Media Governance After AI

Parliamentary testimony on the Implications of Artificial Intelligence Technologies for the Canadian Labor Force

November 9, 2023 by Fenwick McKelvey

Below are remarks made by Dr. Fenwick McKelvey to the Standing Committee on Human Resources, Skills and Social Development and the Status of Persons with Disabilities

Introduction

I am an Associate Professor in Information and Communication Technology Policy at Concordia University. My research addresses the intersection of algorithms and AI in relation to technology policy.

I submit these comments in my professional capacity representing my views alone.

Background

I am speaking from the unceded Indigenous lands of Tiohtià:ke/Montréal. The Kanien’kehá:ka Nation is recognized as the custodians of the lands and waters on which we gather today.

I want to begin my connecting this study to the broader legislative agenda then provide some specific comments about the connection between foundational models trained off public data or other larger data sets, and the growing concentration in the AI industry.

Canada is presently undergoing major changes to its federal data and privacy law, C-27, that grants greater exemptions for data collection. These exemptions enable greater use of machine learning and other data-dependent classes of AI technologies, putting tremendous pressure on a late amendment, the Artificial Intelligence and Data Act, to mitigate high-risk applications and plausible harms. Labour, automation, workers’ privacy and data rights should be important considerations for this bill as seen in the US AI Executive Order. I would encourage this committee to study the effects of C-27 on workplace privacy and consequences of a more permissive data environment.

As for the relationship between labour and artificial intelligence, I wish to make three major observations based on my review of the literature and a few recommendations.

AI will affect the labour force and these effects will be unevenly distributed.
AI’s effects are not simply about automation, but about the quality of the work.
The current global arrangement of AI is concentrating power in a few technology firms.

AI will affect the labourforce and these effects seem to be unevenly distributed.

I grew up in Saint John, New Brunswick under the shadow of global supply chains and a changing workforce. My friends all worked in call centers. Now these same jobs will be automated by chatbots or at least assisted through generative AI. My own research has shown that a driving theme in discussing AI in the telecommunication services focuses on automating customer contact.

I begin with call centers because as we know from the work of Dr. Edna Brophy that “workforce…is female, precarious, and mobile”. The examples serve as an important reminder that AI’s effects may further marginalize workers targeted for automation.

AI’s effects seem to already be affecting precarious outsourced workers according to reporting from the Rest of the World. Understanding the intersectional effects of AI is critical to understanding its impact on the labour force. We are only beginning to know how Canada will fit into the global shifts, how Canada might export more precarious jobs abroad as well as find job growth in new sectors or regions.

Finally, workers are increasingly finding themselves subjected to algorithmic management. Combined with a growing turn toward workplace surveillance as being studied by Dr. Adam Molnar, there is an urgent need to understand and protect workers from invasive data gathering that might reduce their workplace autonomy or even train their less-skilled or automated replacements. According to the OECD, workers subjected to algorithmic management have a larger reported feeling of a loss of autonomy.

Quality of work

All the promises of AI hinge on being able to do work more efficiently. But who benefits from this efficiency? OECD studies have found that “AI may also lead to a higher pace and intensity of work.” The impacts seem obvious and well established by past studies of technologies like the Blackberry that shifted workplace expectations and encouraged an always-on expectation of workers.

Other research suggests that AI has the biggest benefits for new employees. The presumed benefit is that this enables workers to make a contribution more quickly, but the risk is that AI contributes to a devaluing / deskilling of workers.

These emphasize the need to consider AI effects not just on jobs, but the quality of work itself.

Concentration of power

The introduction of generative AI marks has arrived embedded into key productivity suites, named Microsoft Office, Google Docs, and Adobe Creative Cloud. My final comment, which is less about AI than about its particular configuration now, is that we are seeing a growing reliance on a few technology platforms that have become critical infrastructure for workplace productivity. AI might lock in these firms’ market power as their access to data and cloud computing power makes it difficult to compete. Past examples demonstrate that communication technology favours monopolies without open standards and efforts to decentralized power.

I am happy to discuss remedies and solutions in the question and answer period, but I encourage the committee to:

Investigate better protection of workers and workers’ rights including greater data protection and safeguards and enforcement of workplace surveillance especially to ensure works can’t train themselves out of a job;
Consider arbitration and greater support in bargaining power especially for contracts between independent contractors and large technology firms arrangements;
Ensure efficiency boosts are fairly distributed, such as considering a 4-day work week, raising minimum wages, and A right to disconnect from work.

Thank you for your time and this opportunity.

Mitigating Systemic Online Harms: The Challenges Posed by Internet Personalization

July 11, 2023 by Fenwick McKelvey

Greg Elmer, Fenwick McKelvey, Ganaele Langlois, and Seungwoo Baek

Introduction

This short research brief and accompanying literature review serves as an ongoing effort to identify not only trends in research on online harms, but also areas in need of development, funding, public attention, and debate. The broad umbrella of “systemic online harms” highlights a common set of factors that contribute to issues as broad as internet-based conspiracies, disinformation, troll farms, and fake news sites as well as the targeting, harassment, and abuse of marginalized and racialized individuals and communities. While politics and some extremist ideologies have clearly contributed to such online harms, this literature review is more narrowly focused on socio-technical factors: how the internet and other forms of networked computing have amplified such harms.

Some might think that our focus on “the internet” may limit the scope of mediated social harms, but under this umbrella term we include other essential technologies of networked computing: the web, social media platforms, search engines, user interfaces, and, indeed, personal computers, tablets, and handheld devices such as smartphones. Content that used to be exclusively accessed by television cables, newspapers, or over-the-air broadcasts all now converge — almost universally “streamed” or otherwise posted — on these internet properties and technologies. What’s more, we contend that online harms are often cultivated by — or mediated through — internet-based forms of news and information. Given that 79% of Canadians get their news/info from the internet, a rigorous and vibrant research agenda is urgently needed to offer potential solutions to online harms.

The shift to “the internet” further involves changing media habits and an overall convergence of entertainment, educational, professional, interpersonal, and news content on the same device, on the same app. This apparent collapse in generic distinctions among content has led to competition for scarce attention between news and entertainment. These trends have been ongoing since the advent of cable and the early internet, but the ubiquity of smartphones has further altered the context of media habits from passive watching to distracted scrolling. A decade after the introduction of the Blackberry changed expectations of worker availability and the iPhone made computing portable, we know media habits have changed, but how these changes matter to public life and to the prominence of online harms remains unclear.

