Panel:
Moderated by: Aman Nair, Researcher, Digital Futures Lab
Generative AI systems like ChatGPT and Midjourney have recently established themselves within the global popular consciousness. The emerging narrative around these systems has focused on their groundbreaking abilities and potential to revolutionise multiple facets of modern society including work, communication and trade.
However, a crucial element omitted from the discourse around these AI systems is the hidden labour required to ensure that they remain operational. Data annotators play a pivotal role in ensuring that data captured from multiple sources across the internet are labelled correctly so that they can be utilised properly by these AI systems. In the case of generative AI systems, the massive quantity of training data utilised requires substantial human labour to be labelled properly. Much of this work is currently outsourced from the Global North and done by workers in the Global South - particularly women.
This has raised questions about whether such work can act as an emancipatory tool by providing traditionally marginalised individuals and communities with financial gains, or whether they reproduce systems of oppression and exploitation against those at the bottom of the economic and social ladder.
The panel attempted to unpack this question through a discussion on the challenges currently present in the sector, the responsibility of various actors involved as well as the path forward to create ethical data work. Some key points from the discussion are listed below:
Safiya spoke about the vision behind Karya to alter the paradigm of existing data annotation work from one of exploitation to one of community empowerment. She explained how Karya has implemented an adapted cooperative model to account for the microwork and gig work nature of data annotation work.
She emphasised the necessity to foster continued community ownership over labelled data as a mechanism to ensure that workers are fairly compensated for their work. Karya has evolved to identify and enfranchise individuals who have historically been excluded from the digital economy, for example, through their focus on providing work for lower caste women. Building trust among communities occurs through the help of local NGO partners that have existing close relationships with communities. Findings from Karya’s work have pointed to the success of this model in promoting a cooperative sense of work where individuals work together rather than compete with each other.
Safiya noted that while Karya functions as a not for profit organisation, it is not necessary for all data annotation firms to do so. Existing market conditions allow for the simultaneous existence of fair working conditions and wages for workers as well as sufficient profits. She also suggested the improved technology, particularly in the context of data validation, could improve the efficiency of work and thereby lead to greater pay to workers for the same effort.
Billy shed light on some of the conditions faced by data workers across the globe, noting how many of them were subject to long working hours, possessed few legal protections, and were often (in the case of content moderators) exposed to distressing material that could lead to post traumatic stress. He explained that while many data annotation companies propagate a narrative of empowerment, they operate through the exploitation of the worker. Furthermore, given the intricate nature of intermediaries and subsidiaries within these organisations, there is little transparency between the AI companies in the Global North soliciting such a firm and the workers on the ground in the Global South performing the work - creating conditions where such exploitation becomes easier. It is the presence of many of these intermediaries that absorbs much of the pay that would have otherwise been allocated to workers.
Billy noted that while efforts at unionisation and collective bargaining have begun to emerge among data workers in the Global South, there continue to be structural impediments to widespread action. The inherent precarity associated with the work, the worry unionisation will drive up prices and drive away clients and retaliation by firms against workers that attempt to unionise all limit efforts of establishing collective solidarity among workers.
Speaking on the role of media to frame the narrative around these issues, Billy emphasised the necessity to move away from a narrative of AI systems as being purely automotive systems to one that highlighted the human labour behind them. He also highlighted the need for consistent and repetitive messaging on these issues to ensure that they seep into the popular consciousness and are eventually addressed by policymakers.
Noopur drew attention to how narratives around data annotation as gig work often frame this work as banal, performable by anybody and easy to do anywhere that would be better off being automated or mechanised. However, this ignores that within the current economic landscape innumerable individuals are reliant on such work as a significant income source - and that calls to automation result in a clear loss for these workers. Moreover conceiving of data annotation through this rubric of ‘easy to do work’ fundamentally underestimates the skill involved in performing these tasks, resulting in an undervaluation of the work being performed and a subsequent underpayment of workers. In reality however, not only does the data annotation work require skill, but generates value far in excess of the pay that is being provided to the workers who are creating this value.
Noopur explained how it is essential to integrate numerous presuppositions that underlie data annotation work within the Global South. Firstly, it is essential to reject the framing of the Global North as the centre of technological innovation and the Global South as the labour that actualises that innovation. Secondly, legal and social protections afforded to workers must be seen in the context of a globalised neo-liberal economic order where corporations are constantly in search of new cheaper labour markets. This can only be countered by emphasising the rights of workers as going beyond simply data or internet rights, and viewing them as workers that deserve strong labour protections.
Speaking on the role of academics in working towards creating fairer conditions for workers, Noopur stressed the necessity to educate people and re-orient the narrative around AI and data work - a narrative that is currently dominated by the words of technocrats and entrepreneurs in the Global North.