Search

The article develops a critical approach to AI in content moderation adopting a decolonial perspective. In particular, the article asks: to what extent does the current AI moderation system of platforms address racist hate speech and discrimination? Based on a critical reading of publicly available materials and publications on AI in content moderation, we argue that racialised people have no significant input in the definitions and decision making processes on racist hate speech and are also exploited as their unpaid labour is used to clean up platforms and to train AI systems. The disregard of the knowledge and experiences of racialised people and the expropriation of their labour with no compensation reproduce rather than eradicate racism. In theoretically making sense of this, we draw influences from Anibal Quijano’s theory of the coloniality of power and the centrality of race, concluding that in its current iteration, AI in content moderation is a technology in the service of coloniality. Finally, the article develops a sketch for a decolonial approach to AI in content moderation, which aims to centre the voices of racialised communities and to reorient content moderation towards repairing, educating and sustaining communities.

In this paper, we introduce a novel sociotechnical moderation system for Reddit called Crossmod. Through formative interviews with 11 active moderators from 10 different subreddits, we learned about the limitations of currently available automated tools, and how a new system could extend their capabilities. Developed out of these interviews, Crossmod makes its decisions based on cross-community learning---an approach that leverages a large corpus of previous moderator decisions via an ensemble of classifiers. Finally, we deployed Crossmod in a controlled environment, simulating real-time conversations from two large subreddits with over 10M subscribers each. To evaluate Crossmod's moderation recommendations, 4 moderators reviewed comments scored by Crossmod that had been drawn randomly from existing threads. Crossmod achieved an overall accuracy of 86% when detecting comments that would be removed by moderators, with high recall (over 87.5%). Additionally, moderators reported that they would have removed 95.3% of the comments flagged by Crossmod; however, 98.3% of these comments were still online at the time of this writing (i.e., not removed by the current moderation system). To the best of our knowledge, Crossmod is the first open source, AI-backed sociotechnical moderation system to be designed using participatory methods.

This article analyses the European Commission’s Guidance on Article 17 of the CDSM Directive, the CJEU's YouTube/Cyando judgement and the Advocate General’s Opinion in C-401/19. Together, these documents outline a legal framework for content moderation for digital platforms that tries respecting the copyright-inherent balances and minimize its negative impacts, without however eliminating all problematic issues relating to fundamental rights and other general principles of EU law. Especially after the Opinion in C-401/19, Art. 17's compatibility with freedom of expression seems extremely fragile as it is entirely conditional on how the provision will be implemented by Member States and what safeguards they will provide against overblocking of legitimate uses. Therefore, it is argued that in order to establish a virtuous content moderation system for digital platforms in the EU, it would have been essential to address more fundamental concerns around Art. 17, and in particular the fact that privately operated algorithmic tools will be deciding what content should be available online instead of independent assessors making such decisions based on copyright law’s equilibrium, and to take into account of the inherent limits and flaws of technology.

The Directive on Copyright in the Digital Single Market might fundamentally change the way works and other subject matter protected by copyright will be used on online platforms. Article 17 of the Directive obliges such platforms, based on information they receive from rightholders, to ensure that copyright infringing uploads made by their users are prevented and/or removed by an internalized procedure over which platform operators exercise control. Important enforcement mechanisms are thereby left to private entities solely deciding what content can (or cannot) be available on their platforms. More problematic, the immense masses of uploads seem to be only manageable by automated AI-based filtering technologies incapable of properly distinguishing between lawful and unlawful uses, which leads to a serious danger of overblocking of perfectly legitimate content and thus to potentially important limitations of fundamental rights such as the right to freedom of expression. At the same time, Article 17 also forbids general monitoring of the content available on platforms and clearly mandates to safeguard user’s rights protected by exceptions and limitations to copyright such as quotations, criticism and review or caricature, parody or pastiche. This obligation of result leads to an unsolvable conflict and in a great difficulty for Member States of implementing the provision in a fundamental rights compatible way. Moreover, the lack of clarity of the provision has led Member States to interpret it in significantly diverging ways, jeopardizing the objective of harmonisation behind the Directive. Therefore, the interpretation guidelines by the European Commission foreseen by Article 17(10) to help Member States in their implementation effort were impatiently awaited, in particular since the provision is facing an action for annulment before the Court of Justice of the European Union for potential violating the right to freedom of expression. The Guidance on Article 17 finally issued by the Commission in June 2021 after several postponements, although it provides for some useful clarifications and certain safeguards for users of protected works, but unfortunately does not depart from a system based on monitoring and automated filtering and thus is likely to fail protecting fundamental rights in an appropriate manner. This paper analyses the main additions and proposed interpretation tools that the Guidance brings to platform’s content moderation as mandated by Art. 17 CDSM. It argues that in order to establish a virtuous content moderation system and to help Member States to implementing Article 17 in a balanced way, it would have been essential to address the more fundamental concerns when it comes to Article 17, and in particular the fact that privately operated algorithmic tools and not independent assessors based on copyright law’s equilibrium are deciding what content should be available online and to acknowledge the inherent limits and flaws of technology. We conclude that in the absence of a truly independent arbiter between the interests of users, platforms and rightholders, Article 17 of the Directive is likely not to comply with European fundamental rights and the basic principles of EU law.