This short research brief focuses on one of the main obstacles to tackling online harms, media personalization. Where we once referred to the social power of the mass media, stemming from common images, narratives, and stories told through film and broadcast television, mass personalization refers to the near universal customization and filtering of news and information for individual users. The term “personalization” might seem positive at first glance in that it evokes adaptive and intelligent technologies able to sift through massive amounts of information to find what a person actually needs, wants, or desires. However, the form of personalization that we are confronted with online today involves the constant surveillance and profiling of users to find ways to effectively turn and cultivate users to react to, or against, something.

Personalization should therefore be understood as another form of manipulation, much like mass propaganda was a dominant form of manipulation in the age of mass media. Again, unlike the previous broadcasting model of communication, the internet has long been characterized as decidedly decentred or at least distributed among many nodes, offering individuals much greater access to — and participation with — start-up media organizations, online diaries (blogs), or other forms of public communication.

The rapid proliferation of many sources of information provoked an old concern about “information overload” that iterations of internet services sought to address. Early solutions to this problem of content categorization mimicked preexisting systems found in libraries, like thematic categories (adopted by Yahoo) or the use of keywords to search archives (e.g., Google’s search engine). Such systems addressed two key problems introduced by decentralized, networked media, namely the need to find relevant information in a convenient manner. A second generation turned to public participation — crowdsourcing, gate-watching, and collaborative filtering — that popularized early excitement about a participatory internet and the possibility of social media. With the consolidation of social media by a few firms, these participatory elements became automated and dependent on surveillance and profiling to make content recommendations, a move that aligned closely with revenue models dependent on targeted advertising.

User interfaces have become increasingly governed by a back-end software substructure. While many have referred to this computational or “algorithmic” level as opaque, secret, or “black boxed,” the logic of mass personalization has long been exposed by technology journalists and researchers for many years, including by some information scientists responsible for their design. Starting with web cookies, code that web pages (often e-tailers) store on visitors’ personal computers to better customize their future transactions, internet personalization started to take hold as the overriding logic of algorithms that governed the visibility of content on social media platforms such as Facebook, Twitter, and Instagram.

Two other factors are commonly cited as having contributed to the massification of internet personalization, the widespread use of user registration systems (accounts for e-tailers and online service/information providers) and internet interfaces that reinforced personalized recommendations and content from online “friends.” The current internet landscape can in short be characterized as a back-end, algorithmically driven system that gives preference to personalized content, visualized by interfaces that make such personally “relevant” content convenient to find, follow, and engage with.

This brings us to the social harms produced by such personalized systems and the need for research to address potential solutions and mitigating factors. As the internet underwent a “user-generated content” revolution, greatly expanding the opportunities to contribute content to the internet’s many platforms and properties, interfaces have struggled to represent the accompanying explosion in content. Threaded conversations that gain added prominence from users’ votes have privileged hot takes, outrageous opinions, or worse. Personal attacks have consequently become commonplace alongside the personalization of internet content and social communication. Much of this research has focused on the internet as a distinct space or economy of attention, where harms are produced by those seeking to outshout or otherwise capture the attention of others online.

Personalization has been actively constructed in marketing as a common sense solution. Personalization sounds intuitive but is anything but. Moreover, personalization simplifies automated decision-making itself contingent on user surveillance, big data, and machine learning. Personalization is a layer of thin ice covering a deeper pool of questions relating to how platforms calculate user engagement, what counts as optimal behaviour, and what aspects of someone’s multifaceted identity are brought out by algorithmic recommendation.

The more prominent argument in the literature however questions the social effects of personalization and the resulting system of “homophily,” an information system that recommends users “more of the same.” In this context, algorithmic personalization is said to produce a polarized population in clusters of hardened worldviews and keeps individuals in ideological “bubbles,” distant from unfamiliar, uncomfortable, or potentially disagreeable content.

Much of the current literature focuses on specifying the impact of this homophilic form of algorithmic communication, on qualifying it, or, in a small though growing number of articles and books, on rejecting its central premise, arguing that users are free to choose content as they see fit. While individual agency is undoubtedly an important component of everyday consumption of internet-based information and news, the prevalence of personalized interfaces and content cannot be overcome or adequately addressed by individualized solutions. That users have agency and may be exposed to different sources of information does not diminish the need to question personalization as the de facto solution to content regulation online.

The individual power to find “different” content must still swim against the current of personalization, and choosing to do so makes internet browsing and search decidedly more work and less convenient. A thread of research that recognizes some users’ need and desire for different and unfamiliar content can be found in studies of content discoverability and search, an approach that addresses forms of personalization and commercial optimization that monopolize search engine results and, thus, the visibility of content and services.

Ultimately though, the core logic of a default and algorithmically governed process of media personalization needs to be addressed head on, with competing models of news and information urgently required to complement educational strategies, calls for privacy, and research on other political and social factors that mitigate harmful homophilic media effects. Such research requires a rethinking of the visibility and discoverability of heterogeneous information sources and opinions, while maintaining a degree of usability, relevance, and convenience. In short, we believe that studies of mitigating media personalization will never fully redress or prevent the harms of media personalization without recognizing and further studying user practices that prevail beyond their own customized media platforms and interfaces.

Personalization Research Project Bibliography

We see an emerging gap between a refutation of popular accounts of algorithmic harms and a growing call to research other, less-studied algorithmic harms. Filter bubbles and polarization are two key debates in contemporary communication studies. These debates invoked high-choice media environments, social media, and algorithmic filtering as factors in a politicization of world views and greater animosity to those with different opinions. In other words, filter bubbles and polarization introduce algorithmic harms, reflecting concerns that the algorithmization of media environments has detrimental effects on society and politics. We see limitations in these specific algorithmic harms and highlight other harms based on our broader literature review.

While we acknowledge there is no consensus about the significance of radicalization and polarization nor whether these are the right terms to discuss risks to democracy, we find that evidence in Europe finds little to support claims that algorithms are significant factors. Algorithms are commonly perceived to encourage radicalization and polarization through filter bubbles — or homogenous bubbles of information organized by algorithms. A major literature review conducted by the Reuters Institute concludes that “echo chambers are much less widespread than is commonly assumed” and “finds no support for the filter bubble hypothesis and offers a very mixed picture on polarisation and the role of news and media use in contributing to polarisation” [p. 5]. Rather, the effect is largely the opposite, with algorithms just as likely to result in serendipitous and incidental exposure. The review continues, “studies in the UK and several other countries, have found that algorithmic selection generally leads to slightly more diverse news use — the opposite of what the filter bubble hypothesis posits — but that self-selection, primarily among a small minority of highly partisan individuals, can lead people to opt in to echo chambers, even as the vast majority do not, and document that limited news use remains far more prevalent than echo chambers are” [p. 20]. This latter comment suggests that deeper research is needed to understand the effects of algorithmic recommendations on publics demanding more partisan or homogeneous information.

