’s Resolution No. 23.732/2024, Singapore’s POFMA, and Turkey’s Law No. 2022-7418. Guided by the UN’s and UNESCO’s human rights-based recommendations for platform governance, the analysis focuses on five dimensions: definitions of content and misinformation, mod- eration practices, transparency requirements, penalties, and independent oversight. Our findings reveal variations on how misinformation is defined, with most jurisdictions adopting vague formulations. Only Brazil’s resolution explicitly addresses AI-generated content. NetzDG emphasises platform-led enforcement; French and Brazilian jurisdic- tions rely more on judicial orders; POFMA and Turkey’s law grant discretionary powers to state authorities. Independent oversight – a key safeguard for human rights – is for- malised only in France (Arcom) and Germany (regulated self-regulation). Although Turkey designates the BTK as an oversight body, its independence is widely contested. Without an independent regulator, Brazil’s resolution allows judicial assessments to draw on verifications conducted by accredited fact-checkers.

NDAs, originally intended to protect intellectual property, have now evolved into a tool that enable labor repression on data supply chains. They suppress organizing, shield exploitation from scrutiny, and enable tech giants to profit from a system built to conceal harm, allowing companies to claim ignorance while benefiting from abuse.

This paper engages with the problem of toxic speech online and suggests remedies inspired by the value-sensitive design literature (VSD), suggesting that the designers of online platforms should explore methods of adding friction to online conversations. Second, this paper examines a historical case of designing a communications platform to offer methods to users to inculcate norms of acceptable behavior by introducing friction into synchronous conversations. This is the case of America Online (AOL) Instant Messenger, also known as AIM, which included a feature whereby users could "warn" other users, attaching a cost to, and thus disincentivizing, certain kinds of speech. The nuances of the design of this feature make it especially well-suited as a subject of study in value-sensitive design as it seems to be the product of significant reflection and foresight by its designers. In the course of examining this case, this paper proposes two novel and generalizable processes of integrating values into the design of technology, inspired by the approach of valuesensitive design: a "method of decomposition," reconstructing a user journey in order to identify possible moments of intervention; and an iterative "Innovation-Abuse-Innovation" branching diagram, which systematizes the process of anticipating abuse cases and designing responses to them. These methods build upon recent work in the literature on operationalizing ethical values in the design process. I close by illustrating the flexibility and generalizability of these methods and speculating on how they might be applied to contemporary platforms.

This brief exposes the systemic abuse faced by food delivery riders and data workers—such as content moderators and data labellers—who are hired through third-party logistics (3PL) and outsourcing (BPO) firms. Drawing on interviews with over 200 workers across the Gulf, Africa, Latin America, and Asia, it reveals how global platforms like Meta and OpenAI outsource labour to evade responsibility, resulting in widespread forced labour risks, unsafe conditions, wage theft, and retaliation against those who try to organise.

The brief urges the International Labour Conference (ILC) and national regulators to recognise these workers under international labour standards, hold lead firms jointly accountable with their intermediaries, prohibit the use of NDAs to silence abuse, and extend full protections for occupational health, safety, and the right to organise.

‘Scroll. Click. Suffer.’ is an in-depth investigation into the hidden workforce behind AI and social media—data labellers and content moderators. Based on interviews with 113 workers across Colombia, Kenya, and the Philippines, the report exposes the extreme occupational, psychological, sexual, and economic harms faced by those moderating violent content and training AI models for platforms like Meta, TikTok, and ChatGPT.

From PTSD and substance dependency to union-busting and exploitative contracts, it reveals an industry architecture that normalises harm and conceals abuse. Tech giants outsource risk down opaque, informal supply chains, while workers are silenced by NDAs, punished for speaking out, and denied basic protections.

Grounded in worker testimonies and legal analysis, the report outlines a clear roadmap for reform—and calls on lead firms, investors, governments, and the ILO to centre the rights of digital workers.

On March 11, the African Commission on Human and Peoples' Rights (African Commission), an organ of the African Union (AU), adopted a landmark Resolution 630/2025, expressing concern over recent regressions in content moderation and reduction of fact-checking services by social media platforms-developments that undermine information integrity and the protection of free expression and access to information in Africa and beyond. This resolution emerged after major social media platforms such as Meta and YouTube recently announced a 'Community Notes' system to tackle disinformation, which, in turn, is cutting back on professional fact-checking services on their platforms. This blog post first scrutinizes social media platforms' latest rollback in ending professional fact-checking services and replacing them with the Community Notes system. It then explores whether and how such a move may undermine their due diligence obligations under African and international human rights law.

