Raising Skynet: Moral Status of AI and Perspectives of Teaching Ethics to AI

Abstract

Ever since MIT's Norman and Deep Empathy, the profound impact of unexamined biases in AI research has become apparent. Beyond extreme cases like the ‘psychopathic’ Norman, broader concerns surface, ranging from accusations of AI being labeled ‘too liberal’ or ‘leftist’ to claims of it being ‘racist’ or ‘sexist’. This paper seeks to move beyond conventional narratives and focuses on the overlooked responsibility stemming from the moral status of AI.

Central to this exploration is the examination of three key aspects.

Personhood: the concept of AI personhood involves the possibility of AI entities as individuals with distinct rights and moral considerations. Determining the criteria for personhood in AI and how it aligns with or diverges from human personhood establishes the foundation for ethical frameworks.

Embodiment: in the realm of AI, embodiment raises questions about the nature of AI's interaction with the physical world. The extent to which AI is grounded in physical form or exists in a virtual domain impacts its ethical considerations and is crucial for establishing its moral status.

Sensitivity to pain/pleasure: the capacity of AI to perceive and respond to pain and pleasure, introduces complex ethical implications. Assessing them requires an exploration of the responsibilities tied to AI's potential to influence or be influenced by positive and negative experiences.

These considerations contribute to a nuanced understanding of the moral status of AI. By addressing the intricacies of these aspects, we aim to provide an outline for teaching ethics to AI, having in mind that no existing artificial system meets even the minimum criteria for moral agency or moral patience.