We caution that this study indicates a lack of research in Canada. Out of the 114 articles cited in the Reuters review, only 8 mention Canada. One of which explains that, “it was not possible for us to analyze Canada using our approach because it is essentially home to two media systems, one based on French-language media and one based on English-language media” (Fletcher, Cornea, & Nielsen, 2020). The low representation of Canada in international studies should point to a major caveat in the interpretation of results. When compared, Canada is better understood in relation to European counterparts than the United States. A cross-national survey of resilience to disinformation found that Canada was more like its European counterparts “with respect to welfare expenditure, support for public broadcasting, and regulations of media ownership, advertising, and electoral coverage,” and that “Western European democracies and Canada . . . are likely to demonstrate high resilience to online disinformation: they are marked by low levels of polarization and populist communication, high levels of media trust and shared news consumption, and a strong PSB” (Humprcht, Esser, & Van Aelst, 2020, pp. 13-15).

We then caution that radicalization and polarization are likely low algorithmic harms in Canada, because Canada is seen to be a more moderate media system with known countermeasures (such as public broadcasting) and there appears to be a less pronounced demand for highly partisan or extreme news. The Digital Democracy Project, the largest project of its kind to date, found that: “news sources catering to specific partisan audiences play a very small role in Canada”; “Canadians generally tend to be ideologically moderate”; and “Canadians are substantially less likely to pick partisan news sources and stories than one might expect.” Other studies find that Canada has seen slight increases in affective polarization but much less than in the United States.

While these studies suggest that Canada is both a distinct case — with a more complete linguistically diverse media system — and also a resilient media system, mitigating concern about algorithmic filter bubbles as a specific harm. However, we wish to stress that these studies tend to focus on political information and less on media habits, evaluating the effects of algorithms on access to information. In addition, there is consensus that some users might be more prone to these algorithmic harms. Research consistently identifies subsets of Republican and right-wing voters as more inclined to share misinformation and seek out echo chambers. The United States, however, is an anomaly in comparative media studies — Canadians may not share the same prevalence as Americans.

We observe a strong emphasis on journalists’ theories of media consumption — as information seeking — in prompting concerns about polarization and filter bubbles. The so-called realist approach to democracy, in contrast, considers voters to be less apt to seek out information, habitual, and, ergo, less prone to being influenced by the media. Thus, voter apathy and political disinterest in Canada might lead to reduced risk of polarization and filter bubbles — citizens simply cannot be bothered. If this is true, we then see a different set of potential algorithmic harms that warrant greater discussion.

These harms include:

a more nuanced version of filter bubbles, one less concerned with homogeneity than the consequences of personalization to political culture;
a focused concern on specific uses cases or demands (e.g., searching for white nationalist content on YouTube) that algorithmic systems cannot or do not distinguish from other requests;
homophily and algorithmic attempts to draw inferences between users and content, effectively or not; and,
transparency and opacity of algorithmic regulation and the consequences of its unknowability, including the encouragement of conspiracy theories as popular explanations (e.g., “my phone is listening to me”)

These algorithmic harms are less well researched, even less so in Canada.

Bibliography

Boxell, L., Gentzkow, M., & Shapiro, J. (2020). Cross-country trends in affective polarization. Cambridge, MA: National Bureau of Economic Research. https://doi.org/10.3386/w26669

Fletcher, R., Cornia, A., & Nielsen, R. K. (2020). How Polarized Are Online and Offline News Audiences? A Comparative Analysis of Twelve Countries. The International Journal of Press/Politics, 25(2), 169–195. https://doi.org/10.1177/1940161219892768

Humprecht, E., Esser, F., & Van Aelst, P. (2020). Resilience to Online Disinformation: A Framework for Cross-National Comparative Research. The International Journal of Press/Politics, 1940161219900126. https://doi.org/10.1177/1940161219900126

Algorithmic Harms

Adomavicius, G., & Tuzhilin, A. (1999). User profiling in personalization applications through rule discovery and validation. Proceedings of the Fifth ACM SIGKDD International Conference on Knowledge Discovery and Data Mining – KDD ’99, 377–381. https://doi.org/10.1145/312129.312287

Beer, D. (2009). Power through the algorithm? Participatory web cultures and the technological unconscious. New Media & Society, 11(6), 985–1002. https://doi.org/10.1177/1461444809336551

Bucher, T. (2018). If…Then (Vol. 1). Oxford University Press. https://doi.org/10.1093/oso/9780190493028.001.0001

Cinelli, M., De Francisci Morales, G., Galeazzi, A., Quattrociocchi, W., & Starnini, M. (2021). The echo chamber effect on social media. Proceedings of the National Academy of Sciences, 118(9), e2023301118. https://doi.org/10.1073/pnas.2023301118

de Vries, K. (2010). Identity, profiling algorithms and a world of ambient intelligence. Ethics and Information Technology, 12(1), 71–85. https://doi.org/10.1007/s10676-009-9215-9

Garrett, R. K. (2013). Selective Exposure: New Methods and New Directions. Communication Methods and Measures, 7(3–4), 247–256. https://doi.org/10.1080/19312458.2013.835796

Gillespie, T. (2014). The Relevance of Algorithms. In T. Gillespie, P. J. Boczkowski, & K. A. Foot (Eds.), Media Technologies: Essays on Communication, Materiality, and Society (pp. 167–194). The MIT Press. https://doi.org/10.7551/mitpress/9780262525374.003.0009

Just, N., & Latzer, M. (2017). Governance by algorithms: Reality construction by algorithmic selection on the Internet. Media, Culture & Society, 39(2), 238–258. https://doi.org/10.1177/0163443716643157

Munger, K., & Phillips, J. (2022). Right-Wing YouTube: A Supply and Demand Perspective. The International Journal of Press/Politics, 27(1), 186–219. https://doi.org/10.1177/1940161220964767

Nielsen, R. K., & Fletcher, R. (2020). Democratic Creative Destruction? The Effect of a Changing Media Landscape on Democracy. In N. Persily & J. A. Tucker (Eds.), Social Media and Democracy: The State of the Field, Prospects for Reform (1st ed., pp. 139–162). Cambridge University Press. https://doi.org/10.1017/9781108890960

Prior, M. (2013). Media and Political Polarization. Annual Review of Political Science, 16(1), 101–127. https://doi.org/10.1146/annurev-polisci-100711-135242

Thorson, K. (2020). Attracting the news: Algorithms, platforms, and reframing incidental exposure. Journalism, 21(8), 1067–1082. https://doi.org/10.1177/1464884920915352