The paper, 'Regula(ng the Undefined: Addressing Systemic Risks in the Digital Services Act (with an Appendix on the AI Act)' , examines the DSA and its risk assessment and management processes. In this commentary, I aim to highlight the significance of the DSA from a computer ethics perspecDve, explore its limitaDons -some of which are addressed by Loi et al., while others are newly idenDfied -and propose potenDal soluDons in addiDon to those outlined in the original work. To begin, the DSA places new responsibiliDes on major online plaJorms and search engines, which include evaluaDng and managing systemic risks associated with their services and providing jusDficaDons for their methods of addressing those risks. Specifically, the DSA seeks to combat harmful internet content, including conspiracy theories and dangerous and hate speech (European Parliament & Council of the E.U., 2022). 2 However, what precisely differenDates dangerous speech from hate speech, and which ethical challenges do arise?

This comprehensive review examines the intersection of artificial intelligence (AI) and pornography, analyzing how AIdriven technologies such as deepfakes, recommendation systems, and content moderation tools are reshaping the adult entertainment industry. While AI introduces efficiencies in content creation and personalization, it also generates significant ethical, psychological, legal, and societal challenges. The proliferation of non-consensual deepfake pornography raises urgent concerns about consent, privacy, and image-based sexual abuse. AI's role in influencing user behaviour, reinforcing unrealistic sexual norms, and altering perceptions of intimacy is explored through psychological and media effects theories. Additionally, the paper highlights gaps in global regulation, inconsistencies in legal enforcement, and the urgent need for longitudinal and intervention studies to assess the real-world impacts of AIenhanced pornography. Future directions emphasise the development of ethical frameworks, robust technological safeguards, and interdisciplinary research to guide responsible innovation and protect human dignity in digital environments.

This paper argues that an AI judge is conceptually undesirable and not just something that lies beyond the state of the art in computer science. In a nutshell, even if an AI system could accurately predict how a good human judge would decide a particular case, its prediction would be the product of correlations between such factors as patterns of syntax in bodies of legal texts. This approach of AI systems is insufficient for basing their output on the sort of rationales that are expected of valid judicial decisions in any desirable legal system. Thus, by their very nature, AI systems are incapable of providing valid legal decisions in any such system.

While content moderation began as a solution to online abuse, it has increasingly been framed as a source of abuse by a diverse coalition of users, civil society organisations, and politicians concerned with platform bias. The resulting crisis of legitimacy has motivated interest in more participatory forms of governance, yet such approaches are difficult to scale on platforms that lack bounded communities and designated tools to support collective governance. Within this context, we use a high-profile debate surrounding bias and racism in content moderation on YouTube to investigate how creators engage in meta-moderation, the participatory evaluation of moderation decisions and policies. We conceptualise the conversation that plays out across a network of videos and comments as aspirational platform governance, or the desire to influence content moderation without established channels or guarantees of success. Through a content analysis of 115 videos and associated online discourse, we identify overlapping and competing understandings of bias, with key fault lines around demographic categories of gender, race, and
geography, as well as genres of production and channel size. We analyse how reaction videos navigate structural factors that inhibit discussions of platform practices and assess the functions of aspirational platform governance, including its counter-intuitive role in legitimising content moderation through the airing of complaints.

The growth of the digital economy has resulted in unprecedented increase of the amount of data generated by mankind. This vast volume and variety of data is a significant source of knowledge that allows researchers to open new research fields and to analyse existing problems more precisely. It is particularly important for scientists employing machine learning techniques in their studies. From

In this article, we argue that YouTube’s algorithm should be programmed to make a modest but significant percentage (e.g. 2%) of recommendations for the climate. Just as a librarian has a (meta-editorial) responsibility to highlight certain titles and not others, we believe that so should YouTube’s algorithm. The company, we argue, has duties of content moderation, reparation and meta-editing, as well as strong consequentialist reasons to program its algorithm to do so. With 2 billion users, our proposed intervention could be an effective contribution to mitigating the climate crisis in a transparent and accountable way. We consider different setups, with varying degrees of transparency and centralization. We then address the worries that such a project may raise: the risk of manipulation, the threat of a slippery slope, and the concerns for freedom of expression. We conclude that none of these elements seriously undermine the desirability of our proposal.

