This keynote explores the norms and governance of recommender systems on digital platforms like YouTube and TikTok, especially in relation to user-generated content. It addresses concerns about the platforms' algorithmic systems contributing to user harm and challenges the notion of platforms as mere content conduits. There are three main points: Firstly, the need to question the established definitions of recommendation-related harms and to encourage diverse frameworks for evaluating these systems. Secondly, the importance of considering long-term efects of information landscape commercialization and the potential of algorithmic recommendation for elevating historically excluded voices. Lastly, the keynote calls for greater appreciation for the nature of the 'items' being recommended, which opens up possibilities for more sophisticated discussions on normative frameworks for curation.
CEUR
ceur-ws.org
1. Keynote
In my keynote, I discuss the question of norms underlying recommender systems from a platform
governance perspective, and I focus on commercial digital platforms hosting user-generated
content operating in the cultural and entertainment space – platforms like YouTube, TikTok,
Instagram, and so on. Over the past few years there has been mounting concern on the part
of civil society, academia, and regulators about platforms’ algorithmic recommender systems,
forming part of a growing trend to hold platform companies to account for the ways in which
their own algorithmic systems and processes actively contribute to user harm. This challenges
platform companies’ historic tendency to present platforms as ‘mere conduits’ of content,
downplaying their active curatorial role [
In this keynote, using examples from my own research and that of others, I make three main
points. First, I argue that we need to resist locking in framings of recommendation ‘problems’,
especially over the next few years as regulatory processes in the form of algorithm audits become
formalised. What constitutes recommendation-enabled ‘harm’ is not a foregone conclusion:
we must question who is being given the power to define what a recommendation problem
or harm is, and who is being excluded [
© 2023 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0). harmful or borderline harmful content, privacy concerns related to user profiling and targeting are some framings of ‘the recommendation problem’ that seem to have dominated the public debate, to varying degrees. They are also implicit or explicit in recent policies by platforms and third-party regulators. I argue that we need to approach these framings with some caution – not to discard them, but to ask what they exclude.
Second, I argue that there is value in going back to some of the older literature that highlights
concerns about the over-commercialisation of the information landscape and the implications
this has on which voices are being heard over the long term [
Finally, drawing on the work of Rieder [11], I argue that normative frameworks for recommendation must carefully account for the nature of the ‘items’ being recommended. Rieder [11] argues that our tendency to talk in terms of ‘items’, ‘information’ and ‘content’ has a lfattening tendency, it brings “entire [human] domains into the fold of computing” but discounts all the existing frameworks and professional norms we can draw from – frameworks which would allow us to have more sophisticated normative debates about good curation. In my own research, for example, I draw from scholarship on librarianship (e.g., [12, 13, 14]) and non-punitive, speech preserving national approaches to regulating historical narratives in democratic countries [15] to (re)imagine what the responsible algorithmic recommendation of history books on the Amazon Bookstore might look like. [8] A. J. Christian, K. C. White, Organic representation as cultural reparation, JCMS: Journal of Cinema and Media Studies 60 (2020) 143–147. [9] S. Vrijenhoek, M. Kaya, N. Metoui, J. Möller, D. Odijk, N. Helberger, Recommenders with a mission: assessing diversity in news recommendations, in: Proceedings of the 2021 conference on human information interaction and retrieval, 2021, pp. 173–183. [10] N. Helberger, On the Democratic Role of News Recommenders, Digital Journalism 7 (2019) 993–1012. URL: https://doi.org/10.1080/21670811.2019.1623700. doi:10.1080/21670811. 2019.1623700. [11] B. Rieder, Engines of order: A mechanology of algorithmic techniques, Amsterdam University Press, 2020. [12] E. Drabinski, Queering the catalog: Queer theory and the politics of correction, The
Library Quarterly 83 (2013) 94–111. [13] S. U. Noble, Algorithms of oppression, in: Algorithms of oppression, New York university press, 2018. [14] H. A. Olson, The power to name: Representation in library catalogs, Signs: journal of women in culture and society 26 (2001) 639–668. [15] E. Heinze, Beyond ‘memory laws’: Towards a general theory of law and historical discourse, Forthcoming in Law and Memory: Addressing Historical Injustice by Law (U. Belavusau & A. Gliszczyńska-Grabias, eds., Cambridge University Press), Queen Mary School of Law Legal Studies Research Paper (2016).