AI alignment research concerns itself with designing AI systems that are consistent with human values.
However, ethical discourse is typically characterised by disagreement. Moral theories disagree about which actions are right or wrong (and why), as well as which states and objects are ultimately valuable.
But this raises a problem: How are we supposed to align AI with our values if we cannot come to agree upon what we value in the first place?
Yet for all that, there is much more that unites the different ethical theories than divides them.
Strikingly, though, except for a few notable exceptions (e.g. Derek Parfit), moral philosophy has focused almost exclusively on disagreement.
In this project, I aim to provide a systematic account of consensus among different ethical theories which can then serve as the basis for AI Alignment research.
A recent debate in the ethics of AI concerns the value of Human-AI companionship.
In this article, I offer a quasi-Kantian account of good companionship. Following Alison Hills, I distinguish between moral knowledge and moral understanding. I argue that emotions enable us to grasp ethically significant content and thus play a key role in morally worthy action. I then question whether AI is capable of meeting the conditions set out in this quasi-Kantian account. If it cannot, then an AI companion cannot act in ways that possess intrinsic ethical worth.
I am currently revising a draft. Please email me if you are interested.