Selected Papers of #AoIR2024: The 25th Annual Conference of the Association of Internet Researchers Sheffield, UK / 30 Oct - 2 Nov 2024 Valente, J. Ustek Spilda, F. Alyanak, O. Brittain, L. and Graham, M. (2024, October). Fairness in the work behind the AI industry: how action-research approaches can build better labour conditions in AI development. Paper presented at AoIR2024: The 25th Annual Conference of the Association of Internet Researchers. Sheffield, UK: AoIR. Retrieved from http://spir.aoir.org. FAIRNESS IN THE WORK BEHIND THE AI INDUSTRY: HOW ACTION- RESEARCH APPROACHES CAN BUILD BETTER LABOUR CONDITIONS IN AI DEVELOPMENT Jonas C. L. Valente Oxford Internet Institute Funda Ustek Spilda Oxford Internet Institute Oguz Alyanak Oxford Internet Institute Lolla Brittain Oxford Internet Institute Mark Graham Oxford Internet Institute Extended abstract Artificial Intelligence (AI) has gained attention in the last decade because of its ability to transform how we live (Crawford, 2021). This change is not inherently good or bad, but it is not neutral also (Cole et al., 2022). Instead, the ongoing development of AI systems presents us with opportunities and threats that will emerge in different ways in different contexts. The blossoming of AI has been accompanied by an expansion in the discussion of AI ethics (Hagendorff, 2020). As a result, scholars and a broad range of stakeholders have begun to ask how AI affects society and how it should be developed and used (Jobin et al., 2019). Regarding the relationship between AI and work, the opportunities and challenges have also been highlighted in intense controversies, particularly about the effect on workplaces and labour relations. AI has also been seen as a transformative technology capable of boosting productivity and improving the labour process (OECD, 2023). On the other hand, the literature has also revealed risks and harms for workers (Benanav, 2020). One first aspect relates to the consequences of AI deployment on the number of jobs. Studies diverge on the range of reach of occupation substitution, with more significant (Frey & Osbourne, 2013) or limited effects of AI-based automated systems (Arntz et al., 2017). Some studies point out possibilities of job augmentation (Gmyrek et al., 2023). However, less attention has been paid to the work behind the AI industry. An AI System Artificial Intelligence (AI) is a set of technologies that “seeks to make computers do the sorts of things that minds can do. (Boden, 2016). These machine-based systems can make predictions, recommendations, or decisions influencing real or virtual environments. However, in order to do that, an enormous amount of human labour is needed throughout the AI development lifecycle (Steinhoff, 2021). Along this cycle, there are several tasks which demand human labour. Although their names vary throughout the literature and companies, they encompass actions such as data cleaning, data annotation, data labelling, model revision, content moderation, sentiment analysis, tagging or detection, relevance scoring, audio transcription, etc (Miceli et al., 2020). This work is made available by different types of companies. Digital labour platforms are key actors in organising the buying and selling of on-demand workers’ labour power internationally, constituting a planetary labour market (Graham & Ferrari, 2022). Research has shown the growth of these platforms. The ILO (2021) counted an increase in the number of digital labour platforms from 142 in 2010 to 777 in 2020. Regarding online remote work, or cloudwork, platforms, Datta et al. (2023) estimated a range of 154- 535 million workers worldwide. Among them are those working on microwork platforms with many of the tasks mentioned above (Tubaro et al., 2020). Other companies have also been providing a workforce for AI development, such as Business Processing Outsourcing (Muldoon et al., 2023). The risks and challenges in the labour relations behind the AI industry have been raising concerns (Morley et al., 2023; Miceli & Posada, 2022). For instance, digital labour platforms can facilitate precarious arrangements with low pay, poor working conditions, inaccessible and unreasonable conditions, unfair management, and a lack of representation (Woodcock & Graham, 2019). Comprehending the labour relations in the AI industry is an increasingly important task since many critical aspects of AI systems rely on global production networks characterised by atypical and, many times, precarious arrangements (Howson et al., 2022). However, more than investigating, the limitations and problems of these production models pose a further effort of flagging and challenging problematic practices and promoting, on the other hand, fair work on these supply chains. Drawing on these assumptions and concerns, the paper has two main goals. First, the aim is to investigate the current labour conditions in the AI industry and point out the lack of quality of outsourced jobs offered by digital labour platforms and Business Process Outsourcing companies, calling attention to the challenges faced by workers and the human costs of AI systems deployment. Second, it presents world research project results on how not only to carry on international studies on