Tucker, J., Guess, A., Barbera, P., Vaccari, C., Siegel, A., Sanovich, S., Stukal, D., & Nyhan, B. (2018). Social Media, Political Polarization, and Political Disinformation: A Review of the Scientific Literature. SSRN Electronic Journal. https://doi.org/10.2139/ssrn.3144139

van Dijck, J. (2013). The Culture of Connectivity: A Critical History of Social Media. Oxford University Press. https://doi.org/10.1093/acprof:oso/9780199970773.001.0001

Yeung, K. (2017). ‘Hypernudge’: Big Data as a mode of regulation by design. Information, Communication & Society, 20(1), 118–136. https://doi.org/10.1080/1369118X.2016.1186713

Yeung, K. (2018). Algorithmic regulation: A critical interrogation: Algorithmic Regulation. Regulation & Governance, 12(4), 505–523. https://doi.org/10.1111/rego.12158

Filter Bubbles (Homophily)

Andersson, M. (2021). The climate of climate change: Impoliteness as a hallmark of homophily in YouTube comment threads on Greta Thunberg’s environmental activism. Journal of Pragmatics, 178, 93–107. https://doi.org/10.1016/j.pragma.2021.03.003

Aral, S., Muchnik, L., & Sundararajan, A. (2013). Engineering social contagions: Optimal network seeding in the presence of homophily. Network Science, 1(2), 125–153. https://doi.org/10.1017/nws.2013.6

Caetano, J. A., Lima, H. S., Santos, M. F., & Marques-Neto, H. T. (2018). Using sentiment analysis to define twitter political users’ classes and their homophily during the 2016 American presidential election. Journal of Internet Services and Applications, 9(1), 18. https://doi.org/10.1186/s13174-018-0089-0

Cai, W., Jia, J., & Han, W. (2018). Inferring Emotions from Image Social Networks Using Group-Based Factor Graph Model. 2018 IEEE International Conference on Multimedia and Expo (ICME), 1–6. https://doi.org/10.1109/ICME.2018.8486450

Chun, W. H. K. (2018). Queerying Homophily. In Pattern Discrimination (p. 124). Meson Press.

Fabbri, F., Bonchi, F., Boratto, L., & Castillo, C. (2020). The Effect of Homophily on Disparate Visibility of Minorities in People Recommender Systems. Proceedings of the International AAAI Conference on Web and Social Media, 14(1), 165–175.

Figeac, J., & Favre, G. (2021). How behavioral homophily on social media influences the perception of tie-strengthening within young adults’ personal networks. New Media & Society, 146144482110206. https://doi.org/10.1177/14614448211020691

Halberstam, Y., & Knight, B. (2016). Homophily, group size, and the diffusion of political information in social networks: Evidence from Twitter. Journal of Public Economics, 143, 73–88. https://doi.org/10.1016/j.jpubeco.2016.08.011

Hekim, H. (2021). Ideological homophily or political interest: Factors affecting Twitter friendship network between politicians. Journal of Information Technology & Politics, 18(4), 371–386. https://doi.org/10.1080/19331681.2021.1881937

Ingram, P., & Choi, Y. (2017). From Affect to Instrumentality: The Dynamics of Values Homophily in Professional Networks. Academy of Management Proceedings, 2017(1), 15151. https://doi.org/10.5465/AMBPP.2017.15151abstract

Kevins, A., & Soroka, S. N. (2018). Growing Apart? Partisan Sorting in Canada, 1992–2015. Canadian Journal of Political Science, 51(1), 103–133. https://doi.org/10.1017/S0008423917000713

Khanam, K. Z., Srivastava, G., & Mago, V. (2020). The Homophily Principle in Social Network Analysis. ArXiv:2008.10383 [Physics]. http://arxiv.org/abs/2008.10383

Ladhari, R., Massa, E., & Skandrani, H. (2020). YouTube vloggers’ popularity and influence: The roles of homophily, emotional attachment, and expertise. Journal of Retailing and Consumer Services, 54, 102027. https://doi.org/10.1016/j.jretconser.2019.102027

Laniado, D., Volkovich, Y., Kappler, K., & Kaltenbrunner, A. (2016). Gender homophily in online dyadic and triadic relationships. EPJ Data Science, 5(1), 19. https://doi.org/10.1140/epjds/s13688-016-0080-6

Nagar, S., Gupta, S., Bahushruth, C. S., Barbhuiya, F. A., & Dey, K. (2021). Empirical Assessment and Characterization of Homophily in Classes of Hate Speeches. Proceedings of the AAAI-21 Workshop on Affective Content Analysis, New York, USA, AAAI.

Nahon, K., & Hemsley, J. (2014). Homophily in the Guise of Cross-Linking: Political Blogs and Content. American Behavioral Scientist, 58(10), 1294–1313. https://doi.org/10.1177/0002764214527090

O’Callaghan, D., Greene, D., Conway, M., Carthy, J., & Cunningham, P. (2015). Down the (White) Rabbit Hole: The Extreme Right and Online Recommender Systems. Social Science Computer Review, 33(4), 459–478. https://doi.org/10.1177/0894439314555329

Seargeant, P., & Tagg, C. (2019). Social media and the future of open debate: A user-oriented approach to Facebook’s filter bubble conundrum. Discourse, Context & Media, 27, 41–48. https://doi.org/10.1016/j.dcm.2018.03.005

Personalization

Atote, B. S., Zahoor, S., Dangra, B., & Bedekar, M. (2016). Personalization in User Profiling: Privacy and Security Issues. 2016 International Conference on Internet of Things and Applications (IOTA), 415–417. https://doi.org/10.1109/IOTA.2016.7562763

Awad & Krishnan. (2006). The Personalization Privacy Paradox: An Empirical Evaluation of Information Transparency and the Willingness to Be Profiled Online for Personalization. MIS Quarterly, 30(1), 13. https://doi.org/10.2307/25148715

Bennett, W. L. (2012). The Personalization of Politics: Political Identity, Social Media, and Changing Patterns of Participation. The ANNALS of the American Academy of Political and Social Science, 644(1), 20–39. https://doi.org/10.1177/0002716212451428

Bennett, W. L., & Segerberg, A. (2011). Digital Media and the Personalization of Collective Action: Social technology and the organization of protests against the global economic crisis. Information, Communication & Society, 14(6), 770–799. https://doi.org/10.1080/1369118X.2011.579141

Bennett, W. L., & Segerberg, A. (2012). The Logic of Connective Action: Digital media and the personalization of contentious politics. Information, Communication & Society, 15(5), 739–768. https://doi.org/10.1080/1369118X.2012.670661

Bozdag, E. (2013). Bias in algorithmic filtering and personalization. Ethics and Information Technology, 15(3), 209–227. https://doi.org/10.1007/s10676-013-9321-6