The study of fairness in machine learning applications has seen significant academic inquiry, research and publication in recent years. Concurrently, technology companies have begun to instantiate nascent program in AI ethics and product ethics more broadly. As a result of these efforts, AI ethics practitioners have piloted new processes to evaluate and ensure fairness in their machine learning applications. In this session, six industry practitioners, hailing from LinkedIn, Yoti, Microsoft, Pymetrics, Facebook, and Salesforce share insights from the work they have undertaken in the area of fairness, what has worked and what has not, lessons learned and best practices instituted as a result. • Krishnaram Kenthapadi presents LinkedIn's fairness-aware reranking for talent search. • Julie Dawson shares how Yoti applies ML fairness research to age estimation in their digital identity platform. • Hanna Wallach contributes how Microsoft is applying fairness principles in practice. • Lewis Baker presents Pymetric's fairness mechanisms in their hiring algorithm. • Isabel Kloumann presents Facebook's fairness assessment framework through a case study of fairness in a content moderation system. • Sarah Aerni contributes how Salesforce is building fairness features into the Einstein AI platform. Building on those insights, we discuss insights and brainstorm modalities through which to build upon the practitioners' work. Opportunities for further research or collaboration are identified, with the goal of developing a shared understanding of experiences and needs of AI ethics practitioners. Ultimately, the aim is to develop a playbook for more ethical and fair AI product development and deployment.

Part 6: ICT and Education; International audience; Despite the growing importance of digitization in all aspects of teaching and learning, digital moderation (eModeration) has received little attention in research or practice. Considering the secondary school environment, no evidence-based requirements could be found for the development of a digital moderation system. This finding provides the rationale for this investigation into the requirements for an efficient eModeration system for IT and CAT assessments at grade 12 level in South Africa (SA). This study draws on eModeration literature as well as technology adoption literature to identify a set of pre-adoption eModeration requirements against the background of the challenges and benefits of eModeration. The research design involved a single exploratory case study with IT and CAT moderators as units of analysis. Qualitative data analysis using Atlas.ti V8 was conducted on the 61 responses received from an online survey and 4 responses received from a focus group interview. The findings confirmed the absence of customized eModeration systems. The main contribution is the eModeration requirements identified from literature and then refined by triangulation with the empirical findings of the survey. The research has practical value in guiding the design of eModeration systems for the school environment.

The past decade has been characterized by a strong increase in the use of social media and a continuous growth of public online discussion. With the failure of purely manual moderation, platform operators started searching for semi-automated solutions, where the application of Natural Language Processing (NLP) and Machine Learning (ML) techniques is promising. However, this requires huge financial investments for algorithmic implementations, data collection, and model training, which only big players can afford. To support smaller or medium-sized media enterprises (SME), we developed an integrated comment moderation system as an IT platform. This platform acts as a service provider and offers Analytics as a Service (AaaS) to SMEs. Operating such a platform, however, requires a robust technology stack, integrated workflows and well-defined interfaces between all parties. In this paper, we develop and discuss a suitable IT architecture and present a prototypical implementation.