the topic but also how to use an action-research approach to generate impact through public scoring companies and encourage the adoption of best practices by them. The paper’s methods are based on the project’s action-research approach and methodological framework. The project investigates and highlights these enterprises' best and worst examples and examines their implications. Through assessing working conditions, the project sheds light on the current state of these companies and tries to contribute to shaping the future of work in a way that is fair, inclusive, and sustainable for all stakeholders, especially the workers. Companies are scored based on principles that address the major issues that define labour relations, such as pay, conditions, contracts, management, and representation. The analytical framework is adjusted with a specific set of standards within each principle depending on the kind of work analysed (cloudwork and AI-facilitated work, for instance). Each company evaluated receives a score out of ten based on their evidence of adherence to the five principles. Each of the principles has two thresholds, and accordingly, companies can receive scores from 0 to 10. A maximum score of 10 means that a platform meets the minimum standards of fairness.  The methodological framework involves three data sources: desk research, workers’ surveys and platform managers' interviews. The research team then gathers the information gathered for each study, granting or not points when evidence is found that platforms meet each of the ten thresholds. During the process, there is constant engagement with different stakeholders, from workers to platform managers. With the latter, the research team highlights worse practices and problems found before final scores to foster changes in line with the principles and standards. The paper will explain and detail the action-research approach and the methodological framework employed in the studies on cloudwork platforms and AI supply chain companies. Furthermore, it will present findings in line with the two goals mentioned above. First, it will discuss the assessment of the cloudwork platform conducted in 2023, which analysed fairness in 15 leading global and regional web-based platforms. The results include how each platform meets each principle and the picture of the work behind AI in these companies. In addition, the paper will examine labour relations in BPO companies using a case study of the company Sama, which is based in the United States and operates in many African countries. Although the company calls itself an ethical sourcing business, our research found many problems, such as low pay, excessive working hours, unfair management practices, short-term contracts and job insecurity, and problematic surveillance practices. After presenting the studies’ findings on the labour conditions on cloudwork platforms and AI companies, the paper will show how the project’s action-research approach, including engagement with the management teams, led to the implementation of over 50 best practices on fairer working conditions in those companies. The discussion will focus on how action-research strategies can be designed and implemented, its paths and barriers, and how they can contribute to creating effective impacts and indicating how the platform economy can be. References Arntz, M., Gregory, T., & Zierahn, U. (2017). Revisiting the risk of automation. Economics Letters, 159, 157-160. Benanav, A. (2020). Automation and the Future of Work. Verso Books. Cole, M., Cant, C., Ustek Spilda, F., & Graham, M. (2022). Politics by automatic means? A critique of artificial intelligence ethics at work. Frontiers in Artificial Intelligence, 5, 869114. Crawford, K. (2021). The atlas of AI: Power, politics, and the planetary costs of artificial intelligence. Yale University Press. Gmyrek, P., Berg, J., Bescond, D. 2023. Generative AI and jobs: A global analysis of potential effects on job quantity and quality, ILO Working Paper 96 (Geneva, ILO). Hagendorff, T. (2020). The ethics of AI ethics: An evaluation of guidelines. Minds and machines, 30(1), 99-120. Howson, K., Ferrari, F., Ustek-Spilda, F., Salem, N., Johnston, H., Katta, S., ... & Graham, M. (2022). Driving the digital value network: Economic geographies of global platform capitalism. Global Networks, 22(4), 631-648. Irani, L. (2012). Microworking the crowd. Limn, 1(2). Jobin, A., Ienca, M., & Vayena, E. (2019). The global landscape of AI ethics guidelines. Nature machine intelligence, 1(9), 389-399. Miceli M and Posada J (2022) The data-production dispositif. arXiv:2205.11963. arXiv. Miceli, M., Schuessler, M., & Yang, T. (2020). Between subjectivity and imposition: Power dynamics in data annotation for computer vision. Proceedings of the ACM on Human- Computer Interaction, 4(CSCW2), 1-25. Morley J, Kinsey L, Elhalal A et al (2023) Operationalising AI ethics: barriers, enablers and next steps. AI & Soc 38:411–423. Muldoon, J., Cant, C., Graham, M., & Ustek Spilda, F. (2023). The poverty of ethical AI: impact sourcing and AI supply chains. AI & SOCIETY, 1-15. OECD (2019). Artificial Intelligence in Society. Paris: OECD. Steinhoff, J. (2021). Automation and Autonomy. Springer International Publishing.