The development of a “human-centered” AI tends to overlook a component of AI development: data production. Machine learning data sets are used to develop and train algorithms. The resulting algorithms are then used for content moderation, facial recognition, or to train self-driving cars, etc. The amount of data needed for this, has to be sorted, cleaned, annotated, labelled. These tasks are far from automated: they are outsourced to manual labor in countries where wages and working conditions are low. From a data justice perspective, is this reality considered when thinking about regulating AI? Issues related to the exploitation of labour, social justice and power remain unaddressed. In the EU, the 8 legal requirements in the draft AI Act will be operationalized through harmonized standards. We will reflect about the ability of standardization committees to address the issue of data justice.
• In the debate about how best to regulate AI, is there space for a “social and data justice” perspective?
• Should the draft AI Act address this issue? In what way?
• Is there a possibility in the standardization committees to address the issue of data justice for data workers?
• Should standardization bodies be “revisited” in their membership, mandate and production process in order to address this issue?