User-generated online comments and posts increasingly contain abusive content that needs moderation from an ethical but also legislative perspective. The amount of comments and the need for moderation in our digital world often overpower the capacity of manual moderation. To remedy this, platforms often adopt semi-automated moderation systems. However, because such systems are typically black boxes, user trust in and acceptance of the system is not easily achieved, as black box systems can be perceived as nontransparent and moderating user comments is easily associated with censorship. Therefore, we investigate the relationship of system transparency through explanations, user trust and system acceptance with an online experiment. Our results show that the transparency of an automatic online comment moderation system is a prerequisite for user trust in the system. However, the objective transparency of the moderation system does not influence the user's acceptance.

RegulationRoom was an online discussion platform designed to help stakeholders historically missing from the conventional rulemaking processes, such as ordinary citizens and small businesses, engage in thoughtful and informed discussion about a proposed rule. To achieve the goal, RegulationRoom adopted a neutrally engaged and expertise-based facilatative moderation system. The RegulationRoom modertor helped commenters better understand the topic, engage in the discussion more effectively, consider opposing views, and substantiate their own comments. In this study, employing social network analysis (SNA), I analyzed one discussion hosted by RegulationRoom on a debt collection policy. I used crosscutting interaction between different interest groups as a proxy indicator to measure deliberativeness. The results showed that the RegulationRoom moderator played a pivotal role in facilitating active discussion, encouraging both within-group and between-group (crosscutting) interactions. Further research needs to be done to illuminate what moderation strategies are most effective in encouraging crosscutting interaction, how the quality of comments (or participation) is improved, and what roles the moderator plays in the process. RegulationRoom also showed that the diffusion of social innovation is not so much a matter of technological progress alone as of social, cultural, and organizational change, both inside and outside the rulemaking context.

Social Networking Services (SNS) depend on user-generated content (UGC). A fraction of UGC is considered spam, such as adult, scam and abusive content. In order to maintain service reliability and avoid criminal activity, content moderation is employed to eliminate spam from SNS. Content moderation consists of manual content-monitoring operations and/or automatic spam-filtering. Detecting a small portion of spam among a large amount of UGC mostly relies on manual operation, thus it requires a large number of human operators and sometimes suffers from human error. In contrast, automatic spam-filtering can be processed with smaller cost, however it is difficult to follow spams' continuously changing trend, and it may declines service experience due to false positives. This presentation introduces an integrated content moderation platform called "Orion'', which aims to minimize manual process and maximize detection of spam in UGC data. Orion preserves post history by users and services, which enables calculating the risk level of each user and decide whether monitoring is required. Also, Orion has a scalable API that can perform number of machine-learning based filtering processes, such as DNN (Deep Neural Network) and SVM for text and images that are posted in many SNS systems. We show that Orion improves efficiency of content moderation compared to a fully manual operation.

Participation in discussions about the public interest can be enhanced by technology, but can also create an environment in which participants are overwhelmed by the quantity, quality, and diversity of information and arguments. Political participation is at a greater disadvantage than non-political activities in that participants from different parties already start out with established differences, which requires them to reach some form of common ground before progress can be made. Those seeking authentic deliberation are discouraged to participate when confronted with uncivil and inflammatory rhetoric. These issues are often exacerbated in online discussions, where lack of identity cues and low barriers to entry can lead to heightened incivility between participants, often labeled as “flaming” and “trolling”. This paper explores the extent to which moderator systems, tools online discussion forums use to manage contributions, can reduce information overload and encourage civil conversations in virtual discussion spaces. Using the popular website Slashdot as an example of sound moderation in a public discourse setting, we found that users move toward consensus about which and how comments deserve to be moderated. Using these findings, we explore how transferable these systems are for participation in public matters specifically to the unique attributes of political discussion. Slashdot's political forum provides a comparison group that allowed us to find quantitative and qualitative differences in political posting, comments, and moderation. Our results show that large scale, civil participation is possible with a distributed moderation system that enables regularly lively debates to be conducted positively because the system provides tools for people to enforce norms of civility.

SUMMARY One of the thorny issues in social work practicum training is how to maintain fairness in assessment. To address this issue, a grade moderation system was set up. Digital practicum portfolios and on-line assessment were used through the Web CT platform. Two amendments were made: (1) password control to protect access rights and privacy; (2) assessment data transfer through Common Gateway Interface (CGI) to conduct automatic descriptive statistical analysis for monitoring possible deviations from the grading standard. A feedback system was built to enhance the quality of teaching using students' survey data. To promote the quality of learning in this individualized teaching mode, the Social Work Practice Teaching, Learning, and Research site was constructed.

