Abstract
Attempts to remedy issues arising from the development, deployment, and use of Artificial Intelligence systems rely on ethical prescriptions in the form of ethical guidelines, technical solutions, and/or legal or regulatory frameworks. These current remedies might be sufficient for addressing the class of problems that can be easily tracked as direct consequences of the AI systems. However, I argue that these interventions do not go deep enough, nor do they address the root causes of the problem. This dissertation, advanced through four independent but related articles, defends the view that injustices from AI systems should be understood and addressed, in the first instance, as an oppression problem. To track and overcome the hidden oppressions lurking in AI systems and even in AI ethics, we ought to extend our focus beyond apparent injustices like unfair algorithmic outputs to the underlying norms and values embedded in a particular social imaginary that circulates and sustains oppressions. This is the deepest and most foundational justice intervention required for the ethical development and deployment of AI systems. The first article defends the view that algorithms and AI artefacts perpetuate a form of oppression that I call “cognition-disabling epistemic oppression”. I tentatively define cognition-disabling epistemic oppression as oppression of non-dominantly situated knowers that impinges on their cognitive evaluation of the diverse ways of experiencing the world in ways that obscure, enforce, and/or reinforce forces of epistemic oppression. I show the limitations and weaknesses of the current approaches—ethical, technical, and regulatory—to track the oppression that AI artefacts sustain. In the second article, I argue for the decolonisation of AI ethics through a systems thinking approach by going beyond efforts to transform ‘the thing’ itself. Proper decolonisation of AI ethics ought to consider the whole algorithmic ecosystem, which encompasses values, norms, resources, and standards that support algorithmic development and use. I show that the Algorithmic Fairness approach, at best, recognises the ecosystem in which algorithms are embedded but does nothing to transform it. The algorithmic ecosystem is a complex system of socio-epistemic-technological interactions that function in interdependent and often reinforcing ways to shape how algorithms are made and used. The decolonisation view through systems thinking would guarantee epistemic, social, and economic benefits for non-dominantly situated people. In the third article, I advance a novel conception of Ubuntu as a moral framework that can overcome the oppression problems given its emphasis on instituting a just moral environment. This conception of Ubuntu, which I call Unbounded Ubuntu, situates the prime moral properties in the state of the community rather than in individuals’ internal capacities or communal properties, as often defended by proponents of Ubuntu. In the fourth article, I argue that Ubuntu enables third-order justice that can overcome the oppressions prevalent in AI ethics. I take Kristie Dotson’s distinction of the three levels of epistemic oppression as the framing architecture of this article. From this I build out three levels of justice and argue that current efforts in AI ethics at best, satisfy the first two levels of justice. Ubuntu as an ethical framework, in contrast, would address all three levels of justice and as such provide the proper foundation to construct the ethics of AI that is just in all possible ways. Taken as a whole, the dissertation addresses two primary categories of challenges in the epistemology and ethics of AI—those related to the content of AI ethics and those pertaining to its practical implementation.