This academic paper explores how social media platforms regulate free speech through content moderation policies and the ethical implications of their approaches. It examines the use of automated and human moderation systems, highlighting challenges such as transparency, biases, and enforcement inconsistencies.

My previous post explored the many ways in which regional systems serve as important forums for developing digital rights-specifically, their role in promoting culturally sensitive human rights, supporting local and context-driven enforcement, and advancing subaltern norms. I also discussed key normative developments in the African human rights system, funnelled through the African Union (AU), the African Commission on Human and Peoples' Rights (ACmHPR), and regional courts. In this post, I focus on Inter-American perspectives on digital rights and propose regional approaches as a complement to international human rights law's universalistic vision of digital human rights.

Contemporary legal and policy discussions around the social and political implications of algorithmic decision-making and digital governance have increasingly revolved around an aspiration for 'fairness'. 1 While this aspiration may seem at first sight to be a relevant ideal for both law and technology in digital governance, there are political and distributive implications at stake in taking fairness as a given, especially when left undertheorized. The language of fairness is used by a multiplicity of actors in global digital governance and thus serves a wide variety of purposes, from the normalization and stabilization of problematic practices, to attempts at constraining and resisting them. Fairness is an inherently pluridimensional concept within the international law discipline: the debates and discussions

Can journalism become a forum, equating it with the freedom of expression of every citizen, as a good fused in the collective intelligence benefited by virtual crowds?
Is it the role of journalism to accommodate all the ideas that society brings, when the media has a completely different task, from all these ideas, to select those that are based on facts?
Can the media remain withdrawn in its discretion not to infect the facts and events with excess coverage, precisely not to do the same thing that happens in social networks, where from the overdose of communication, nothing can be distinguished anymore, it is lost news, lose their goals, and degrade from virtuoso to vicious?
Can the market of ideas exist without the market of facts, without the language of factuality, of scientific validation?

The activities of large online platforms based in third countries in the internal market pose potential risks to EU users. The EU aims to ensure a safe online environment not only for consumers, but also for all users active in this ecosystem. Increased security, legal certainty, consumer protection, transparency, and several other partial aims have led to the adoption of the Digital Services Package, which includes the so-called DSA Regulation. The present article aims to identify the key impacts of the new regulation on very large online platforms that are part of the daily routine of EU citizens and to highlight the benefits it brings to regular users. There are many changes brought about by the new legislation; therefore, we decided to focus only on those that we consider the most tangible, both from the perspective of the everyday user and for the platforms per se.

A presente pesquisa é desenvolvida de forma a investigar se o propósito de garantia da liberdade de expressão é afirmado na arquitetura operacional das plataformas de mídias sociais ou se, diversamente, ocorre a flexibilização do referido direito fundamental à luz das diretrizes de uso.

This article examines the provisions of Articles 18 and 19 of the European Media Freedom Act (EMFA) in relation to the digital governance framework established by the Digital Services Act (DSA), with a particular focus on the media privilege codified in Article 18 of the EMFA. Section 2 analyses the media privilege of Articles 18 and 19 of the EMFA from three perspectives: the requirements for obtaining this privilege and the media entities eligible for it; the functioning of the privilege and its intersections with the DSA provisions; and the transparency in its appli-cation, including the role of the Board under Article 19 of the EMFA. Section 3 explores the regulatory framework governing  the  media  privilege  in  Article  18  of  the  EMFA  and  addresses  criticisms  of  this  online  media  privilege.  More specifically, it focuses on two main criticisms: the risk of granting this privilege to agents of disinformation and the “constitutional” legitimacy of the privilege. The final remarks summarise the main considerations of the different sections and offer concluding observations.

In this post, we seek to explore the ways in which the CAP treats another important aspect of the applicability of international law in cyberspace – the CAP’s approach to the application of international human rights law in cyberspace and, in particular, to the emergence of digital human rights.

El presente ensayo es resultado de un proyecto de colaboración entre el Instituto de Investigaciones Jurídicas de la Universidad Nacional Autónoma de México y la Asociación de Internet México (AIMX). El objetivo de la investigación fue explorar el ambiente regulatorio de la moderación de contenidos en Internet, identificar las características que hacen compleja la intervención de actores en la moderación de contenidos y proponer un planteamiento práctico para comprender el alcance de las intervenciones o iniciativas normativas o técnicas por parte de múltiples actores a través de las plataformas, dentro de un marco de afectación controlado por estándares globales de comportamiento en las prácticas de expresión y discursos públicos. Nuestro estudio se dirige principalmente a quienes pueden proponer iniciativas legislativas, regulatorias o judiciales, pero también hacia otros mecanismos como los internos de moderación de las plataformas, o nuevos negocios y modelos.

