Some critical and ethical perspectives on the empirical turn of AI interpretability

We consider two fundamental and related issues currently facing the development of Artificial Intelligence (AI): the lack of ethics, and the interpretability of AI decisions. Can interpretable AI decisions help to address the issue of ethics in AI? Using a randomized study, we experimentally show that the empirical and liberal turn of the production of explanations tends to select AI explanations with a low denunciatory power. Under certain conditions, interpretability tools are therefore not means but, paradoxically, obstacles to the production of ethical AI since they can give the illusion of being sensitive to ethical incidents. We also show that the denunciatory power of AI explanations is highly dependent on the context in which the explanation takes place, such as the gender or education of the person for whom the explication is intended. AI ethics tools are therefore sometimes too flexible and self-regulation through the liberal production of explanations does not seem to be enough to resolve ethical issues. By following an STS pragmatist program, we highlight the role of non-human actors (such as computational paradigms, testing environments, etc.) in the formation of structural power relations, such as sexism. We then propose two scenarios for the future development of ethical AI: more external regulation, or more liberalization of AI explanations. These two opposite paths will play a major role in the future development of ethical AI.