Cohen, J. N. (2018). Exploring Echo-Systems: How Algorithms Shape Immersive Media Environments. Journal of Media Literacy Education, 10(2), 139–151. https://doi.org/10.23860/JMLE-2018-10-2-8

Farid, M., Elgohary, R., Moawad, I., & Roushdy, M. (2018). User Profiling Approaches, Modeling, and Personalization. SSRN Electronic Journal. https://doi.org/10.2139/ssrn.3389811

Jain, S., Grover, A., Thakur, P. S., & Choudhary, S. K. (2015). Trends, problems and solutions of recommender system. International Conference on Computing, Communication & Automation, 955–958. https://doi.org/10.1109/CCAA.2015.7148534

Kang, H., & Sundar, S. S. (2016). When Self Is the Source: Effects of Media Customization on Message Processing. Media Psychology, 19(4), 561–588. https://doi.org/10.1080/15213269.2015.1121829

Kant, T. (2020). Making it Personal: Algorithmic Personalization, Identity, and Everyday Life (1st ed.). Oxford University Press. https://doi.org/10.1093/oso/9780190905088.001.0001

Karwatzki, S., Dytynko, O., Trenz, M., & Veit, D. (2017). Beyond the Personalization–Privacy Paradox: Privacy Valuation, Transparency Features, and Service Personalization. Journal of Management Information Systems, 34(2), 369–400. https://doi.org/10.1080/07421222.2017.1334467

Li, T., & Unger, T. (2012). Willing to pay for quality personalization? Trade-off between quality and privacy. European Journal of Information Systems, 21(6), 621–642. https://doi.org/10.1057/ejis.2012.13

Lury, C., & Day, S. (2019). Algorithmic Personalization as a Mode of Individuation. Theory, Culture & Society, 36(2), 17–37. https://doi.org/10.1177/0263276418818888

Matamoros-Fernández, A., Gray, J. E., Bartolo, L., Burgess, J., & Suzor, N. (2021). What’s “Up Next”? Investigating Algorithmic Recommendations on YouTube Across Issues and Over Time. Media and Communication, 9(4), 234–249. https://doi.org/10.17645/mac.v9i4.4184

Oeldorf-Hirsch, A., & Sundar, S. S. (2015). Posting, commenting, and tagging: Effects of sharing news stories on Facebook. Computers in Human Behavior, 44, 240–249. https://doi.org/10.1016/j.chb.2014.11.024

Rosenthal, S., Wasenden, O.-C., Gronnevet, G.-A., & Ling, R. (2020). A tripartite model of trust in Facebook: Acceptance of information personalization, privacy concern, and privacy literacy. Media Psychology, 23(6), 840–864. https://doi.org/10.1080/15213269.2019.1648218

Sundar, S. S., & Limperos, A. M. (2013). Uses and Grats 2.0: New Gratifications for New Media. Journal of Broadcasting & Electronic Media, 57(4), 504–525. https://doi.org/10.1080/08838151.2013.845827

Sundar, S. S., & Marathe, S. S. (2010). Personalization versus Customization: The Importance of Agency, Privacy, and Power Usage. Human Communication Research, 36(3), 298–322. https://doi.org/10.1111/j.1468-2958.2010.01377.x

Sunikka, A., & Bragge, J. (2008). What, Who and Where: Insights into Personalization. Proceedings of the 41st Annual Hawaii International Conference on System Sciences (HICSS 2008), 283–283. https://doi.org/10.1109/HICSS.2008.500

Zeng, F., Ye, Q., Li, J., & Yang, Z. (2021). Does self-disclosure matter? A dynamic two-stage perspective for the personalization-privacy paradox. Journal of Business Research, 124, 667–675. https://doi.org/10.1016/j.jbusres.2020.02.006

Echo Chambers

Acemoglu, D., Ozdaglar, A., & Siderius, J. (2021). Misinformation: Strategic Sharing, Homophily, and Endogenous Echo Chambers (No. w28884; p. w28884). National Bureau of Economic Research. https://doi.org/10.3386/w28884

Baccara, M., & Yariv, L. (2008). Similarity and Polarization in Groups. SSRN Electronic Journal. https://doi.org/10.2139/ssrn.1244442

Mukhudwana, R. F. (2020). #Zuma Must Fall This February: Homophily on the Echo-Chambers of Political Leaders’ Twitter Accounts. In M. N. Ndlela & W. Mano (Eds.), Social Media and Elections in Africa, Volume 2: Challenges and Opportunities (pp. 175–202). Springer International Publishing. https://doi.org/10.1007/978-3-030-32682-1_10

Sasahara, K., Chen, W., Peng, H., Ciampaglia, G. L., Flammini, A., & Menczer, F. (2021). Social Influence and Unfollowing Accelerate the Emergence of Echo Chambers. Journal of Computational Social Science, 4(1), 381–402. https://doi.org/10.1007/s42001-020-00084-7

Mitigating (Algorithmic Harms)

Ananny, M., & Crawford, K. (2018). Seeing without knowing: Limitations of the transparency ideal and its application to algorithmic accountability. New Media & Society, 20(3), 973–989. https://doi.org/10.1177/1461444816676645

Flew, T. (2019). Digital communication, the crisis of trust, and the post-global. Communication Research and Practice, 5(1), 4–22. https://doi.org/10.1080/22041451.2019.1561394

Helberger, N., Karppinen, K., & D’Acunto, L. (2018). Exposure diversity as a design principle for recommender systems. Information, Communication & Society, 21(2), 191–207. https://doi.org/10.1080/1369118X.2016.1271900

Helberger, N., Pierson, J., & Poell, T. (2018). Governing online platforms: From contested to cooperative responsibility. The Information Society, 34(1), 1–14. https://doi.org/10.1080/01972243.2017.1391913

Nagulendra, S., & Vassileva, J. (2016). Providing awareness, explanation and control of personalized filtering in a social networking site. Information Systems Frontiers, 18(1), 145–158. https://doi.org/10.1007/s10796-015-9577-y

Natali Helberger. (2011). Diversity by Design. Journal of Information Policy, 1, 441–469. https://doi.org/10.5325/jinfopoli.1.2011.0441

Nissenbaum, H. (2011). A Contextual Approach to Privacy Online. Daedalus, 140(4), 32–48. https://doi.org/10.1162/DAED_a_00113

Reviglio, U., & Agosti, C. (2020). Thinking Outside the Black-Box: The Case for “Algorithmic Sovereignty” in Social Media. Social Media + Society, 6(2), 205630512091561. https://doi.org/10.1177/2056305120915613

Ziewitz, M. (2016). Governing Algorithms: Myth, Mess, and Methods. Science, Technology, & Human Values, 41(1), 3–16. https://doi.org/10.1177/0162243915608948

Homophily & Filter Bubbles’ Impacts Are Overstated or Overemphasized

Bruns, A. (2019). Are Filter Bubbles Real? Polity Press.