هدف از این پژوهش نقد به رویکردی است که راهکار برطرف شدن چالش‌هایِ اخلاقیِ هوشِ مصنوعیِ را محدود به طراحی و اصلاحات فنی می‌داند. برخی پژوهش‌گران چالش‌های اخلاقی در هوش مصنوعی را همگرا تلقی می‌کنند و معتقدند این چالش‌ها همانطور که با ظهور سیستم هوش مصنوعی پدید آمدند، با پیشرفت و اصلاحات فنی آن مرتفع خواهند شد. در مباحثِ اخلاقِ هوش مصنوعی، موضوعاتی همچون حفاظت از حریم خصوصی و شفافیت در بیشتر پژوهش‏ها مورد توجه قرار گرفته است. در پژوهش حاضر با استفاده از روشِ تحلیلی-انتقادی، همگرایی میان شفافیت و حریم خصوصی در سیستم‌های یادگیری ماشین بررسی شده و رویکردی که رفع چالش‌های اخلاقی را به طراحی هوش مصنوعی محدود می‌داند به نقد گذاشته شده است. نتایج این بررسی نشان می‏دهد که همگرایی میان چالش‌های اخلاقی هوش مصنوعی وجود ندارد. همچنین با مطرح کردن سه چالش کمی‏سازی، محدودیت‏های تکنیکی و مباحث فرااخلاقی به نقد رویکردهایی که تنها بر طراحی و اصلاحات فنی متکی هستند پرداخته شد. نتایج این بخش نشان می‌دهد که نمی‌توان تنها با تکیه بر طراحی، به چالش‌های اخلاقی در هوش مصنوعی پاسخ داد و نیاز به استفاده از سایر روش‌ها مانند قانون‌گذاری و یا پیشرفت در سایر علوم در کنار توجه به طراحی وجود دارد.
This research aims to criticize the approach that considers the solution to the ethical challenges of artificial intelligence (AI) to be only in design and technical improvements. Some researchers consider the ethical challenges in AI to be convergent, which emerge with the advent of AI systems and are resolved with technical progress and improvements. In the discussions of the ethics of AI, issues such as privacy and transparency have been the focus of most studies. In the present research, using the analytical-critical method, the convergence between transparency and privacy in machine learning systems was investigated, and the approach that limits the resolution of ethical challenges to AI design was criticized. The research findings indicate that there is no convergence between the ethical challenges of AI. Additionally, by raising three challenges of quantification, technical limitations, and meta-ethical issues, the criticism of approaches that rely solely on design and technical improvements was addressed. The results also indicate that it is not possible to respond to the ethical challenges of AI only by relying on design, and there is a need to use other methods such as legislation or progress in other sciences in addition to attention to design.

This essay is my contemplation on the labor conditions of migrants in Germany through my job experience as a content moderator. Structured as a metaphorical trilogy, it examines the multifaceted exploitation by outsourcing companies, highlighting the gaps between the judicial system, activism, and the lack of research and knowledge.

As an introduction to this metaphorical trilogy, I pose questions that mirror the working conditions, inviting readers to collectively reflect on the subject matter.

In the first chapter, I guide readers through the content moderation experience, using a metaphorical reference to the dystopian setting of the novel „A Clockwork Orange“ by Anthony Burgess, which was adapted into a film produced and directed by Stanley Kubrick.

The second chapter highlights how the misuse of data protection laws by outsourcing companies in Germany enables them to threaten workers‘ freedom. Drawing on Hannah Arendt’s concept of action as essential for preserving human freedom, I argue that NDAs violate human freedom and threaten the very essence of humanity. This situation is reminiscent of the execution apparatus Kafka creates in his short story „In the Penal Colony.“ In this context, any attempt at disclosure, as an action to preserve human freedom, is punished. Therefore, NDAs function as the execution apparatus of freedom.

The third chapter is a visual exploration reflecting on the impact of working under surveillance on Sara’s artwork. Sara, a ceramic artist from Iran, began her job as a content moderator in 2018. In this visual narration, Sara and I examine her ceramic works produced in 2021 as a means of disclosing and preserving her freedom.

