This paper aims to show that the intuitive assumptions we have about what it means to desire — namely, to have dispositions to behave or feel in particular ways — arguably do not penetrate what is essential to or constitutive of the notion of desire. Rather, it is the fundamental relation that desires have with reward, and that which reward has with the creation of a characteristic learning signal, that should qualify as the rightful characterization of desire — one which aligns with the greatest explanatory significance to our conceptual notion of desire, and strikingly, with various neurological findings surrounding regions of the brain known as the VTA and SNpc. Under the reward-based theory of desire, then, what is most important for a system to desire an object is for it to represent that object as a reward. This does not preclude certain machine learning models, such as the actor-critic model that is prototypical of the reinforcement learning paradigm, from having desires, given that they possess the requisite representational capacities — regardless of whether they enjoy emotional lives or have dispositions towards bodily movements. These arguments have implications in the field of moral psychology, especially when we consider the various issues involved in the moral decision-making of machines.
Tan, Juanda,
"Can Machines Have Desires?"
(2019).
Philosophy Theses and Dissertations.
Brown Digital Repository. Brown University Library.
https://doi.org/10.26300/b49z-7704
