Modeling the concept of intentional action in artificially intelligent systems: moral considerations

My aim in this essay is to provide a multidisciplinary characterization of the folk concept of intentional action for design and use in artificially intelligent perceptual-cognitive systems (PCSs). Immediately relevant to this aim is the empirical work done by Josh Knobe on the concept, which suggest that the concept is laden with moral considerations that often go undetected. But my general framework for understanding the nature of concepts is borrowed from the philosopher Ruth Millikan. This selection follows the strategy of information scientists Fausto Giunchiglia and Mattia Fumagalli, whose writings provide much technical inspiration for my project and inform my core methodology. My essay is organized as follows. First I provide a case study from Josh Knobe that demonstrates the influence of morality on people's judgments about intentional action. Next, I explain my selection of Giunchiglia, Fumagalli, and thus Millikan. I follow with a detailed analysis of Millikan's theory of concepts as "recognition abilities." After then extending this analysis to the recognition of mental states, I endeavor to model concepts, thus characterized, in a PCS. I return to Knobe's case study and attempt to apply the model there without accounting for any moral factors. When this fails, I expand the model to account for the influence of morality. The success of this expansion does not mark the end of research or discovery regarding the aims I set for myself in this essay. Indeed, my hope at the outset is simply to have encouraged practical thinking about how an artificial system might be designed and trained to gain competence with the folk concept of intentional action as it is and as it functions in ordinary human thought.