A young woman, in a simple tank top, appears on the screen. Her long black hair falls loosely on her shoulders. She stands against a white backdrop and the slim, vertical frame suggests that she films herself from her camera phone. 'Hi guys, welcome back to my channel. Today we're gonna be doing a really requested look: a sun-kissed, no make-up make-up look.' These opening lines, from Turkish

Despite opaque automated systems and few formal channels for participation, YouTubers navigate algorithmic governance on the platform through a strategy we call user-generated accountability: the generation of publicity via content creation to reveal failures, oversights, or harmful policies. Through an analysis of 250 videos, we identify common strategies, concerns, and targets of accountability. Creators primarily upload vlogs that acknowledge the platform’s positive aspects, even as they express concern with YouTube’s policies, automated enforcement systems, poor communication practices, and discrimination against certain creators or content. In publicizing critiques of platform operations, videos enroll creators and audiences as active stakeholders in platform governance that can coordinate actions to draw the company’s attention to matters of concern. We argue that user-generated accountability practices offer a productive starting point for understanding how platform governance disputes come to be and how systems might be shaped or rebuilt to better serve the needs of competing stakeholders.

Xenophobia is one of the key drivers of marginalisation, discrimination, and conflict, yet many prominent machine learning fairness frameworks fail to comprehensively measure or mitigate the resulting xenophobic harms. Here we aim to bridge this conceptual gap and help facilitate safe and ethical design of artificial intelligence (AI) solutions. We ground our analysis of the impact of xenophobia by first identifying distinct types of xenophobic harms, and then applying this framework across a number of prominent AI application domains, reviewing the potential interplay between AI and xenophobia on social media and recommendation systems, healthcare, immigration, employment, as well as biases in large pre-trained models. These help inform our recommendations towards an inclusive, xenophilic design of future AI systems.

Dieser Beitrag erläutert die sozio-technische Funktionsweise der Kuratie rung von Inhalten in sozialen Medien. Er zeigt, wie die der öffentlichen Debatte durch künstliche Intelligenz strukturiert wird und diskutiert mög liche regulatorische Antworten. An verschiedenen Beispielen der vergange nen Jahre lässt sich zeigen, dass Dynamiken in sozialen Medien sowohl Menschenrechte als auch demokratische Werte gefährden können. Um diesen Entwicklungen zu begegnen, gibt es unterschiedliche regulatori schen Ansätzen auf verschiedenen Ebenen. Der vorliegende Beitrag erör tert aktuelle Gesetzgebungsvorhaben und gibt einen Ausblick auf neue Möglichkeiten, den Fragen der Kuratierung von Inhalten zu begegnen. Die Kuratierung von Inhalten in der Entwicklung des Internets Auf dem Weg zu sozialen Medien In den frühen Tagen des Internets waren Interaktionen unmittelbar und die Kommunikation auf Kenntnisnahme von Inhalten ausgerichtet. 1 Das Aufrufen von Seiten setzte Adressen voraus, die man oft über Verzeichnis se organisierte. Die Kommunikation über E-Mail oder Chats bezog sich ebenfalls auf Personen mit Adressen oder Namen, die man bereits kannte oder auf Listen fand. In dieser Zeit kuratierte jeder Nutzer sein eigenes Internet über Adressbücher und Linklisten. 2 Damals wuchs aber die Zahl der Inhalte und der Nutzer so stark, dass es kaum möglich war, einen Überblick über alle relevanten Informationen und Personen zu behalten. Verschiedene Trends befeuerten diesen Wandel. Zum einen war es die rasante Kommerzialisierung des Internets, die auch zu einer stetigen Suche I. A.

This research addresses the crucial challenge of effectively measuring threats in social media comments targeting voting, public officials, and institutions in the United States. Our understanding of these online threats and their links to real-world risks is limited, making it difficult to assess their seriousness. To overcome these limitations, we propose a comprehensive threat level scale from 0 to 5 and collect a dataset of 1.3 million Telegram responses for developing and rigorously testing these threat levels. Additionally, we explore OpenAI-human annotation to efficiently label this vast dataset. Our innovative two-step transfer learning approach initially employs a pre-existing, pretrained model for labeling, followed by expert validation. Next, we use the AI-annotated samples to develop independent models, and expert annotators verify their predictions. Notably, our findings demonstrate that the GPT-2 model, despite its fewer annotated training set, performs comparably to OpenAI's annotations, showcasing its potential for cost-effective threat detection with more annotated samples. With the long-term objective of establishing continuous threat-level monitoring, we identify the strengths and limitations of our current approach and propose a roadmap for enhancing threat detection.