Southeast Asia is a standout amongst the most presented districts to unnatural weather change dangers even they are not principle worldwide carbon dioxide (CO2) maker, its discharge will get to be significant if there is no move made. CO2 wellsprings of Southeast Asia are mainly by fossil fuel through era of power and warmth generation, and also transportation part. The endeavors taken by these nations can be ordered into administrative and local level. This paper review the potential for carbon catch and capacity (CCS) as a part of the environmental change moderation system for the Malaysian power area utilizing an innovation appraisal structure. The country's recorded pattern of high dependence on fossil fuel for its power segment makes it a prime possibility for CCS reception. This issue leads to gradual increment of CO2 emission. It is evident from this evaluation that CCS can possibly assume a vital part in Malaysia's environmental change moderation methodology gave that key criteria are fulfilled. With the reason to pick up considerations from all gatherings into the earnestness of an Earth-wide temperature boost issue in Southeast Asia, assume that more efficient measures can be taken to effectively accomplish CO2 diminishment target.

We consider the problem of building online machine-learned models for detecting auction frauds in e-commence web sites. Since the emergence of the world wide web, online shopping and online auction have gained more and more popularity. While people are enjoying the benefits from online trading, criminals are also taking advantages to conduct fraudulent activities against honest parties to obtain illegal profit. Hence proactive fraud-detection moderation systems are commonly applied in practice to detect and prevent such illegal and fraud activities. Machine-learned models, especially those that are learned online, are able to catch frauds more efficiently and quickly than human-tuned rule-based systems. In this paper, we propose an online probit model framework which takes online feature selection, coefficient bounds from human knowledge and multiple instance learning into account simultaneously. By empirical experiments on a real-world online auction fraud detection data we show that this model can potentially detect more frauds and significantly reduce customer complaints compared to several baseline models and the human-tuned rule-based system.

Online communities provide a social sphere for people to share information and knowledge. While information sharing is becoming a ubiquitous online phenomenon, how to ensure information quality or induce quality content remains a challenge because of the anonymity of commentators. This paper introduces moderation into reputation systems. We show that moderation directly affects strategic commentators' incentive to generate useful information, and moderation is generally desirable to improve information quality. We find that when being moderated with different probabilities based on their reputations, commentators might display a pattern of reputation oscillation, in which they generate useful content to build up high reputation and then exploit their reputation. As a result, the expected performance from high-reputation commentators can be inferior to that from low-reputation commentators (reverse reputation). We then investigate the optimal moderation resource allocation and conclude that the seemingly abnormal reverse reputation could arise as an optimal result. Our study underscores the importance of moderation and highlights that the frequency of moderation should be properly chosen for better performance of online communities.

Search engines have evolved along with the internet, and now control massive amounts of data that they often manipulate. In addition to these deceptive search practices, search engines are currently facing a slew of trademark lawsuits from owners of protected marks as well as criticism from privacy scholars. As gateways to the modern formulation of the internet, search engines resemble a bottleneck for internet traffic ideally suited for a streamlined regulatory approach to cure the current woes of search.This paper proposes a method of modified self regulation where search engines are required to create a feedback system for their search results not unlike eBay's reputation scores or Slashdot's comment and moderation system, effectively giving the community of users the ability to rate the relevance of search results, rather than a black-box algorithm. This, of course, would not be done without legislation. The simplest way to accomplish this feedback regime would be to require search engines to implement it or else lose their immunities provided under §230 of the Communications Decency Act and §512 of the Digital Millennium Copyright Act, which currently immunize search engines for torts that are committed by sites to which they link.Once a feedback regime is implemented, search engines would no longer be able to misrank search results, return results for counterfeit goods, or link to a false story damaging a person's reputation without being noticed, talked about, and rated by the watchful eye of the online community.

This paper presents an integration architecture and a technical framework for moderation of e-discussions, a specific type of scenario for collaborative learning. The approach for monitoring and moderating discussions is supported at a generic level and defines interfaces to integrate diverse discussion environments smoothly into the generic framework. The usage of the developed monitoring and moderation system has been tested practically with two different stand-alone discussion environments and first empirical results of the moderation approach have been collected. We conclude with future work on the framework system and additional discussion tools to be integrated into the architecture.