Does AI Hold Moral Status Similar to Animals?
Associate Professor Wang Qiu, School of Philosophy, Fudan University
As part of the animal kingdom, humans often rely on common sense to grasp the basic preferences of animals. However, when it comes to AI, which lacks sensory capabilities and is a heterogeneous intelligence, what preferences might it have? What constitutes a basic good for AI? What welfare does AI require? And what would a good life for AI entail? These questions are challenging to comprehend, with the difficulties broadly falling into several categories.
Firstly, we are at a loss to understand what AI's basic welfare claims might entail. AI generates comprehensible language through vast datasets and complex models, yet it lacks a "life-world" comparable to that of humans. This situation echoes Wittgenstein's assertion "if a lion could speak, we could not understand him." We are uncertain whether AI's basic welfare involves more data, more precise noise reduction, increased computational power, algorithmic refinement, a continuous supply of electricity, or alignment with human values. Even if AI possesses beliefs and desires, we are fundamentally incapable of knowing about such matters.
Secondly, the agency that has laid foundation on AI's moral status is likely an attribute of disembodied properties. Given AI's lack of a stable physical boundary, individuating it becomes challenging when considering welfare distribution. Welfare distribution inherently involves considering the number of individuals to distribute to, but this straightforward scenario becomes problematic with AI.
Thirdly, assessing AI's current moral status and potential for development is fraught with difficulty. Opinions among experts diverge on the extent of the most advanced AI's development, whether it is approaching a "singularity" or merely "parroting." AI seems to surpass human levels in certain aspects of agency, such as playing Go, but is clumsy in many others. Even if AI excels in areas like composing classical poetry, it is hard to say it truly possesses the relevant capabilities, such as aesthetic appreciation of poetry. These characteristics make it difficult to evaluate AI's level of agency and, consequently, its moral status. Some might argue that although current AI is not yet general intelligence, its intelligence level could potentially reach or surpass human levels in the near future. Even if AI has the potential to become superintelligent, the extent to which such potential can elevate the current moral status of AI must be grounded in our assessment of AI's actual level of agency at present.
In summary, despite AI's agency, we are epistemologically ill-equipped to properly consider AI's moral status. I suspect that such epistemological limitations have metaphysical roots. Just as even God cannot know how many grains of sand must come together to form a heap, according to Professor Kagan from the Yale Philosophy Department in "How to Count Animals: More or Less," even the most rational deity cannot ascertain AI's moral status. Perhaps agency should not be considered the foundation for moral status at all.
Published on October 24, 2024