Responding to frustrations with the enforcement of copyright on YouTube, some creators publish videos that discuss their experiences, challenge claims of infringement, and critique broader structures of content moderation. Platform callouts, or public complaints about the conduct of or on platforms, are one of the primary ways creators challenge the power imbalance between users and corporations. Through an analysis of 135 videos, we provide a rich empirical account of how creators publicly define the problem of copyright enforcement, propose solutions, and attribute responsibility to other creators, the platform, and external actors like media conglomerates. Creators criticise the prevalence of "false" copyright claims that ignore fair use or serve ulterior motives like harassment, censorship, and financial extortion, as well as the challenges of communicating with the platform. Drawing inspiration from organisational theory, we differentiate horizontal and vertical callouts according to the institutional positioning of the speaker and target. Horizontal callouts, or public complaints between peers, offer a mechanism for community self-policing, while vertical callouts, or public complaints directed towards organisations, provide a mechanism for influencing centralised content moderation policies and practices. We conclude with a discussion of the benefits and limitations of callouts as a strategy of creator-driven platform governance. Issue 2 This paper is part of Locating and theorising platform power, a special issue of Internet Policy Review guest-edited by David Nieborg, Thomas Poell, Robyn Caplan and José van Dijck.

Technologies, media, and creativity are strictly intersected in creative society. This study sheds light on how new media and technologies affect creativity and creative communication, through research articles published in the Web of Science (WoS-Clarivate Analytics) database. The primary objective of this research is to provide a systematic literature review of existing scientific publications on new media and communication in the creative society and to provide proposals for future research. A total of 173 articles, published between 2012 and 2024 in scientific journals indexed by the WoS database, were considered in this literature mapping. The results of this research report the following: (1) the structure of this field of study with regard to publications, authors, journals, and countries; (2) an analysis of the structure and content of selected articles; (3) an analysis of the new trends in the field (Artificial Intelligence, Algorithms); (4) a critical discussion of the current publications together with proposals for future research. This study aims to offer an overview of current scientific research on new media and communication in the context of creative society, providing for the first time a systematic literature review of this topic.

O Presente Policy Brief é fruto do programa de extensão "Field
Project" coordenado pela Profª. Yasmin Curzi (FGV Direito Rio) e
supervisionado por Giullia Thomaz (CTS-FGV). O relatório é
resultado do trabalho de etnografia digital elaborado e
produzido pelos alunos da FGV Direito Rio, autores do produto.
Este material, seus resultados e conclusões são de
responsabilidade dos autores e não representam, de qualquer
maneira, a posição institucional da Fundação Getulio Vargas /
FGV Direito Rio.

In the rapidly evolving digital landscape of the Middle East and North Africa (MENA) region, content moderation has emerged as a critical issue with profound societal implications. This essay delves into the multifaceted challenges faced in moderating online content in the MENA region, examining the lack of consistent legal frameworks, cultural and religious sensitivities influencing content policies, and the complexities arising from linguistic diversity. Moreover, it explores the technological obstacles hindering effective content moderation, such as limited access to advanced tools, language barriers for automated systems, and encryption challenges. Lastly, the essay analyzes the far-reaching impacts of these challenges, including the proliferation of misinformation and hate speech, threats to freedom of expression and online activism, and the potential for social and political unrest stemming from unregulated content dissemination.

The conversation around content moderation in Africa became prominent following the Arab Spring revolutions, which began in 2010 in northern African countries such as Egypt, Libya, and Tunisia. Since then, social media platforms have faced criticism for their inadequate content moderation efforts in many cases across Africa. These platforms have been particularly ineffective in protecting users from external interference during election periods in countries like Kenya and Nigeria, and have also struggled to effectively regulate hate speech in Ethiopia and Sudan, which has contributed to inciting violent ethnic conflicts and civil strife. Given the absence of effective content moderation by the platforms, discussions regarding the regulation of information disorder and hate speech mostly center on state-centric strategies.

This post seeks to examine the limitations of prevailing content moderation systems, which are built on international human rights law (IHRL). Of course, these systems have well documented linguistic and context blind spots, especially in Africa and other places where social and cultural systems are predominantly communal. This post advocates for a radical shift to an African approach to content moderation, calling for an alternative reading of platform governance from an African perspective that emphasizes the application of collective rights and duties and a communal ethos in content moderation practices.