Dandekar, P., Goel, A., & Lee, D. T. (2013). Biased assimilation, homophily, and the dynamics of polarization. Proceedings of the National Academy of Sciences, 110(15), 5791–5796. https://doi.org/10.1073/pnas.1217220110

Dubois, E., & Blank, G. (2018). The echo chamber is overstated: The moderating effect of political interest and diverse media. Information, Communication & Society, 21(5), 729–745. https://doi.org/10.1080/1369118X.2018.1428656

Gargiulo, F., & Gandica, Y. (2017). The role of homophily in the emergence of opinion controversies. ArXiv:1612.05483 [Physics]. http://arxiv.org/abs/1612.05483

Kwon, H. E., Oh, W., & Kim, T. (2017). Platform Structures, Homing Preferences, and Homophilous Propensities in Online Social Networks. Journal of Management Information Systems, 34(3), 768–802. https://doi.org/10.1080/07421222.2017.1373008

Möller, J., Trilling, D., Helberger, N., & van Es, B. (2018). Do not blame it on the algorithm: An empirical assessment of multiple recommender systems and their impact on content diversity. Information, Communication & Society, 21(7), 959–977. https://doi.org/10.1080/1369118X.2018.1444076

Zuiderveen Borgesius, F. J., Trilling, D., Möller, J., Bodó, B., de Vreese, C. H., & Helberger, N. (2016). Should we worry about filter bubbles? Internet Policy Review, 5(1). https://doi.org/10.14763/2016.1.401

Opening Remarks to the Senate Committee on Transport and Communications on the Online Streaming Act

October 5, 2022 by Fenwick McKelvey

Introduction

I am an Associate Professor in Information and Communication Technology Policy at Concordia University. My research addresses the intersection of algorithms and AI with media policy. I submit these comments in a professional capacity representing my views alone.

Background

I am speaking from the unceded Indigenous lands of Tiohtià:ke/Montréal. The Kanien’kehá:ka Nation is recognized as the custodians of the lands and waters on which we gather today.

That acknowledgment has great urgency. As a participant in the National Culture Summit, I witnessed a consensus to reimagine cultural policy post-Massey. A future where investments in the Arts and Culture are something other than a settler-colonial project, a future where Cultural Policy is key to governments’ responsibilities and their relations to the many peoples who find their home in Canada.

That future is also bound to public service media in Canada. For all the talk today about disinformation and threats to democracy, the academic evidence is clear. Canada’s resilience depends on public service media. Public service media requires funding and innovation. Change that cannot wait. Every year I am reminded how irrelevant public service media is becoming as I ask my students their own media habits.

These students are part of a new generation of cultural creators working on new platforms under precarious working conditions. We need only be reminded about the coordinated harassment against Clara Sorrenti most recently to know that who gets considered a cultural worker and protected as a cultural worker. I know this better thanks to Margaret McDonald and Dr. Valerie Webber’s research on sex workers online and my collaborations with Saskia Kowalchuk on the meme makers on Tik Tok. These creators need rights.

These are issues of great importance but not directly addressed, or at least clearly addressed, by C-11. I recommend that C-11 is narrowed in scope, and that subsequent legislation addresses the objectives of Canadian cultural policy, renews public service media, and develops rights and protections for digital creators.

C-11

The Act today has one clear objective: ensure that the CRTC has the capacity to regulate large, economically powerful domestic and international firms involved in broadcasting distribution. Online streaming regrettably, remains a broad term, but the mission-critical function of the new Act must address the convergence of large online video-on-demand services and traditional broadcasting distribution undertakings.

The maturation of the streaming service to a few dominant players indicates that online services have become cable by other means. C-11 must ensure the CRTC has the jurisdiction to regulate this converged broadcasting system.

To do so, I see 4 areas for improvement.

1. Discoverability

2. Privacy

3. Reforms to the CRTC

4. Scope

Discoverability

First, the lack of a clear definition of discoverability – as opposed to prominence or catalog quota – undermines a larger movement for algorithmic and AI accountability. I recommend the Act defines discoverability its consequences for regulated entities and its feasibility under the government’s commitments to the USMCA. In addition, I recommend a more coordinated effort across government to implement an Algorithmic and AI Accountability Act.

Privacy

The discoverability question is a proxy question that might be rephrased: does being a private data collector grant broadcasting distributors unfair market power? I think so.

As I have written with my co-author Bram Abramson, we recommend the act adds the following policy objective: “contribute to the protection of the privacy of persons and recognize the public and cultural significance of information about them.”

The amendment recognizes both the need to harmonize privacy and broadcasting policy, but also recognizes the significance of cultural data as a public commons that require care and protection, especially in an age of DALL-E and Stable Diffusion.

CRTC

Given the complexity of these matters, I am discouraged that the Act has not followed the BTLR report’s recommendations for CRTC reform especially the establishment of a Public Interest Committee. As someone who cares and participates actively at the CRTC, I worry the expectations that C-11 places on the CRTC are unrealistic, gives too much voice to corporate interests and too little to public interests. I recommend the BTLR recommendations, Section 1, be considered as part of C-11 or another Act in tandem.

Scope

Finally, I continue to not know the scope of the Act. The lack of clarity and unwillingness to narrow its application to Canada’s media system remains C-11’s enduring shortcomings. I recommend that Section 4.1 be clarified as well as how platform affiliates would be subject to regulation.

These matters are for today. What matters to me most are the future of cultural policy I discussed at the start that remain stalled due to the lack of reforms to C-11.

Navigating the AI stack: Situational Analysis and Collective Mapping to Study AI in Canada’s Media System

February 1, 2022 by Algorithmic Media Observatory

By Meaghan Wester and Sophie Toupin

The Making of a Report on ‘Media Governance After AI’
Workshop Memo December 14, 2021

Participants: Janna Frenzel, Luciano Frizzera, Rob Hunt, Saskia Kowalchuk, Fenwick McKelvey, Jonathan Roberge, Sophie Toupin, Meaghan Wester

Collective Mapping, Translation, Partial Knowledges, and Situational Analysis

Where is AI being used in Canada’s media system? More so, what is Canada’s media system? Before Omicron closed campuses again, our team gathered for a workshop based on situational analysis to collaborate and theorize AI’s regulatory uptake in Canada. Here we want to share our approach oriented around situational analysis.

