A/HRC/57/70 otherwise structural problem, which consequently fuels the manifestation of inequality and violent outcomes. 26. Gift Mwonzora, Willy Brandt School of Public Policy at the University of Erfurt, focused on health and healthcare, nutrition, and food security. He presented findings from his research and highlighted examples of precarity in the labour market as a result of automation and digitalization in agriculture. In South Africa, citrus farming (horticultural sector) mostly employs a large part of the female labour force, in particular women of colour. While picking fruits largely remains a physical activity still involving human intervention, mechanization in some processes of production, such as in sorting and grading, has seen women losing or in fear of losing their jobs. Further compounding the existing vulnerabilities of the casualization of labour, low wages and the seasonal nature of work in agriculture. Increasingly, fewer women are employed in such sectors as a result of being replaced by AIdriven production processes. In other sectors, the use of drone delivery systems in the medical field in Malawi and Rwanda was addressing some of the challenges of lack of access to healthcare in remote areas, constituting positive use AI and technology. However, ethical concerns remain due to a lack of duty of care regarding people of African descent in medical trials. B. Racial bias in the technology sector 27. The technology sector has been criticized for its lack of diversity, favouring white, affluent males. Large-scale AI systems are developed almost exclusively in a small number of companies and elite university laboratories which engage mostly white males and have a history of discrimination against and exclusion of ‘others’, including people of African descent. Technology that is developed and produced in fields that disproportionately exclude people of African descent are more likely to reproduce racial inequalities. 28. The creation of AI systems begins with data — its extraction, organization, and subsequent modelling. Each step in this process holds the potential to introduce or perpetuate racial bias, significantly impacting the healthcare outcomes for people from racial or ethnic groups. AI systems are trained on enormous quantities of data, mostly on non-Black populations, which are used to build models of behaviour. The designers and developers of machine learning and AI systems can therefore intentionally or unintentionally introduce biases into their algorithms through the utilization of prebuilt models which contain racial biases, as evident in some generative AI systems being unable to create accurate and realistic depictions of Black people. How developers obtain such critical data raises ethical questions. Data acquisition practices often lack transparency, with instances where data is obtained without proper consent or through exploitative means. 29. Facial recognition software used by governments and the police disproportionately affect people of African descent to learn and propagate biased associations between race groups and negative attributes, exacerbating racial inequality. In 2015, for example, Google had to apologize after its image-recognition app mistakenly labelled African Americans as “gorillas”. 30. The surveillance practices from times of enslavement and colonization which persist up to today, can and have been made worse with the use of AI as research has consistently shown greater inaccuracies among non-white populations. This has already led to several dangerous situations for people of African descent, such as being falsely identified as a suspect for a crime. Accounts of the disproportionate levels of harm from face recognition software experienced by people of African descent are well-known. 31. The lack of transparency and accountability in AI development exacerbates these issues. Many AI systems are developed and deployed by private companies that do not disclose their algorithms’ inner workings, citing proprietary concerns. This opacity makes it difficult for independent researchers, policymakers, and the affected communities to scrutinize and challenge biased algorithms. Without transparency, it is nearly impossible to hold developers accountable for the adverse impacts of their technologies on marginalized groups. Moreover, there are often no mechanisms in place to audit or regulate AI systems effectively. Regulatory bodies lack the technical expertise and resources needed to assess the 8