The Stack as Situation

The workshop marks the beginning of our SSHRC funded-project Media Governance after AI investigation into AI governance and governmentality in what we have come to colloquially call the Canadian AI ‘stack’. We use the term stack since the Internet is often imagined as a mega-structure that is composed of layers (such as applications, networks, etc.). The stack usually has 7 technical layers with politics and governance often described as the higher layers. It’s a metaphor, a diagram to begin our thinking about what is Canada’s media system.

We selected four topics as probes to shed light on this AI stack: 1) the use of AI in data centres, 2) the Canadian AI suppliers list and AI powered botnets, 3) the deployment of artificial intelligence in advertising, and 4) the deployment of artificial intelligence in TikTok’s content recommendation systems.

From these probes, our workshop sought to find their relations and commonalities. We were inspired by situational analysis as an innovative extension of grounded theory. In the approach, the situation becomes the main unit of analysis. It consists in mapping the situations that constitute each topic to layout the major human, non-human, discursive, affective, geopolitical and other elements in the four research situations under study. In our workshop, the stack became our situation.

Situational Analysis in Practice

On December 14, 2021, we collected ourselves to assemble a Canadian AI stack. Four research leads had investigated our four topics over the summer and the workshop began with them reporting back findings.

Situational analysis guided our interactions and discussion. As a methodology, situational analysis seeks to analyze a particular situation of interest through the specification, representation, and subsequent examination of the most salient elements in that situation and their relations. It was developed by the American sociologist Adele E. Clarke, and it emphasizes the following elements all of which have served as an orientation for the workshop and the subsequent report writing:

Mapping as an analytic strategy
Emphasis on reflexivity of researchers
Emphasis on difference in data
Emphasis on silences and absent positions
Emphasis on nonhuman elements
Emphasis on power

The workshop was composed of 8 participants including 4 area experts (i.e., students whose findings following their research were discussed). Accordingly, we proceeded to do four rounds producing each round one map on the topic at hand. Each round took place in the same fashion.

The area experts presented their summer findings for 15 minutes. While talking the other participants wrote on post-it notes salient points, keywords, related questions in order to visually and spatially represent the provided information. When the area expert was done presenting a summary of their findings, other participants asked questions, raised interesting observations and connections. During this discussion phase, post-it notes were moved around on the cardboard map while some were added to reflect the discussion that was taking place. Importantly, this process of discussion and representation engaged principles of partial knowledges, translation as well as reflexivity as we collectively co-created the messy map.

Once we had produced four area experts’ maps, we moved to the second phase of the workshop which was meant to orient how this information may be represented as a diagram or a ‘stack’. We collectively crafted categories to rehash and re-organized the post-its. The aim with the categories was to create cross-references; we came up with 4 pairs of topics (controversy/policies, infrastructure/machine agencies, time/space, suppliers/buyers) that we felt would enable us to unfold major themes and help structure our report. We freely posted the post-its inside the categories or on their borders. Once we had moved all post-its from the maps to the categories, we discussed the clusters that emerged and our choice of category headings.

Everyone attending left with an assigned category and the adjacent post-its. As a homework, participants were tasked to reflect on the category they had chosen and its relation to the post-its with the aim of synthesizing discussions and creating transversal themes across the four topics under study.

The purpose of the workshop was to move from individually imagining a situation to a collective understanding of that same situation. The richness of the collective process brought to the table each of the participants expertise and standpoints. Throughout the process, we paid close attention to what was missing on our post-its as we came up with the categories, as well as what did the categories failed to capture in turn producing leftovers.

Outputs

During this workshop, we co-produced:

4 maps of the four expertise areas
8 sister categories (4 pairs)
8 memos following the workshop that will serve as a basis for the report’s structure.

We are planning to use this workshop as the basis for a report to be developed in 2022. Less tangible, the workshop not only assembled an AI stack, but re-oriented our research team around a shared theorization.

Limits

As a first try, we learned a lot of how to improve future workshops. After discussing with the participants, we identified two main points to improve future workshops.

Some participants voiced the lack of clarity on how to put together common expertise. Before we began the workshop, participants were unsure how the exercise would unfold in practice. This could be remedied by showing slides at the beginning of the workshop where we briefly explain past workshop iteration, and model what is expected from the participants.
Another limit we encountered was the production of four different maps. One suggestion to avoid the multiplicity of maps, is to work on the same piece of paper for all area experts presentations. There are two main reasons why this could improve the overall experience and output of the workshop. First, by working off the same map, the participants could more naturally link together ideas by placing post-its in proximity with one another and placing arrows to show relationality. As such, we would produce a more relational pool of data which is what we were aiming for in the synthesis part of the exercise. The second reason why working on the same map would be beneficial is because the group could start producing the transversal theme together instead of leaving it on the participants during their memo drafting. This would again force a very productive process of translation and reflexivity.

Next Steps

Following this workshop, each participant is expected to send their memo on their assigned categories to a member of the team to start organizing the structure of the report.

We are hoping to test and improve this methodology with other research processes.

Comment on Canada’s Online Harms Regulatory Framework

September 28, 2021 by Algorithmic Media Observatory

25 September 2021

To Whom It May Concern:

My comments address the technical paper that will inform the government’s approach to online harms. The technical paper proposes to:

establish new information sharing between police, security agencies, and OSCs as well as create new obligations for TSPs;
mitigate five distinct categories of criminal activity or online harms; and,
propose a new regulatory framework for the oversight of commercial content moderation online.

These three objectives, debatable in their own right, invoke distinct policy traditions and fields of expertise. I am concerned that the technical paper conflates these three separate issues into one legislative agenda.

My submission focuses on the third objective, the new regulatory framework. Before focusing on the third objective, I note:

Changes to the administration of the criminal code for OCSs must be treated separately from the regulatory framework for content moderation

At present, the technical paper too often frames online harms as a policing problem at a time when the biases and oversight of Canada’s policing services are evident and calls for reforms clear and needed. The distinction between online harms and criminal activities remains ambiguous in the technical paper. Conflating harm and criminal acts risks deputizing OSCPs with both enforcement and police reporting for criminal activities (1). Proposal 20 specifically requires a separate consultation phase and it should not be assumed that because automated content takedowns are happening that automated content takedowns are an effective or central instrument to address online harms. Furthermore, the technical paper’s overall focus on OSC regulation is diluted with its discussion of new measures for Internet services and new blocking/filtering obligations for Telecommunications Service Providers. These powers are out of scope and arguably within the power of the CRTC to implement if needed already.

Comparability of these five online harms is debatable and more targeted legislation may prove more effective

The five online harms need further definition. Furthermore, the nuances of each online harm, such as the national and international dimensions of terrorist activities, for example, may not be well suited for an omnibus framework (2). Protecting Canada’s democracy, ostensibly another online harm, has been addressed through reforms to Canada’s Elections Act.

Online harms require a whole of society approach that is out of scope with aspects of this bill focused on OSCs

More accountability to commercial content moderation has not and will not resolve the root causes of online (3). Rather, better regulation of already-existing content moderation is enough of a regulatory accomplishment without the added challenge of suggesting that content moderation as a first response to systemic injustice.

With these primary concerns in mind, I move to the administrative aspects. I acknowledge that content moderation is a needed part of inclusive communication systems, but certainly not more important than matters of access, affordability, and inclusion. As part of these the broader reforms to Canada’s communication system, the technical paper that:

Defines the regulatory category of OSC in line with the CRTC’s suggested reforms in CRTC 2017-359 for new domain/industry-specific media regulation categories distinct from a TSP;
Establishes a Digital Safety Commission that includes a Digital Resource Council of Canada, and an Advisory Board to monitor compliance and administer OSCPs enforcement of online harms;
Sets new obligations for OSCPs to report and filter 5 types of illegal content; and,
Grants the DSC new powers including AMPs for OSCPs as well as inspection powers.

The regulatory framework seems a viable opportunity if primarily seen as a mechanism to enhance oversight and transparency for commercial content moderation (4) and algorithmic filtering (5). The DSC’s powers to investigate are welcome additions to Canada’s media institutions especially since the DSC is subject to the Access to Information Act. As access to data is a primary barrier to research into OSCs (6), the DSC may enhance public knowledge of largely opaque moderation practices.

The DSC’s mandate must be defined with clear policy objectives. I recommend its policy objectives follow Dr. Suzie Dunn who suggests that, “Canada’s approach to regulating platforms should centre human rights, substantive equality, and intersectionality, and employ a trauma-informed approach.” (7)

I am expressly supportive of DSC’s power to investigate how an OSCP monetizes harmful content (14f) and encourage more attention to this function in the subsequent act.

The technical paper, at present, does not provide a timeline for the constitution of the DSC. I recommend that:

Reporting and inspection powers be prioritized as a first step before automated takedown obligations come into effect;
The DSC establish a Technical Standards Committee with Measurement Canada, the CRTC, OSCPs, civil society, and academics to develop information gathering and potentially an equivalent of the CRTC Monitoring Report to establish a public record about the threat of online harms and the state of commercial content moderation and automated content recommendation;
All matters of composition (46, 64, and 71) be changed from considering inclusive membership to requiring inclusive membership and establish better democratic oversight of candidate selection and vetting as proposed in the BTLR report;
Ensure sufficient budget and effective reporting mechanisms before implementing blocking and filtering regimes.

My timeline emphasizes focusing the DSC first on enhancing transparency first on commercial content moderation then secondly considering its effectiveness to combat online harms. Ideally, other measures or more dedicated initiatives could develop simultaneously taking a whole-of-society approach to the 5 identified online harms.

The present risk is that the 24-hr takedown requirement along with a lack of penalties for false positives may encourage OSCPs to further invest in automated content moderation, especially artificial intelligence as a form of media governance.

The consideration of automated content regulation is lacking in the current working paper and needs substantive consideration. The technical paper does not address its responsibility nor its legitimization of artificial intelligence as used by OSCPs to classify, filter, and demote harmful content. The technical paper proposed a regime legitimating automated content regulation at scale without sufficient records of the efficacy of the systems in Canada’s both official languages and in Canada’s multicultural society. The technical paper needs a substantively expanded discussion of AI accountability including times when the potential risks require the prohibition of automated solutions (8).

The DSC may need powers to designate standards for content moderation work that then prohibit AI as high-risk applications and better accountability mechanisms. Inversely, outsourcing and ghost labour in commercial content moderation require better labour standards and safer working environments. At present, the labour of moderation is assumed to be automatable and without long-term harm to the workers.

The DSC must be seen as a promising beginning that must proceed cautiously to build knowledge, expertise, and autonomy before implementing content takedowns (11a) recognizing instead that better accountability to already-existing content moderation, reporting and oversight is a needed first step before assuming that content takedowns will be an adequate form of online harms that require a whole-of-society approach.

The technical paper and discussion papers mark an early first step that hopefully leads to a more fulsome consultation, public record, and clearer legislative agenda. I continue to support these efforts in my research and my opinion here.

Sincerely,

Fenwick McKelvey

References:

(1) For a distinction, see: Tenove, Chris, Heidi Tworek, and Fenwick McKelvey. “Poisoning Democracy: How Canada Can Address Harmful Speech Online.” Ottawa: Public Policy Forum, November 8, 2018. https://www.ppforum.ca/wp-content/uploads/2018/11/PoisoningDemocracy-PPF-1.pdf.
(2) For a more detailed discussion of focused reforms, see: Khoo, Cynthia. “Deplatforming Misogyny.” Technology-Facilitated Violence. Toronto: Women’s Legal Education and Action Fund, 2021. https://www.leaf.ca/publication/deplatforming-misogyny/.
(3) McKelvey, Fenwick. “Toward Contextualizing Not Just Containing Right-Wing Extremisms on Social Media: The Limits of Walled Strategies.” SSRC Items (blog), July 13, 2021. https://items.ssrc.org/extremism-online/toward-contextualizing-not-just-containing-right-wing-extremisms-on-social-media-the-limits-of-walled-strategies/.
(4) Roberts, S. T. (2019). Behind the screen: Content moderation in the shadows of social media. Yale University Press.
(5) Hunt, R., & McKelvey, F. (2019). Algorithmic Regulation in Media and Cultural Policy: A Framework to Evaluate Barriers to Accountability. Journal of Information Policy, 9, 307–335. JSTOR. https://doi.org/10.5325/jinfopoli.9.2019.0307
(6) Tromble, Rebekah. “Where Have All the Data Gone? A Critical Reflection on Academic Digital Research in the Post-API Age.” Social Media + Society 7, no. 1 (January 1, 2021): 2056305121988929. https://doi.org/10.1177/2056305121988929; Tworek, Heidi. “Open Access to Data Is Critical in a Democracy.” Centre for International Governance Innovation, August 25, 2021. https://www.cigionline.org/articles/open-access-to-data-is-critical-in-a-democracy/.
(7) Dunn, Suzie, William Perrin, and Heidi Tworek. “What Can Canadian Law Makers Draw from the New UK Online Safety Bill?” Centre for International Governance Innovation, May 20, 2021. https://www.cigionline.org/articles/what-can-canadian-law-makers-draw-new-uk-online-safety-bill/.
(8) Balayn, Agathe, and Seda Gürses. “Beyond Debiasing: Regulating AI and Its Inequalities.” Brussels: European Digital Rights, 2021. https://edri.org/our-work/if-ai-is-the-problem-is-debiasing-the-solution/.